There’s a growing amount of information on video cards for rendering these days. That being said after spending some quality time with a Quadro 5000, Tesla 2050, and GTX 470 I thought I’d take a minute to share some of my findings/opinions on how these cards stack up in my own daily use with 3ds Max and iray. Perhaps this info will benefit anyone that’s trying to figure out which video card(s) to purchase.
I recently built a new workstation based on the EVGA x58 FTW3 motherboard. I chose it because it seemed to provide a decent amount of newer features like SATA 3.0, USB 3.0 and of course plenty of PCIe slots for multiple video cards. For the video card I chose an EVGA GTX 470 “gaming card”. I chose this card because it is a FERMI based card and at the time seemed to offer the best bang-for-the-buck as it was on sale. 
A few months later I added an NVIDIA Quadro 5000 and Tesla 2050 to the mix. Now, the first thing to note is that a combination of GTX470, Quadro 5000, and Tesla 2050 is an odd combination. I say “odd combination” because the GTX 470 with it’s 1280mb of on-board memory can bottleneck the rendering process since the 3ds Max scenes (including texture maps) must fit into the memory footprint of EACH video card. The on-board memory of the video cards are NOT considered cumulative amounts. So, while the Quadro 5000 has 2.5GB of on-board memory and the Tesla 2050 has 3GB of on-board memory, the scenes can not exceed the 1280MB on the GTX 470 if I wish to use all three video cards at the same time.
As you can imagine since the scenes have to fit onto each GPU, the amount of GPU on-board memory is critical. The more memory your GPU has, the more complex your scenes can be. Of course you can render with iray and use on the CPU and system RAM…but it’s going to be painfully slow in comparison to using GPUs, especially the new FERMI based cards.
So we’ve established that on-board memory is critical but the number of CUDA cores and their speed is also critical. The more CUDA cores (Stream Processors) your GPU has, the better. The faster they are, the faster your scenes will render. However, on the flip side the faster they are the more heat they generate. Here’s a comparison of CUDA cores & clock speeds on these three cards:
GTX 470 = 448 CUDA cores (core clock: 1215 MHz)
Quadro 5000 = 352 CUDA cores (unsure of clock information)
Tesla 2050 = 448 CUDA cores (core clock: 1150 MHz)
As you can see the “Gaming card” has the same amount of CUDA cores as the Tesla and they are clocked at a higher speed. This is reflected in the performance as well. When rendering with iray the GTX is noticeably faster than the Quadro 5000 and a little faster than the Tesla 2050 (provided I can fit the scene onto the GTX card). The next consideration and what might be the most important to many of us is cost.
Here’s a cost break down between these three video cards:
GTX 470: $259.99
Quadro 5000: $1,779.99
Tesla 2050: $2,499.99
In looking at the prices and with the knowledge that the gaming card is faster to render a scene than the Quadro 5000 or the Tesla 2050 the obvious question then becomes: How can anyone justify the huge cost difference of the Tesla & Quadro cards vs. the 400 series FERMI based “gaming cards”?
Answer: The main consideration (for me) is memory limitations of the GTX type cards. The Quadro & Tesla series house more on-board memory therefore I can render larger/more complex scenes. Of course I do have to say that it’s still difficult to fit complex scenes on the 3GB of RAM of the Tesla 2050 but it’s a lot easier than trying to fit a scene onto a 1gb GPU like my GTX 470. On the upside you can find some GTX cards with up to 2gb of on-board memory so I’d definitely look into those over the 1gb models if you plan on rendering with the card.
The other issue to consider is heat. I can tell you that my GTX 470 generates WAAAAY more heat when rendering than the Tesla 2050 or Quadro 5000. I do worry about that damaging the GTX card as I’m sure it wasn’t designed for this kind of abuse. I could water cool it, but to be perfectly honest since these gaming cards are common, somewhat inexpensive, and updated/redesigned frequently I would probably replace it within a year or so anyway. So if it lasts that long I’m good to go. The Quadro & Tesla cards are built to last longer under the stress of rendering. So your investment in a Tesla or Quadro card should last longer in that aspect.
NOTE: Seriously, if you render with GTX/gaming cards please make sure you have ample cooling and/or the workstation isn’t near anything flamable! I’m not saying they will instantly burst into flames when you hit the “make pretty button” but these things get CRAZY hot when rendering for long periods so please be aware of that.
The final consideration for me is 3ds Max viewport performance. In this aspect the GTX 470 pales in comparison to either the Quadro 5000 or the Tesla 2050. Don’t get me wrong, my 3ds max viewport performance isn’t bad with my GTX 470 but it’s noticeably more responsive/smooth with either the Quadro or Tesla cards.
Bottom Line: Which GPU(s) should you purchase? All I can offer on that question is advice on what I’d do. I do like the Quadro 5000 and Tesla 2050 because they are quieter & run cooler than the GTX card and of course I can render larger scenes with the Quadro/Tesla cards. That being said if I were building a new machine I would probably start with 1 GTX (FERMI based) video card with 2gb of on-board memory depending on the reviews of those cards. If the reviews didn’t look good for those cards I would start with a 1.5gb FERMI based gaming card. Try that with 3ds Max to see if most of your scenes fit into that 2gb of on-board memory. If most of your scenes do render with a 1.5gb or 2gb GTX type card then buy one or two more & enjoy some very quick rendering (but lot’s of heat). Just don’t be surprised if the gaming cards don’t hold up under the extreme stress of long periods of rendering, unless you water cool them.
If most of your scenes don’t fit onto the 1.5/2.0 gb GPU’s THEN you should explore the Quadro/Tesla series…OR wait & hope for someone to develop a GTX based video card with a lot of on-board memory. The other option is a Tesla 2070 that has 6GB of on-board memory. It would be awesome to have two or three of those! However that configuration would be really expensive but it might pay for itself if you can turn projects around quickly with that configuration.
NOTE: You may also check my iray/GPU F.A.Q. HERE.
Example iray animations:
Vizdepot Entourage Movement and Life test from Jeff Patton on Vimeo.
iray animation test 01 from Jeff Patton on Vimeo.
285 comments
Trackback e pingback
-
Comparação entre diferentes GPUs para computação gráfica | Allan Brito
[...] pelo artista 3d Jeff Patton, em que ele coloca as suas considerações depois de fazer testes de render no ... -
Best GPU for architectural visualization rendering? | Blender 3D Architect
[...] found an article about the use of GPUs for architectural visualization written by an artist called Jeff Patton, where ... -
Wise Rebel | Architectural visualiser- Pros and cons of GPU-accelerated rendering
[...] http://jeffpatton.net/2010/11/gtxquadrotesla-my-opinion-on-todays-gpu-selections-for-rendering/ [...] -
EVGA 560 ti 2 win
[...] [...]
Great write up Jeff! I do wonder if the the GTX cards will get better with the heat issue. Like the recently released 580 which seems to answer that. How well it stacks up against the Quadro and Tesla is another question.
The fact the 580 has 512 cores if I remember correctly.
By the sounds of the specs the 580′s sure sound like they will be quite a bit faster (for GPU rendering) than just about anything else we have available to us today.
Unfortunately it looks like they come with 1.5gb of on-board memory and while that may be fine for games it just won’t cut it for rendering anything other than fairly basic/stripped scenes.
Would be great to see a “rendering” video card with the speed of the GTX’s, build quality of the Quadro/Tesla series, and 8 to 12 gb of ram.
Hi Jeff thanks for all the very informative blogs you write. I have a question to ask you.
When you say “since the 3ds Max scenes (including texture maps) must fit into the memory footprint of EACH video card.” how can you calculate or know the total size of the above mentioned items?.
Thanks
See point #14 in my iray/GPU FAQ HERE.
I see. I thought there was a properties function in 3Dmax or similar.
Thanks for your reply.
Interesting, thanks for sharing…
Which renderer are you using ?
I mention iray in the article but I think the info may be applicable to any GPU based rendering application.
Nice! thanks for sharing. I am buying a new graphic card right now but I don’t think GPU so called “real time” renderer are ready yet for my usage (I do mostly special effect and movie CG). To many features that I need are still not allowed by those renderers. I was really impressed by the benchmarks in 3dsmax of the Ati firepro v7800 and it’s so cheap that I want one for display and I’ll try to mix it with a Fermi 580 with 1.5 for GPU rendering with vray RT GPU.
Thanks for this interesting comparison.
Users can also look at the new GTX 460 with 2 GB RAM, there are about 200 $, and 2 or 3 of this GPU are very powerfull ( 336 cuda core on each one ) and power consumption / heat are “limited”.
Not as good as tesla 2070 with 6 GB RAM, but you can buy it for 600 $.
Well with technology like this i am not sure you can even call it an “investment” in pure sense since it will be “old” in 1-2 years.
You can buy 6,5x $260, for the $1800 Quadro and not even accounting for selling the cards in second hand and money rates in the bank. So you can buy an updated GPU for 6,5 years every year. After 6,5 years you do have a up to date GPU vs a old quadro 5000 outdated.
The only reason to buy any of expensive ones is if memory is crucial, further if Iray is really necessary with its limitations now.
Jeff, you got to find a way to get your hands on Nvidia’s new GTX 580 card and see how that compares with the ones you already tested.
Also concerning the memory limitation, isn’t there a way to use/share main system memory? Seems like that would be a big addition to these renderers if they somehow allowed you to share your system memory when your videocard memory was running low.
Also, I have to ask why so much memory is needed overall – geometry in general isn’t all that memory intensive, so are you really using about a GB worth of textures?
Geometry is memory intensive if you have curved objects. Depends on polygons numbers. A 3D car is much more memory intensive than a 3D big building that if it has only planar surfaces is just a bunch of planes.
The meshes I’m working with these days are in the 7 to 10 million polygon range. That alone will eat up around 1gb of GPU memory. That in combination with a large frame buffer (3k and higher) is what’s preventing me from using this technology.
I copied this information on frame buffer size from Shane Griffith’s blog at the AREA site: http://area.autodesk.com/blogs/shane/the_iray_faq/comments
“Using a Quadro FX 4800 with 1500MB of dedicated video memory:
1000 x 1000 = 425 mb
2000 x 2000 = 555 mb
3000 x 3000 = 750 mb
5000 x 5000 = 1125 mb
6000 x 6000 = 1200 mb artifacts start to appear as memory limits are reached
7000 x 7000 = render fails”
So iray isn’t independent of the render size. Don’t forget to factor that into your memory usage calculations. :/
With that in mind, personally speaking if I can’t render with a 3gb Tesla I’ll have to pass on the 1.5gb 580′s at least for rendering my “work” scenes.
Also don’t forget that all textures need to be in memory uncompressed so regardless if you’re using JPG or PNG or TIF they will all use the same amount of memory.
Nice article Jeff, I’ve been keen on the idea of the GTX boards because of their low cost but I’m now wondering if the 460 2GB might be a better alternative to the 480 1.5GB just for the extra amount of memory
J.
J.
I was hopeful that someone would make a 2gb 580, but apparently that wasn’t the case. So, if you plan on rendering with the GTX type cards I’d have to say yes the 2gb 460 sounds like it might be a better option than a 1.5gb 480 since it provides a little more headroom in the memory department.
Hi Jeff,
Thanks for providing good knowledge. It is helpful in making decision on cards.
But i still not strong on it that what about the number of cuda cores on the card.
What i should select if i want both view port and rendering both perform best.
thanks
rahu
That’s tough question to answer since the quadro cards offer great viewport performance but slower clock speeds whereas the GTX cards offer higher clock speeds but less viewport performance. Then you also have to factor in the cost difference between the two
IMHO memory footprint of the card is the most important aspect because if your scenes don’t fit onto the card, or the card can’t render the size images you need then the number of cores the card has doesn’t really matter.
As for viewport performance it depends on your scenes. If you generally work with very complex scenes then a high end Quadro should work well. If your 3ds Max scenes aren’t super complex then a GTX type card should offer the best bang for the buck and decent performance.
Hi ..
Jeff it really usefull information you share with us, recently i decide to upgrade my card as my current card are not work properly with complex scene specially
on viewport.
with limit of budget 500 $ i find the best 2 option.
1- Nvidia GTX 580 which look to be the stronger in Gtx cards .
2- Quadro 2000.
Gtx have strong preferences even it reach to the high preferences of Quadro 6000 except memory size sure .
see this comparing please
http://img600.imageshack.us/img600/1885/97411471.jpg
Quadro 2000 as it in same level of price and support all new tech it will be good option …
now i have GTX 580 with memory size 1.5 GDDR5 still card specialized for games but look with high preferences and the Quadro 2000 with memory size only 1 GDDR5 but under professional card specialized for 3d . i’m confused so please tell your openion and what you suggest me to choose
If you plan on rendering with any GPU based application (iray, VRay GPU, Octane, etc.) I would lean towards the 580 card since it has more memory & will probably be faster to render than the Quadro 2000. BUT, keep in mind these GTX type cards weren’t designed for the abuse of rendering so the heat could damage the card over time.
If Viewport performance in 3ds Max is more critical to you or if you don’t plan on using any of the GPU rendering applications, I would probably lean towards the Quadro card.
That being said please keep in mind that’s just a guess as I haven’t personally tried either card and I can’t say how well (or not) the 580 handles 3ds Max viewport performance when compared to a Quadro 2000.
I know both of those cards are fairly new but I would still ask around on some of the popular CG forums for thoughts on those cards from people that already own either the 580 or Quadro 2000 and actually tried them with 3ds Max.
Hi Jeff,
Thanks a lot for sharing your findings in this matter. Based on your conclusions I was thinking on getting a EVGA x58 FTW3 and 5 cards GTX580 (7680 GB in total) in conjunction with a good water cooling system and a big radiator… what do you think?
1. If you’re planning on rendering with this setup you can’t combine all the GPU memory together to come up with 7680gb. The rendering applications will only see what is on each GPU since the scenes have to be loaded onto each GPU. Again to be perfectly clear, the GPU memory is NOT cumulative. Your 3d scenes have to fit onto each card individually so it’s 1.5gb for all 5 GPU’s.
2. I don’t know if the motherboard would support five of those cards. Keep in mind that some video cards take up two slots then you have the water blocks/tubing to consider as well. Maybe you’ve already measured and know it would work, I dunno…just thought I’d mention it. You may want to start with two or three GPU’s at most just to ensure you have room/power for them. You can always order more later once you verify there’s room & adequate power/ventilation.
thank you Jeff for your reply with this clarification.
i heard that Rendering with GPU dosn’t give good result like with CPU is that true.
Good result meaning render time or rendered images? There are more limitations/restrictions on GPU based applications…but for users that can live within the limitations and that have the right hardware the GPU results can be very impressive.
render images, i just heard about that . by any way i got your answer and
and thankx again Jeff
Dear Jeff,
Thanks a lot for this information.
It seems that we need someone to purchase the 580, hack it open, place 6 gigabytes of RAM in it, add an excellent cooling system, and sell it to us for one third the price of Tesla. He would surely make a lot of money.
Seriously though, games are becoming more demanding of the GPU by each release. My guess is that the Geforce series will, in a few years, will have to posess over 6 Gb of RAM anyway.
Since the demands of architectural visualization will NOT change that fast, Geforce series will catch up with our demands and perhaps even surpass them. This might be the case in 3 to 5 years. Until then, we shall clench our teeth and persevere.
Thank you ever so much for your valuable input and time.
Bummer…
I guess I’ll have to figure out something else…
Thanks a lot for taking the time to answer
Cheers
Hi Jeff
Thanks for the great article.
I’m in the middle of wanting to replace one of my graphics cards. At the minute I’ve got a ATI FireGL v8650 and it just doesn’t perform as well as it should. The drivers don’t work and as far as a can tell it just doesn’t work well with 3ds max.
So I’m going to sell it and buy something new, with a mind to use Iray.
In your option what do you think would work better,
1 x Quadro 5000 2.5GB
or
2 x Quadro 4000 2GB
Any help would be great
Thanks
Si
That’s a tough one. It may be wise to initially buy a 2gb GTX type card (460) and see IF you can render with that before spending the money on a Quadro card.
That will let you know if you can render your scenes with 2gb of memory or not. If you can’t render with a 2gb card then you’ll have to decide if a 2.5gb card would work and is that extra .5gb worth the cost?
good idea, I was look at one last night.
Just been lookin at the Quadro 6000 Graphic card now that would be nice, if I had £3,735.33.
Hi Jeff,
Thank you for your explicit explanation. It is very helpful for the beginner like me. I am a student in architecture working a lot with image and rendering. I have a computer which dates back ten years ago.
Now, i am planning to buy a new computer which will be good enough to handle with the rendering: Vray, Maxwell, Fryrender, and at the same time be friendly with 3D software: Rhino, 3DS Max, I am quiet interested in GTX580. Do you know any motherboard that work smoothly with this card? And what is the difference between having two or three card at the same time and a single one? Can i have GTX 580 with GTX 460 to make rendering faster with bigger pixels?
Thanks in advance.
M.
Hi Jeff!
Greetings from Poland ;]
Did you have any problems using GeGorce GTX together with Tesla?
I’ve got GTX460 2GB card and I’m planning to buy Tesla C1060 card, because of its 4GB of memory onboard.
Will they work together?
hi kcpr,
I think will buy GTX 460 in few days, but wondering on viewport performance.
Now i use HD2600 and its much faster than GTS250 in viewport.
Did your gtx460 run smooth in viewport?
I’ll use for architecture & interior but not a large scene, its only about 900k poly
Jeff, thank you for this usefull sharing.
hi antok!
I didn’t noticed any issues suring displaying my scenes, but it could be real subjective feeling, especially when I don’t have cards you’re using to compare with.
Bit your scenes are pretty light, so I wouldn’t expect main issues…
P.S. I’ve just bought second GTX 460 with 2048 MB, becaose my first is working grat.
…And wailting for GAINWARD GTX 580 with 3GB of RAM
Greetings!
Michael – I would imagine the GTX580 would be compatible with most any motherboard that supports PCI Express 2.0. There should be lot’s of boards available for consideration. Sites like newegg.com and tigerdirect.com are good places to look for PC parts.
Multiple GPU’s really makes little difference to the applications until you hit the render button. At that point more GPU’s = faster renders with GPU accelerated applications like iRay, VRay GPU, and Octane to name a few.
If the memory footprint on the 460 and 580 are the same you’ll find little difference between a rendered image from either. Of course I would expect the 580 to render the image faster than the 460, but as far as the output itself, it should be the same.
If the 460 has 2gb of on-board memory while the 580 has 1.5gb then you’ll be able to render slightly larger/slightly more complex scenes with the 460 than the 580.
kcpr – I have encountered no hardware/software issues with running a GTX card along side the Tesla or a Quadro card.
I believe the only issue you’ll encounter is that if a 3d scene is too large for the memory on the GTX card. When that happens either the lower memory video card will not be used by the application or it may create an application crash situation.
Thanks for answer!
Theoretically in the situation you’ve described iray should say “theres no memory on cude device 0″ and render with Tesla only. If it crashes it would make buing Tesla useless for scenes bigger than 2GB.
Thank you, Jeff, very helpful!
Wow, great thread!
Regarding memory Ive always wondered why none offer a graphic card that you can update. This buying and discarding thing doesnt seem neither wallet nor environment friendly…
Very good point, totally agree.
What about underclocking gtx cards to get lower temps and longer lifespan? Maby even try to match its quadro counterpart. I´ve read some gtx 580 reviews and they seem to peak at something like 80-90 deg C which seems hot but apparently is alot lower than the gtx 480. What temps is quadros and teslas running at?
Hi jeff i’m typing from Palermo (Sicily) i have on board Nvidia Quadro 5000 2.5 gb, i use 3ds max 2010 and i work in mental ray, but i do not know setup my video card at the best with max i have tried same setup (direct3D o OpenGl) but the rendering is alwais the same time (es. 3 min 30 sec) can you help me giving same tips for let work my video card at the best, you understand that i have spend lot og money in italy.
My os is windows 7 64 bit, cpu processor i7, video card Quadro 500,3ds max 2010 64bit,
Thank you for help
I read somewhere that the reason why quadro cards are expensive its coz they are more stable while rendering. But if i may ask, an i7 950 @3.03GHz and 6GB DDR3 1600MHz and a GTX 580, how would they render a HD 1920×1200 resolution animation (1 minute) , putting in mind a scene in 3dsmax 2011 with 5million plus polygons using iray…. Is the rig a good combo?? thanks for the tips too.
Hey Jeff,
My appologies upfront if you covered this question already. If I load two quadro 5000 in SLI can I expect IRay to render on all the cuda cores? Thanks, J
Jody – Yes, iRay will use as many cores as it can. I’m not sure if it will use 100% of all cores on the card(s) you use to drive your monitor. I would think some would be set aside for the monitor…but that’s just a guess.
Also, no need to SLI the cards for iRay. It will use them without SLI.
Timjan – Yes, under-clocking would surely lower the temps of the GTX cards. It would also reduce the performance a bit depending on how much it’s under-clocked. The main problem is that you’ll still have to work within the memory limits of the GTX type cards.
I’ll post the Quadro/Tesla render temps (and ambient/room & case temp) as soon as I get a chance. I’m tied up with a few short deadline projects at the moment.
giovanni54 – Make sure you have the latest drivers installed and then configure 3dsmax to use Direct3D/DirectX10. That seems to work well for me.
Just keep in mind that mental ray doesn’t use the video cards for rendering at this time. For that, you’ll need to use a GPU accelerated rendering application like iRay, Octane, VRay GPU, etc..
Nesim Mosh – I can’t really say what type of render times you’d get because render times would depend on the scene. Architectural interior scenes typically take longer to clear from noise than product vis/exterior scenes.
I will say that simply opening a 1920×1200 frame buffer will eat up about 500mb of memory from the GTX50 so right off the bat you’re down to 1gb of memory.
A 5 million polygon scene (geometry only) will probably consume about 300-500mb of on-board memory. So that doesn’t leave a lot of room for textures, especially if you plan on using a HDR.
That’s why I keep saying that memory restrictions are the main problem with using GTX type cards for rendering. If someone would release a reasonably priced 6gb-12gb under-clocked for cooling (or water cooled) GTX card…the CG world would make them very wealthy.
hey jeff,
very, very interesting thread!!!
just one question – when rendering with iray and a setup of two gtx580ies – does the scene have
. .. to fit on both card’s memory
or do they double – say 1.5 gb on the gtx580 x2 = 3 gb
Quotes from my article & replies above:
“3ds Max scenes (including texture maps) must fit into the memory footprint of EACH video card. The on-board memory of the video cards are NOT considered cumulative amounts.”
“you can’t combine all the GPU memory together to come up with 7680gb. The rendering applications will only see what is on each GPU since the scenes have to be loaded onto each GPU. Again to be perfectly clear, the GPU memory is NOT cumulative. Your 3d scenes have to fit onto each card individually so it’s 1.5gb for all 5 GPU’s.”
Hi Jeff, thanks for sharing !
do you have (or plan to) benchmarking rendering scence and than get quantitative comparision between Tesla/Quadro/GTX separatly? I would be very interested in such comparision,
thanks again,
A.
Yes & no. IMHO it would be nice to see one person independently test many of the GPU’s available today. It’s not feasible for someone like me to do as I would need to purchase all these GPU’s just for testing. Instead it would need to be someone that can freely get GPU’s for testing purposes.
I thought about contacting websites like Tom’s Hardware or similar to help setup a standardized test since I think they are provided GPU’s for testing.
In other words I guess I just need to locate the right person with hardware contacts that’s willing to tackle such an ongoing project in an honest/unbiased reliable way.
Hi Jeff,
I asked that, because I metnion that you already have those chip with you…
anyways… thanks again for this blog!
Hi jeff,
I want to know if I can put 2 different GC on my motherboard sabertooth X58 , i.e. GTX 460 2GB and GTX 580 1.5GB?
Or it has to be identical?
No, they do not have to be identical.
Of course in that GPU combination if your using iRay and your scene is >1.5GB and < 2.0GB then the GTX580 won’t be used.
@ Michael: Which video card is best for Animation using Maya 2011 .. i already am using FirePro v7800 .. wanted to give Quadro 4000 a chance .. what do you say?
3ds max use DX9 for use more it features, means the gaming card could possible reach the workstation card. The workstation use opengl optimized driver mode for better viewport. But the fermi series have a lot complaints at viewport performance lately / or 3ds max 2011 it self?
The maya uses opengl extensively, so the viewport perf. of the workstation will be very far away than the gaming card. the Fermi series has a slowbug openglreadpixels.
Well i had database performance of iray, viewport, and cpu render of 3ds max. I hope someone could tested his/her rig and fill the performance, so we can choose which better config for the new one.
http://spreadsheets.google.com/ccc?key=tW4TrlUCS7gS9aaoctBh5Og&gmrcpt
Hello Jeff.
Just to inform you with this new : Gainward announce a new Geforce 580 with 3 GB Ram and optimized cooling system.
Look great for Iray.
http://www.gainward.com/main/vgapro.php?id=454
Palit has also announced a 3gb card.
http://www.guru3d.com/news/palit-releases-geforce-gtx-580-with-3gb/
Hi Jeff,,
Im quite new to iray in general, but ive been using mental ray in 3ds max for almost 3 years now, im planning to build a pc in a few months, from what i have read in this thread, i cant go wrong with nvidia cards may it be gtx, quadro or tesla, what im interested in knowing is will iray work on ati cards? (radeon and firepro) the reason i asked is because i noticed these 2 cards from ATI, Radeon 6950 and 6970 which is way cheaper than the GTX cards, i hope you can help me out on this one.
As far as I know (I haven’t checked lately) ATI doesn’t support CUDA which iRay uses for GPU rendering. Therefore no, the ATI cards would not work with iRay.
Thanks for clearing that up Jeff, i guess ill just go with the GTX 570 or 580 for now, i would like to take this opportunity to ask you regarding which version of windows should i purchase for my new pc build, im still not sure if i want windows 7 64bit, (i havent used any version of it yet) im a big fan of windows xp 64bit mainly for its stability, from your experience, which one would be more worry free in terms of compatibility with 3ds max?
Which O/S to use is very subjective. I’ve been using Win7 64 for quite a while now without any problems so I would go with it, but that’s just me and your experience with it may differ.
Hi Jeff, after doing a lot of product type renders recently in MR we tried the same scene in iray and got some really good results. So good infact that I am going to invest in some more gpu’s.
After reading your comments and looking at best value against gain I was thinking of adding 3 460′s to my Boxx workstation then tonight I found this little gem on a UK gaming site.
http://www.overclockers.co.uk/showproduct.php?prodid=GX-144-GW&groupid=701&catid=1914&subcat a 580 graphics card that comes with 3gb of ram. Further down the list is another 3gb 580 that has dual fan but i like the look of the full length heat sinks on the one above.
Anyway I think thats the route I will try, I’ll let you know how it performs.
Steve
Hi Jeff,
I’d have a question to the Fermi/CUDA based difference in Tesla/Quadro and GeForce Cards. I’ve just read a technical article in german computer magazine “c’t” that one of the main differences between Tesla/Quadro high-end cards and GeForce GTX cards is the DP (double precision) calculating speed. Nvidia artificially limits the GeForce Cards to 1/8 respectively 1/12 of the Tesla/Quadro-speed when calculating double precision operations. Do iRay or v-ray (state of the art renderers/raytracers) use such kind of operations internally? That would be a big argument for the higher price… if the renderes are only using single precision operations I think I’ll also invest in one of the new GTX 580 3 GB cards…
lars
Hi Lars,
I would imagine that iRay/VRay GPU do take advantage of the double precision operations, but that’s just a guess on my part. I will see if I can track down someone that can provide a more definitive answer.
Great Article.
Definitely the new GTX 580 3GB Chips sound promising. If someone gets one of those please provide us with some test results in terms of complexity of the scenes it can handle.
Greetings.
I not see any reason for IRay.. it not full real time render for my. I can not see final image!!!
So.. I need use Quicksilver render any time! -for this need more GPU and SLI mode!
I hate this day, when bay Quadro FX 5800 and 2x Tesla C1060 -this cards do not put to SLI mode! So, in this moment, I use for render only one GPU!!!
hi there,
can anybody explain how i can calculate the need of memory out of my scene. just want to know how much graphic card memory i need for my scenes. is there a table or something?
thanks a lot.
I don’t know of a table or chart for typical size but I posted some information in my comments/replies above about my scene sizes and frame buffer memory footprints, etc. and that should give you an idea based on your scenes & typical output sizes.
FWIW a typical scene for me these days runs about 12 to 16 million polygons. Sprinkle in a few texture maps, an HDR, and a 6k frame buffer output and I use about 3.8 gb to 4.6 gb of GPU memory.
thank you so much for sharing information with us.
just a last one. as i can see on cgtalk you`re playing around a lot with mental ray / iray and your video tutorials are also about rendering and lighting with mental ray.
some of your superb works, like the mercedes cgi`s are done with vray. why is that?
are you experiencing stuff with mental ray and working commercial on vray? is it a question of speed and material resources…?
thanks and kind regards.
m
Last year I used VRay on the MBUSA projects because I was asked to do so for compatibility with their own VRay workflow.
However, last month I was able to switch all my productions over to iray/mental ray! I will start posting those images in the near future as I get clearance to do so.
Hi Jeff!
Just wanted to ask something. I own a GTX460, mainly bought for render performance since it’s a 2GB model and i’ve been working with it flawlessly for the past few months, it’s really fast for the final quality of the render.
I want to buy a mid-range Quadro for better viewport performance, FX1800 to be exact, because the GTX honestly sucks at this. I want to use it in parallel with the GTX, meaning I want to use the Quadro for Max but the GTX for ocasional head busting in games (work demands relaxing when it’s done
). How can i do this without phisically switching between boards (i.e. take one out, put the other one in). I don’t want to load the quadro pointlessly while gaming, will simply switching between display ports do the trick?
Hope it’s not too confusing, my grammar suffers sometimes since i’m all the way over from Romania
Thanks in advance and keep up the good work, really nice work by the way! You just made a fan.
Afraid I’m not sure how you could achieve this without physically moving the primary/secondary cards around as you mentioned. Maybe someone else will chime in with a solution.
Hi guys, been shirking work looking for some information for my new pc build. I’ll mainly be using Solidworks but want to get more into rendering, particularly gpu based solutions. The past 3 weeks I’ve been trawling the internet for solutions to the problem of running quadro and geforce cards on the same system. I think I have finally found a no compromise solution!
EVGA have motherboards available, SR-2 classified I think, that have multiple slots, upto 8 pcie slots! The clincher here is that it has enable/disable switches on the board for the pcie slots. Voila! Enable each as you please, no driver conflicts. I will probably route the switches to the outside of the case, but I have some electronics skills, so if you’re not sure, get help.
The EVGA is available now but I will be waiting for the MSI Big Bang Marshal which is due any day now and uses the latest Intel Sandy Bridge platform. It also has a LucidLogix Hydra chip to enable different cards from ATI/Nvidia to run concurrently. I know…ATI/AMD…shhhh…
Hope that helps or at least gets you excited. Good thread, very interesting.
I want to build my first workstation for 3d renderind(cad, 3ds max) an would like your opinion
1x qudro 4000 ($780)
or
3x SLI gtx 460 1Gb (3x $200)
The quadro 4000 is based on GTX 460 architecture so they are basicaly the same card except for the optimized drivers on the quadro.
I really don’t think an optimized driver can catch up with two extra cards in any area.
Thank you.
Depends on what your goals are for the setup. If it’s rendering with iray, the multiple GTX460′s would be faster than a single quadro 4000. However, they are limited to 1gb of ram whereas the Quadro 4000 has 2gb. So the Quadro would allow you to work with slightly larger scenes and/or rendered output size.
I would imagine the viewport performance would be better with the quadro card as well but that’s very subjective and may not be an issue for you.
Hi Jeff,
your thread is exactly what i’ve been looking for!
very informative and insightful,
I work in arch viz and i am building a machine, i use vray 2 and 3ds max,
I’d really value your opinion on a pc that i’m serously considering;
dual socket setup with 2 Xeon 5650′s, a 180 gb ssd with a 1Tb raid Zero config, 16Gb Ram, and 1, possibly 2 GTX 580 phantoms,
at work my machine has a quadro 3800 which works well for viewport display, will the 580 be noticably worse?
is it possible/advisable to pair a quadro with a gtx?
if so which Quadro would give me the best bang for my buck?
Thank you,
Tim
“at work my machine has a quadro 3800 which works well for viewport display, will the 580 be noticably worse?” – I’m not sure as I don’t have a 580 here for testing so I can’t provide a solid answer. Perhaps someone else that does have a 580 will chime in.
“is it possible/advisable to pair a quadro with a gtx?” – I ran a quadro 5000 + GTX480 together for a while and iray didn’t seem to mind. However, I’ve also heard from other people that they did have problems when mixing the cards so it’s hard to say for certain.
“which Quadro would give me the best bang for my buck?” – Depends on whether or not you plan on rendering with the GPU. If so, then you’ll need to get a GPU that has enough memory to render your scenes at the size you need. If you don’t plan on doing any GPU based rendering then you have a bit more flexibility in choices.
Thanks for your reply Jeff, today i’m going to test our Quadro 3800 with a gigabite gtx 460, 3ds max 2011 and vray 2,
I have been looking at the 580 phantom and now i hear there is a gtx 590, 1000+ cuda cores, the thing is it has 2X 1.5 GB memory so will recognise as just 1.5, according to what i’ve read on this thread i’m tempted to go with the phantom because of the 3GB memory
will post my results
thanks,
tim
Hello Jeff, I have a simple question, perhaps it is too simple
ATM I do not have an iRAY compatible graphic card in my workstation.
Please have a look at mental ray render message window:
RC 0.3 info : allocated 23 MB, max resident 27 MB
In short: Does this message mean that my testscene would require max. 27MB of graphic card RAM? I would like to load some older scenes and have a look if I could render these scenes with iRAY and a 2GB or 3GB graphic card.
Thanks in advance!
In short: Does this message mean that my testscene would require max. 27MB of graphic card RAM? – Yes & no. I’m not exactly sure if that size includes the texture maps or not but it probably does…however either way that’s not the total size you’d need for GPU rendering.
You also have to factor in the output image/frame buffer size. There’s some information in previous posts above that outlines frame buffer size & memory amount required so don’t forget to add that into your memory footprint calculation.
Thanks for answer. AFAIK you have Quadro 6000 in your workstation. Do you have a tool like GpuCapsViewer/GPU Shark which can read out used graphic card RAM?
I am asking because I would like to know what happens if you load any of your scenes (with some textures) and compare the info from message log to real consumption (measured with e.g. GpuCapsViewer/GPU Shark or something else). Of course it would be interesting to test the scene with e.g. 800*600 and something bigger like 4000*3000.
Alright, here’s some tests:
Scene statistics = 525,318 Poly’s | 392,858 Verts | HDR environment/illumination and a couple of other texture maps.
1080p resolution render = 543mb GPU ram consumption.
render message window: http://jeffpatton.net/temp/iray_1080p.txt
Tesla c2070 data during render: http://jeffpatton.net/temp/Tesla_1080p.jpg
6k resolution render = 839mb GPU ram consumption.
render message window: http://jeffpatton.net/temp/iray_6kres.txt
Tesla c2070 data during render: http://jeffpatton.net/temp/Tesla_6k.jpg
10k resolution render = 1425mb GPU ram consumption.
render message window: http://jeffpatton.net/temp/iray_10kres.txt
Tesla c2070 data during render: http://jeffpatton.net/temp/Tesla_10k.jpg
Hope that helps.
Hi Jeff,
I am sorry for thanking you with some delay regarding your iRay RAM tests… too much trouble the last days
In fact this issue aroused interest from my side – more than before. Of course it sounds promising that the render message window predicts much more RAM than the GPU load is in reality. I thought that perhaps the reason could be that GPU RAM snapshot represents just one short point in
time and was not monitored the whole rendertime?
What do you think about Rivatuner monitoring capabilities or what about this nice and easy tool, it just one executable with no installer:
http://gpuz.techpowerup.com/11/01/01/8kw.png
http://www.techpowerup.com/downloads/1969/TechPowerUp_GPU-Z_v0.5.2.html
Do you know if there are any buggs related to the GTX 590 3g memory cards? Seeing that they are based on double GPU’s?
I’m not sure of any bugs but my friend Jennifer O’Conner mentioned a snafu if you’re considering rendering with those double GPU type cards. Apparently they may split the GPU memory into 1.5gb for each GPU.
Provided that’s true (and I don’t doubt Jenni) that would probably mean GPU based rendering applications like iray would only see 1.5gb of memory instead of 3gb.
Yea, when looking at this:
http://uk.geforce.com/hardware/desktop-gpus/geforce-gtx-590/specifications
Looks like thats the case. 3072MB (1536MB per GPU)
Hopefully that won’t catch anyone off guard that’s considering those for rendering purposes.
Hi
.
Fantastc blog
Q: So whats the best way to get both worlds with regards to quality for both viewport performance and rendering to the best price. It seems to me that one has to choose, now that i know that SLI will not provide any additional performance when f.ex. combining two GeForce GTX 470 or other.
Is the best solution to purchase two cards; one cheap for rendering and the other for viewport performance? And let the computer choose which one to use?
The goal is to get one cheap card that has 3gb for rendering and hopefully a Fermi card for vp performance.
I guess in that scenario I would first purchase the cheap 3GB video card and ensure it would work as expected for rendering.
If the cheap 3GB GPU works as expected for rendering your scenes then try it with viewport performance. Maybe you’ll get lucky and it will work ok for that too and if it does, buy another, and probably some cooling upgrades.
If it doesn’t work well for viewports then you know you’ll need something different for viewport performance. Then you’ll just need to use the iray manager script to tell iray which GPU to use provided the primary GPU can’t handle actually rendering your scenes.
Iray manager: http://dimensao3.com/al/
Ok thanks. So i guess that i can purchase an affordable 3gb GeForce card and then a affordable Quadro Fermi for Viewport performance? And let the computer choose which one to use, or use Ir Manager.
Is there a big performance difference with the Fermi on GeForce and Quadro? Maybe a littel early to ask?.
Hope i dont ask to much at once. Much apprechated that you anser these questions. Just amazing
.
The point being trying to save money by combining cards, otherwhise one could just purchase the high end nvidia cards hehe.
Wow what a job this is finding the righ combination and saving money in the process…
Yes, you should see a speed benefit in rendering with the newer fermi based cards.
NOTE: Just make sure the 3GB card has unified memory and doesn’t split it into 1.5×2 like the 590 discussion above your posts.
And what about the viiewport performance (fps) in 3dsmax or maya? Using the same specifications in the questons above.
Can one expect an affodable 3gb card to anywhere near the high-end cards from nvidia. The best solution might then be a 3gb Fermi card disregarding if its Quadro or GeForce?
In this post I’ve just been trying to provide some information for anyone considering GPU’s for rendering with iray. Viewport performance is a different animal all together.
I personally don’t feel confident in saying viewport performance would be like X or Y with any GPUs. I say this because IMHO there are far too many variables that play into how your 3ds Max viewport experience feels.
I’ve read about people having the same GPU’s in similar workstations yet their experiences with viewport performance were dramatically different.
Hi Jeff.
I’m in process of assembling new CUDA computer for my wife for iray rendering (she works in 3ds, most in Mental). All is done except CUDA cards.:) Motherboard Asus P6T7 with 7 PCIe slots, and I think to buy her 4 CUDA cards. I think of four gtx560 2GB or gtx580 3GB. But – 560 I can buy right now, and 580 with 3GB is not present in stocks at this time. So we must make decision: buy 2GB cards right now or wait a long time for 3GB ones.
And we are stopped here – we can’t understand what for amount of memory will be OK for her typical scenes… She has 8600GT 512M card at now – and most her scenes do not fit in it and I can’t test exact memory consuming…
So, my question is: can you please to load this scene for rendering to GPU (with res 1600×1200) and write here what amount of memory on GPU this scene consumed?..
If your answer is “yes” – the scene file is here:
http://narod.ru/disk/9315496001/test.rar.html
Thanks in advance and sorry my English.:)
I downloaded your scene however I can’t give you an exact GPU memory consumption value since I don’t have all the associated texture maps and of course those will also consume memory.
There was also a few maps/shaders in the scene that aren’t supported by iray, like the VRayMtl, BA_color_raylength, and ColorCorrect.
FWIW I can tell you what the geometry alone with a single material applied globally will consume @ 1600×1200 res: 1239mb
Thank you very much Jeff!
It is very helpful, even without textures (sorry I forgot about them, I’m C-programmer, not 3D-professional:), I’m doing that for my wife:)).
So, we can now realize estimated memory consumption of her typical interior scenes. Maybe it will be helpful to someone who investigates the same question – one can look at that scene, and given the consumption of its memory make conclusions… For these – a little info about scene: it has ~9.5M triangles, but only ~1.5M of them is the interior geometry, the rest (and LOT) of triangles are two carpets.
And we make such a conclusions (implying we do not want to spend the money to four Quadro):
- 1.5GB (and less) cards are not applicable to ~10M triangles geometry if we add to it textures, HDR, and bear in mind that we must leave ~300MB of free memory to GPU functionality (to avoid artifacts).
- 2GB cards are applicable to such a complexity – we have enough free memory to add to this scene ~400-450 MB of textures and HDR (and leave that ~300MB of free memory).
- 3GB cards are the best choice for such a complexity – they have a little margin to possible complication of geometry and/or textures. And the best price/quality I think. Bad news is only that here (Russia) we can’t buy it – no one of stocks sells these Palit cards at the present time.
Thanks again Jeff!
Now I would the main diffrence between the gtx cards and quadros are quadros can be stacked to increase memory bandwidth cuda cores working together for professinal applications where as gtx card can only be slid for gaming…also drivers is a huge diffrence too
Hello Jeff, thank you for this article. At this moment I’m going to buy a new nvidia based card and I’m not sure if iRay supports SLI based cards (thinking of GTX 590), because render engines like Arion or Octane does not support SLI. What would be a better option for me? Buying GTX 590 or Quadro 4000? I really dont care so much about stability/heat or power consume, all I need is speed and good price
)
Thank you for answering.
Damn, I haven’t read all the discussion. It seems that the 590 splits memory to 2x 1.5 Gb, which is quite small (I need to make renders >4000×3000 pix
), but it has twice more CUDA cores. It seems that the best sollution would be buying two Phantom 580′s with 3GB..
Just for the sake of clarity, SLI isn’t required for iray to use the GPU’s. Simply install the compatible GPU’s and you can use them in iray without needing to hook up the SLI bridge/cable.
I’ve already bought four Palit GTX-580 3GB – and it works good with iRay, my wife is HAPPY.:)) Renders are very fast, 1600×1200 interior without noise – 30-40 min (with glossy materials and glass in scene)… MentalRay renders this scene in 47 min on four machines (i7 and Q6600, 8-16 GB memory) with caustics DISABLED. So if caustics (that in iRay is “for free”) will be enabled, it takes even more time…
Arion – do not want to render on these cards..:( Its behavior is buggy – it shows that it renders on these cards, but GPUs load is 0% on all cards, and speed of render is as CPU render only.
But be careful – Phantoms take 2.5 slots of PCIe, so you can’t install 3 or 4 such a cards in one motherboard. This is the reason I’ve searched Palits for this purpose.
Sorry my English.:)
Hi JimStar!
That’s great that everything is working for you.
My question is: Does for sure iRay sees 3GB of VRAM of Palit GTX 580 3GB, or it is 2×1,5 – like GTX 590 ?
I have to be sure before byuing it, 1,5 GB of VRAM would make it useless for me.
Thanks in advance!
kcpr
Yes, it sees all 3GB of VRAM. One GPU, one memory bus, one chunk of memory. 590 has two GPUs, so two memory buses and two chunks of memory (1.5 + 1.5).
http://www.imagepost.ru/?v=gpu_1.jpg
I’ve underclocked all cards a bit – a bit slower but more reliable…:)
Keep in mind that the GTX590 is simply a pair of GTX570s on a single board. it is far cheaper to buy 2 570s and SLI them or not as the case may be. But for RAM needs the 3GB on the 580 will always outperform either the GTX570 SLI or the GTX590 single board.
Jeff, I recently purchased 2 GTX 590′s for my workstation.
VrayRT performance is amazing, Iray seems pretty good, but… these units put out a LOT of heat..
As of now there are no Cooler units available (although I could have bought them with coolers but like an idiot, decided not too). Great article.
David
I’ve been testing iRay on some simple scenes on my home computer which has a GTX 460. Now we are upgrading our 3dsmax at work and I am in the market for a new GPU. I’ve been looking all over forums and it seems that processing cores and memory are most important for iRay. Therefore, I have settled between the EVGA GTX 570 1.2GB and Palit GTX 460 2GB. The only reason I am looking at the Palit is for the extra memory. Now here is my main question, my scenes are all archviz, 7 million polys max, and a maximum of 3500px for the frame buffer. So, its seems like the 2GB card might be necessary to fit the scene/textures, or will the 1.2GB suffice?
Thanks for the help
8 million triangles = about 1gb of memory so your 7 million poly geometry will probably consume around 950mb. In looking at some of the test render information I’ve posted I think a 3500 pixel frame buffer will consume about 200mb. So, before textures are added you have used approx. 1.15gb and you haven’t added textures yet. NOTE: To be safe I rounded up my memory usage estimates.
Provided my estimates are somewhat close to accurate the 1.2gb 570 card won’t give you very much play room for texture maps/HDR environments so the 2gb card may be the better choice for you. Of course the 460 won’t render quite as fast as the 570 would, but if your scenes don’t fit onto the 570 then it’s no use to you no matter how fast it is.
Thanks for the quick reply. After reading through this article again I found this and it confused me a bit:
“3ds Max scenes (including texture maps) must fit into the memory footprint of EACH video card. The on-board memory of the video cards are NOT considered cumulative amounts.”
So, does this mean I should throw out my old Quadro FX 570 and only install the new video card since no scene I render will fit within its 256MB memory? Or if I leave it installed, will the scene load onto the new GPU and use the Quadro for additional CUDA cores?
Thanks again for the help, your site is really great.
Correct, if the ENTIRE scene will not fit onto a particular graphic card, it will not (and can not) be used to render with iray. It will not split the scene up to fit onto GPU’s with less memory, etc.. It’s all or nothing (well it’s all or CPU only in this case).
So, during render iray can use only one cards (memory and GPU).
In this case what advantage to have more than one cards, if iray can use only 1
Or iray can use memory from ONE card (how it choose?) and GPU from ALL available cards?
No, iray uses all compatible GPU’s and CPU’s by default or you can assign any combination of compatible GPU’s/CPU’s manually via scripts. However, the scene must fit within the memory footprint of any GPU you assign.
I’ve found that Inno3D and Zotac produse 580 GTX 3GB varm and it’s possible to replace the cooler to make water cooling and after that it will occupy ONLY 1 slot because DVI connectors position one line –> I thought about all 7 water cooled and P6T7
Hi Jeff,
Interesting post here on your site, I have been looking for a custom setup for Max, to render with Iray. I bought a evga GTX 590 with 3 gigs of ram (1.5 each chip). And this monster card is way to hot in my small form factor case, and without proper cooling, fan or water cooling, I have the impression my card will melt only with 5 minutes rendering!!! It goes up to 90 degrees and smells funny lol. I ordered a gaming case (Cooler Master HAF 932 Gamer ) with powerful intake, hopefully that will help my case with 2 more of those cards…
Hi Jeff,
And thank you for all those very interesting infomations.
I learnt a lot about which card i should chose for iray rendering.
But the important point for me is the viewport abilities.
To have a good frame rate in a complexe scene with, for example, an heavy particle flow animated.
which spec i should focus on.
so i whant to change my old card.
and i have to choose between a geforce 580 a 590 or a quadro 4000.
I didn’t understand the thing that a 580 3gb has better spec in every point than a most expencive quadro 4000 with 2gb.
Is it because of the maxtream driver ?
so which card should i choose, and which spec is importante for viewport ?
thanks this will be helping me a lot.
PS: sorry for my english i’am from France.
That’s a tough question to answer because I’ve seen people with similar hardware specs have completely different viewport experiences because they work on different types of scenes and perhaps different drivers as well. For example, wade through this 3ds Max 2012 thread HERE to see the various comments on viewport performance with 3ds Max 2012/Nitrous.
I rarely work with particle flow scenes so I can’t provide any useful info on how those type scenes work with quadro vs. the latest GTX type cards. You may want to ask over at the CGtalk/3ds Max forum to get feedback from people that frequently work with those type scenes & hardware.
Hi Jeff,
I’ve skimmed through the comments so I may have missed some info, but I’m confused over some of your conclusions on combining the Quadro and the GTX. You do know that combining these in these same machine disables the Quadro drivers, right? I’m upset about this myself as I refuse to pay for teslas at 20 times the price for the same amount of CUDAs, but would like a Quadro for display.
I would be very interested in hearing your take on the perceived viewport difference between the Quadro 5000 and your GTX running Max? Not so much in terms of speed as in smoothness, antialiasing and generalbdisplay glitches?
“but I’m confused over some of your conclusions on combining the Quadro and the GTX. You do know that combining these in these same machine disables the Quadro drivers, right?” – All I know is that at the time of writing this article I ran the Quadro & 470 GTX (and even a tesla + GTX) without any noticeable problems. Maybe the GTX did degrade the performance of the Quadro but if it did, I didn’t notice it. At any point it’s all water under the bridge at this point since I’m currently using a different motherboard and GPU’s.
“I would be very interested in hearing your take on the perceived viewport difference between the Quadro 5000 and your GTX running Max? Not so much in terms of speed as in smoothness, antialiasing and generalbdisplay glitches?” – About the only viewport difference I’ve seen from my past experience with GTX cards is how much smoother my viewport rotations are (higher FPS) with heavy scenes/geometry when using the Quadro’s. Especially true when using the 3ds Max performance drivers for the Quadro card. Of course all of this is subjective and from what I’ve read/heard my viewport experience may be completely different if I didn’t mainly deal with static scenes and basic animations.
From what I understand you wouldn’t get any problems, but the Quadro Maxtreme driver would be disabled so you would loose the performance. It is interesting though, I have a history with SGI machines and I remember the graphics being so much smoother than what I see today and assumed Quadros would bring that back. Given the quality of your work it would seem to me that you would notice if your viewport in Max suddenly got jagged or if orbiting occurred to be less smooth, but you didn’t notice anything? From your reply there seems to have been a noticeable difference between the Quadro and the GTX, however if I understand this correctly, a Quadro basically becomes a GTX if a non-Quadro card is present as it then will revert back to a stand Geforce driver.
It all seems to come down to sly marketing and greed from Nvidia. A perfect system for me would have been a Quadro 4000 and three GTX460 giving over 1000 CUDAs. One would have assumed Nvidia would have been content with that as this kind of system would have a practical ceiling for how much memory could be used, as established professionals like yourself would have to buy tesla to get 5-6GB memory anyway, but no, a Quadro requires Tesla. Makes one wonder why GTX cards have CUDA to begin with, as they undoubtedly confuses Tesla sales, but perhaps it is only meant for the 2D and Adobe market.
Question about GeForce GTX580 for animation rendering!
——————————————————-
I have installed a PC with Motherboard Socket 775 “MSI P45Platinum”;plus one little and old GPU GeForce 9400GT-512MB-DDR2, that I want replace… with 1 or 2 GPU Nvidia GeForce GTX580.
I have 8GB DDR2 Kingston as system RAM, and CPU Intel-Core2Quad Q9650, OS Win7-Ultimatex64Bit, x3 HHD 500GB WesternDigital SATA2 on 3,25 internal slot bay + 1 SSD OCZ-Vertex2 SATA2 120GB on 3,25 internal slot bay.
I want buy one “GPU-Xpander Desktop 2” because I want to install 1 or 2 GPU “Gainward GTX580 “Phantom” with 3GB-DDR5” that fill 3 slots in height because it have a radiator with heat-pipe system and 3 cooler fans:
http://www.gainward.com/main/vgapro.php?id=454
Or I will want to buy a “Palit” GTX580-3GB-DDR5, but it have just 2 simple fans mounted (without radiators, nor heat-pipes, and occupies just 2 slot dimensions height), and I am not sure if the heat is critical or not with such of simple gaming GPU!:
http://www.palit.biz/palit/vgapro.php?id=1700
I think “Gainward” is the right choice for heat dissipation…
1) But with those 2 GPU is it really possible to render a 800-1000 frames animation sequence, with max time x frame of 8-10 minutes, and render for 5days and 5 nights continually, or the stress is too high and these GPU will burn and damage immediately ?
2) Is it possibile to made those little animations in terms of stress, or you suggest GeForce just for static photorealistic image renderings?
3) I don’t want a much expensive cost for GPU solutions, such Tesla+Quadro GPU combination. So my choice is GeForce GTX580-3GB:
4) What is better from Gainward and Palit (GTX580-3GB)? (I think Gainward…)
Congratulations for your animations on Vimeo!
Please response me…
Many thanks in advance!
Dario Cavallo.
“is it really possible to render a 800-1000 frames animation sequence, with max time x frame of 8-10 minutes, and render for 5days and 5 nights continually, or the stress is too high and these GPU will burn and damage immediately?”
“Is it possibile to made those little animations in terms of stress, or you suggest GeForce just for static photorealistic image renderings?”
To address both of those questions I’d have to say I’m not 100% sure. Since all of this GPU “stuff” is relatively new, at least to most of us render guys, I think the jury is still out on whether or not it’s alright to use GTX type cards for serious/long term rendering.
Autodesk has gone on the record as stating the GTX cards may not be recommended for long term rendering due to the heat/stress factor. I have personally heard from a handful of people that said their GTX card bit the dust after some GPU rendering…but they were older cards anyway so that needs to be considered as well.
I really hope someone out there is doing some long term GPU rendering tests with liquid cooled GTX cards. I’m thinking that may be a nice option provided a person can live within the memory limitations.
FWIW the Quadro/Tesla cards I’m using typically hover in the 85c – 90c range when rendering (and that’s both long term & short term rendering). I imagine that if you can keep the GTX cards in that range or lower you’ll be better off in the long run. But please keep in mind that’s still a guess on my part. I haven’t personally tested this or spoken with anyone that’s done any long term render tests with the GTX type cards to see how they hold up over time to the stress of rendering.
“What is better from Gainward and Palit (GTX580-3GB)? (I think Gainward…)” – Between those two I’d probably go with the Gainward too since it seems to have a slight edge in cooling. Plus from what I read about the Gainward it ran a little cooler but with a slight overclock. Therefore you could underclock it a bit to make it cooler and still have good performance.
But again keep in mind that long term effects of heavy rendering with GTX cards is yet to be seen by the masses. If you (or anyone else reading this) is actually doing a lot of rendering with the GTX cards please let me know. I’d like to create a v2.0 update to this article in a couple of months and provide some real-world findings from people that have been using the GTX cards for some heavy rendering.
I wrote here about these “Palit GTX580-3GB”.
I’ve bought four 3GB Palits (my wife works in Max, not me:) and installed them in a big case with extra cooling (this is an old photo, later I’ve replaced this big fan with 6x120mm and installed a fan controller):
http://imageshack.us/photo/my-images/143/cimg1768m.jpg
http://imageshack.us/photo/my-images/831/cimg1770i.jpg
I’ve underclocked all cards to 1710/1215MHz 0.963V.
So, in a long render the temp of the topmost card is ~92C, the bottom card is ~72C (all fans in case are on maximal 2200 rpm speed in that moment, with my home conditioner enabled, VERY LOUD:)). Sometimes this comp renders all the night – no problems so far… Except the noise..:)
This is a test render (iRay, on all 4 cards 2048 cores, 1:15 hour long, 6400×4800 rescaled to 1600×1200 to reduce noise). Not so good materials and the carpet is disabled but it can be used to see the time needed with four 580′s…
http://imageshack.us/photo/my-images/851/64004800115resize160012.jpg
Sorry my English.:)
Thanks for the detailed info! I’ll be sure to include this data in my update.
This was the most insightful article I’ve read in a few months on GPU’s! Also the most helpful and easy to read since all the comments make sense.
I purchased a GTX 560Ti a few months ago and its got 2GB of mem on it. I haven’t really put it to the test in terms of rendering, but I have a few scenes I plan on testing over the weekend. I run Max2010 and am a recent architecture grad (but I like to do test renderings of weird models I make on the side). My previous card was a GTX 260 (8xx MB of GPU RAM or so). There is a question in this…
I was wondering if the max drivers that the Quadro’s have REALLY improve render times. I’ve been drooling over the Quadro 4000 but haven’t been able to commit to buying. I’ve only used mental ray as a renderer and will continue to do so until I figure out how to use iray…
But if I may chime in on the viewport performance of the gtx 560 card. It’s fairly smooth. I have a particle flow scene with +500,000 polys and its workable. Edges are a bit jagged when I rotate, FPS does take a hit. Rendering time of that scene is somewhat unacceptable. GTX 260 vs. GTX 560Ti render time difference is 10 minutes. Doing the GTX to GTX comparison piqued my curiosity about the quadro cards and their drivers.
I run these in a mid-tower with no liquid cooling. just a stock 120 fan, a V8 heatsink over the processor (i7 920 clocked @ 3.0 Ghz), 9 GB of 1066 Mhz Nanya RAM, and 120GB SSD.
If anyone could provide some insight on the performance boost and shorter render times due to the Quadro drivers, than I’ll proly stop dabbling with gaming cards and make the leap into the Quadro’s. Great blog!!
“I run Max2010″ – Keep in mind that if you want to use a GPU based rendering system with 3ds Max 2010 you’ll need to purchase an external one because 3ds Max 2010 does not include iray.
“I was wondering if the max drivers that the Quadro’s have REALLY improve render times.” – No. Render speed comes from the number of CUDA cores and their clock rate.
“If anyone could provide some insight on the performance boost and shorter render times due to the Quadro drivers” – The quadro drivers help with viewport performance but it’s the hardware itself that counts when rendering.
The basic benefits of Quadro cards are: Usually better viewport performance, but that can be subjective. Longevity when used in stressful scenarios (like rendering), and higher memory capacity (on some models) than the GTX cards.
Hi Jeff,
Great Article
Just for the sake of clarity.
If you need 3GB graphics card you shouldnt buy the GTX 590 because it has two chips of 1.5GB. right?
If your goal is rendering with it, then correct. The 3gb GTX 590 would only provide you with 1.5gb memory.
Hi Jeff,
Thanks for the quick reply
And what about this one, do you now if it is a single core?
Gainward GeForce® GTX 580 3072MB “Phantom”
I think the 580′s are single GPU/core models so I would assume the 3GB models do not have to split their memory allocations like on the 590′s.
Keep in mind that’s all assumptions on my part as I don’t have those here to personally say for certain and Google turns up some unclear answers as well.
Tjey don’t have split memory. There’s nothing to split them between
They just use double density chips, that’s all.
Truth be told, there’s nothing that would stop them from making a 6 GB card, with chips on the back side as well. But i don’t know if NVIDIA would like that since it would destroy most of their Quadro lineup and only lose in some floatin point stuff.
Thanks for the clarification!
Hi Jeff,
I found some valuable information here, thanks for taking the time to write this. I currently own a GTX 460 2GB card, running on an ASUS P6X58-E WS motherboard. I use it with 3DS Max 2010. I’m more concerned with render times then with viewport handling, while both are important the 460 seems to meet my viewport needs. I would like to buy another identical card very soon to run SLI. But, I’m finding a lot of conflicting information about running NonQuadro video cards in SLI with 3DS Max. Mostly I’m seeing people say that there is no benefit, that Nonquadro SLI is only good for gaming. Your article seems to suggest I could see a benefit in 3DS Max by running duel GTX 460 2GB cards in SLI mode. Is this true? I understand that the memory is not cumulative. But are the GPU effects? Would I see faster render times for GPU Image renders? I’m not a gamer so the cost would not be worth SLI for me without some added benefits in Max or Ps CS5 extended.
Thanks,
Dan
“Your article seems to suggest I could see a benefit in 3DS Max by running duel GTX 460 2GB cards in SLI mode. Is this true?” – I hope not because I’ve mentioned several times in my replies here and in some of my other iray posts that SLI is not needed for GPU rendering with iray. AFAIK SLI’ing GPU’s does not help with 3ds Max viewport performance either. I personally do not know of any benefit to SLI’ing GPU’s, Quadro or not, as far as 3ds Max is concerned (I don’t know how that relates to CS5 extended). If SLI’ing GPU’s does help in some way then I personally don’t know what it is.
Yes, you will definitely see faster iray render times by adding compatible GPU’s to your system. However, as of today I’ve not seen any rendering benefit from having the SLI cable attached to the GPUs.
Hope that clears it up.
Thank you Jeff
That did help my understanding. This is my first go around with multiple GPU’s. I’m trying to beat the learning curve.
Thanks for your patience
Dan
Guys, do yourself a favor and read my benchmark test over at if you plan to use 3dsmax:
http://www.the-area.com/forum/autodesk-3ds-max/autodesk-3ds-max–3ds-max-design-2012/wasted-1700-on-a-quadro-5000-when-a-350-card-works-just-as-good-help/
Basically, a $1700 card is the same as a $350 ATI card in 3dsmax 2012 with the new Nitrous viewport. My $1700 Quadro 5000 2.5GB ran as smooth as my $350, ATI 6970 2GB!!!
Get a HD6990 4GB for $800 and it should annihilate the Quadro 5000 2.5GB at $1700.
I’m going to allow this post but please keep in mind that this is primarily an article on RENDERING with GPU’s (as the title of the article states).
There’s intentionally not much info about viewport performance because that’s a completely different topic and very subjective from person to person and computer to computer.
So, let’s stay on point here because this article is rather lengthy as it is without cluttering it with unrelated info.
Hi Jeff! what a great article, thank you for sharing this information! I recently bought a Gainward GTX 560 Ti with 2GB of memory and iray is rendering more slowly than octane render [ I have 3ds MAX 2012 Design, made a test scene and compered the 2 ] so is this because iray is also using my other graphic card [ I have a 220 gt which I use for display only ] and thus is slowed down [ I cannot tell iray which graphic card to use, it appears 2 CUDA devices when rendering, in octane I can select which card to use for rendering ] or is iray more slow than octane render?
Thank you again Jeff!
p.s – if you already talked about this, please direct me to that link
It may be that iray is just slower than Octane. I haven’t used Octane in a while since I borked my license by upgrading parts on my workstation. Prior to that when I was using it I know one thing, Octane was crazy fast. I think I’ll shuffle over to their website now and get my license fixed so I can try it again.
PS – You can use the iray manager script to assign specific GPUs
In high end business and in production environment u need two things; speed,flexibility and cost effective. A tesla is around 2k us dollars, a good gtx around 200, so 10x gtx to 1 tesla. From what i know u can use 4x video cards max on a motherboard till now. The heat is not a real problem, whit 50$ u change the cooler to each one and even if u change all 4 video cards after 1 year u still are under the price of a single tesla. For the second factor, flexibility, u need to render and manage your scene for maximum efficiency and doing that u gain fast export time and rendering. In real production environment almost no one renders a big complex scene in 1 pass.So in the end Tesla is just a very very expensive non efficient video card for enthusiasts but not to good in real production environment. To give an example, 3 years ago on my workplace from that time we bought some workstations for motion graphics,commercials,arch-renderings, dell ,intel whit 2x quad core Xeons, 8 g RAM , 1 TB raid , and each whit 2x Quadro FX 1700 512 vram each,hardware sli, 1000$ bout video cards at that time for a sngle workstation. No matter what tests i have done,or in any big scene, even 3ds max view port or after effects, the video cards where much slower then a single nvidia 7300 gt 128 bit whit 256 vram. on a dual core amd home pc, regarding the video test and the number of poligons in viewport or opengl in after effects. So i learn my lesson, and in real production environment, u need speed and flexibility and cost effective solution firts.
Hey! Nice article! Greetings from Brazil! =]
This is a one of a kind article I’d say. Interesting. On a budget, do you think the GTX 550ti (2 GB Version) could also be a good choice? It does have lower number of cores (192) but has 2 GB memory. Although, I read someone saying that his Max 2012 Design crashed upon rendering and he even had the latest CUDA 4.0 drivers. Before that he had an older version of CUDA and at that point, his CPU was only 1% utilized during rendering which means only the GPU was used by iray.
So what would you say about the GTX 550ti?
“his CPU was only 1% utilized during rendering which means only the GPU was used by iray.” – Then he has something odd going on because by default iray uses the CPU(s) in addition to any compatible GPU(s). There are tools & scripts available to manually configure those options.
“So what would you say about the GTX 550ti?” – I’d probably skip the 550ti for a 3GB 580. For something less pricey than the 3GB 580′s you can look into something like THIS that also has 2GB of memory but has 672 stream processors. As always, if you plan on rendering with any GTX type card make sure it has ample cooling.
Yes, the 460 could be considered but I guess the scene should not be more than 5 – 6 million polygons plus the texture quality also counts.
As a workaround on a 2 GB card’s limitation; can one render in layers to avoid memory overflow and then composite the multiple rendered layers back in something like Premier? It would be a hassle I know. All those alphas, deciding on using straight alphas or premults etc.
What would you recomend for iray (3ds max) and autodesk showcase a quadro 600 or a geforce gtx 460
For rendering you’d probably be happier with teh GTX 460. Provided your scenes fit onto the memory of it, the 460 will render your scenes faster than the Quadro 600. I can’t comment on how it will perform in showcase as I’ve never used it.
If you do go with a GTX card, watch the temps while rendering and try to keep it as cool as you can.
Well, if you where talking about Octane – I have what to say about it now.
My wife bought it to try, and after testing, she likes it SO much, that now she works almost entirely in Octane. Last ~20 projects are made in Octane and even time needed for one project decreased dramatically (material tunning is very convenient and fast). And for the same quality of the same image – Octane gets dramatically faster on the same hardware (on four GTX580-3GB), especially for scenes with lot of glass, noise disappears dramatically faster (especially last beta 2.49 even more increased it). Not so good integration in 3ds Max as iRay has, but I wrote a new plugin for it (instead of official one), and Mary works without inconvenience.
Unfortunately iRay at the time do not has even little part of Octane abilities and is not growing as fast as Octane (the last one gets new version with new abilities every 2-3 weeks)..:( Even that Octane is in beta stage at the time (Mary uses last beta 2.49). Even the good 3ds Max integration of iRay at the moment can’t beat Octanes abilities above iRay… Tested in practice.:) The best of all GPU renders at the moment – tested Indigo 3.0, Arion, iRay, LuxRender… searched GPU-render with most abilities, speed and quality to efficiently load all purchased GPUs. Octane wins at the moment…
Will wait before iRay gets better… I think it will be in the future, but not now yet…
Sorry my English.:)
I just can’t get comfortable with using those external rendering applications. It just feels counter intuitive to export my scene as .obj and then light/texture in another application. Of course that’s just my own personal hang-up and I will have to keep trying to get around that.
I also wasn’t able to load my complex scenes in Octane. I may need to export them in sections and load them into octane to rebuild the scenes but that seems like more work than just sticking to what’s in 3ds Max (even if it’s slower).
Again, that’s just my own experiences and shouldn’t discourage anyone else from trying/using Octane, etc.. On the test scenes I have been able to load into Octane it was crazy fast. Plus it has some nice material options, etc. so it definitely is a great option for many people.
Yes, integration of iRay is a big deal. And the fact that it can use some Mental Ray materials etc… very convenient… This is the reason we will track new versions of iRay, waiting it will be faster and will become more needed features…
I have one question about VRAY RT and GPU rendering… the GPU can be used for production to save all frames for example in an animation? Or to make a high resolution render to be able to save this from ACTIVE SHADE ? I plan to buy a GTX 590 video card for rendering, but I dont know if is good for this, or just for previews
Please let me know. Thanks
Yes, you can save the rendered images from VRay RT just like you can with Vray.
Thanks for the fast reply
. If you can give me one more advice would be great. What you recommend for a new render computer
2 dual xeon 5650 system, or
i7 system with some very good video card
I heard that ati 6990 is a very good video card. The second option would be 590gtx
http://www.guru3d.com/article/radeon-hd-6990-review/22
“What you recommend for a new render computer” - Depends on your scenes and what render engine you plan on using. The GPU’s can be blistering fast on some scenes (exterior / product-vis) and rather slow with others (architectural interiors). If you mostly work with architectural interior scenes then I’d probably stick wtih CPU based rendering applications at the moment so the dual xeon system would be good. If you work on scenes where you can benefit from GPUs (and ignore their current limitations), or you just don’t care about render times for arch-interiors, then I’d go with more GPUs.
“I heard that ati 6990 is a very good video card. The second option would be 590gtx” – Kind of the same answer as above, it totally depends on what render engine you plan on using. If you plan on using iray then currently you’ll need some kind of nvidia card for CUDA support. On the other hand if you’re going to use VRay RT then I think it may be compatible with it.
Thanks for your answer
I use only Vray as rendering engine, and I want to speed my render-time a bit, that’s why I wanted to switch to GPU rendering because I heard is faster than CPU rendering. I am working on some large renderings 9k max, mostly cars, and maybe some animations. I have seen your computer spec on the blog, and you use 3 GPU cards with a i7 processor. Also I see you did a lot of great car renders. You made the final renders with the GPU processors only? Also would you recommend QUADRO ,TESLA or GTX video card for this? If is possible to recommend something not very expensive
Yes, most all of the images on my website here are rendered with iray/GPU. For GPU rendering I’d only venture into the Quadro/Tesla arena if you find you need more than 3GB of memory (current max of GTX cards) to render your scenes. If you do use GTX cards for rendering then make sure to have plenty of ventilation/cooling and/or under-clock them a little to help keep them cool if they run for long periods of time. For VRay RT you may want to ask around on the VRay RT forum about ati card performance. I think they are compatible with VRay RT, run cooler/draw less voltage, and may be less expensive than GTX/Quadro/Tesla options. Keep in mind that’s second hand info to me, I have no personal experience with ati gpu’s.
NOTE: All of this advice is specific to rendering with GPUs. 3ds Max viewport performance is another animal all together and I avoid those discussions here because even with similar hardware people’s viewport experiences can be completely different due to so many varying factors (drivers, types of scenes, etc.)
speaking about CUDA , how do we enable it or use it ??? i just got my self a GTX 570 but havent noticed any differances in rendering, i wanted the telsa but i was shocked when i saw the price
Render engines like iray use the CUDA cores on the GPUs so you’d need to use a rendering application like iray in order to render with your GPUs/CUDA.
you’re right, i noticed iray rendering got faster a little, but every time i render .. a bright white dots starts to display on the image, so i stayed with mental ray, any way thanks for the reply
I know this post is about graphic cards mostly but, I just bought 16GB memory to my computer. How do I find out in 3DSMAX how much it is currently using for textures, (I can check in windows how much whole max application is using at the moment) but is there a way to see inside max what it is currently using for textures only?
In my main work station I have the same motherboard, i7 2600k OC, and 16 gigs of memory. I have 2 Quadro 4000 sli. It works great.
my render nodes x4
i3k oc to 3.75ghz and 8 gigs of memory. I use mentalray 98% of the time.
I was thinking of getting GeForce GTX 580 (Fermi) 3072MB
on paper they looks great.
GPU : GeForce GTX 580 (Fermi)
Core Clock: 832MHz
Shader Clock: 1747MHz
Stream Processors: 512 Processor Cores
Memory
Effective Memory Clock: 4200MHz
Memory Size: 3072MB
Memory Interface
384-bit
Memory Type
GDDR5
Any input?
I can’t offer much advice on using a render farm with the GPU’s as I only have my primary workstation configured for GPU rendering. That being said I would imagine that provided your scenes will fit onto the 3GB card, you should be good to go. The 580 should provide some fast results (depending on the scene of course). Also, try to keep the GPU(s) running as cool as you can. Good airflow in your case and even a slight GPU under-clock can help with that if needed.
jeff, first of all let me thank you for your great blog. my queries regarding a GPU rendering SETUP have been almost sorted out thanks to the relevant (though at times repetitive) questions and your very patient and enlightening responses.
unfortunately, i guess i ll have to wait for some time for GPUs with higher memory. cant afford tesla
having said that, i would like to put forth some questions related to software part of CG and rendering with GPUs.
1) i read on the spot3d website – http://www.spot3d.com/vray/help/rt100/render_3dsmax.htm “Note that V-Ray RT must be selected as an ActiveShade renderer; it will not work as a production renderer”. my question – is there any way to get a good production output from v-rayRT? if yes then how? if not, then is it possible in i-ray since ur really doing amazing stuff with it.
2) v-ray RT GPU doesnt render displacement using VRayDisplacementMod (http://www.spot3d.com/vray/help/rt100/unsupported_features.htm) except with the 3dsmax default displace which is not good at all. does iray on GPU render good displacements?
3) for architectural renderings of very very high quality and huge resolutions, would u advise me to try iray rather than v-ray (GPU or otherwise) which i use presently?
4) if you have any data about comparitive memory usage while rendering betn vray & iray, kindly share with us….
waiting with fingers crossed for a 6GB GTX …….
““Note that V-Ray RT must be selected as an ActiveShade renderer; it will not work as a production renderer”. my question – is there any way to get a good production output from v-rayRT? if yes then how? if not, then is it possible in i-ray since ur really doing amazing stuff with it.” – I think you’ve hit on something that has confused a lot of people. They aren’t saying you can’t use VRay R/T for production work. They are saying you need to assign VRay R/T as the active shade rendering application in 3ds Max instead of the production rendering application. Assigning VRay R/T as the active shade rendering application allows you to get the real-time updates, viewport activity, etc…it has nothing to do with your rendered output from VRay R/T.
“does iray on GPU render good displacements?” – Yes, iray does handle displacement. However, it does consume GPU memory so you have to factor that into your GPU memory usage and make sure you have enough.
“for architectural renderings of very very high quality and huge resolutions, would u advise me to try iray rather than v-ray (GPU or otherwise) which i use presently?” – If you mainly render architectural interiors and/or images larger than 10k resolution then I’d recommend sticking with Vray if you already own a copy or possibly Thea Render if you don’t already own a copy of VRay. I didn’t say mental ray because of the area light emission issues it has as I feel that has a significant impact on architectural interior scenes.
Rendering architectural interior scenes, especially those with small windows and/or few small light sources, with GPU applications can require a lot of processing time and/or GPU hardware. For reference I posted some examples with the render times HERE. Forty minutes for the daytime version isn’t too bad, of course it would be if this needed to be an animation. But the night render ran for six hours using a Quadro 6000, Tesla 2070, and Tesla 2050. It’s only using two small light sources so that’s why it took longer to render than the daylight version. That type of scene is more efficient to render with CPU based applications (IMHO).
“if you have any data about comparitive memory usage while rendering betn vray & iray, kindly share with us….” – At this point in time I don’t but I can try to add that in the future.
I guess I ll stick to v-ray for the moment.
But guess its not gonna be a long wait for a superior iRay and a GTX with lots of mem…..
Thanks again…..
Hi jeff.
Its really nice article down here.
I’m planning upgrading my cards. Right now i’m using 8800 GT, 3DS Max 2011, and V-Ray.
Just targeting 560ti (1 GB or more) for replacement.
Please let me know your opinion for this.
Thanks jeff.
Regards.
Bread.
If you plan on trying out GPU rendering with VRay R/T then I’d look into a 2-3GB GTX card (3GB if at all possible). I wouldn’t bother with a 1GB model because you’ll find it difficult to fit a scene, textures, and large frame buffer on 1GB. If you’re not going to use VRay R/T but VRay instead, then it doesn’t matter as much since VRay is CPU based and won’t use your GPU for rendering.
Thanks 4 the reply Jeff.
I’ll take the 2 Gigs card then.
3Gigs card too much for me.
Thank you.
Regards.
Bread.
hi jeff
Thanks for nice article , can i sli quadro 6000 or 5000 with gtx 590 or two 590 (quadro 6000 + 2*gtx590 ) ? If it is possible , Which memory used (quadro`s 6gb memory or gtx 590 3gb ) ? sorry me for my eng
I don’t think you can SLI a Quadro with a GTX, and I’m not sure why you’d want to. Today in GPU rendering the memory of the GPUs is not shared and can not be shared. SLI won’t help anything, in fact it can actually decrease rendering performance slightly. So as of today SLI is still best suited for increasing gaming performance with GTX type cards as it doesn’t help with GPU rendering.
thanks for reply
what mix would you suggest is the best choice for rendering using iray under 2000$ budget (for vga ) ? to the extent of my knowledge a set of four GTX 580s (3gb) from nvidia is ideal… because i saw a case with four 580 , in previous comment , is it good for rendering ?
and i think possible to sli quadro with gtx : http://www.titanuscomputers.com/ProductDetails.asp?ProductCode=X150
and another question , which memory will be used when i sli ( 3gb + 1.5gb ) ? bigger memory will be use or less memory ?
“to the extent of my knowledge a set of four GTX 580s (3gb) from nvidia is ideal… because i saw a case with four 580 , in previous comment , is it good for rendering ?” – It should work fine as long as: 1. Your scenes fit onto the memory footprint of whatever card(s) you use. 2. You keep the GTX cards running cool enough as to not damage them during long render sessions.
“and i think possible to sli quadro with gtx” – It may be, I dunno…but again I just don’t know why you’d want to. It will not improve GPU render speeds. From what I’ve read SLI’ing GPU’s for rendering will actually decrease the render performance slightly. SLI’ing GPU’s will not allow you to combine GPU memory for rendering. And AFAIK it doesn’t help with viewport performance either.
and another question , which memory will be used when i sli ( 3gb + 1.5gb ) ? bigger memory will be use or less memory? – Currently SLI does not help you with GPU memory when it comes to rendering. Your scenes need to fit onto each GPU. Therefore if a scene doesn’t fit onto a GPU it will not be used for rendering. If it does fit onto the GPU memory footprint it will be used, whether it’s SLI’d to another GPU or not.
Hi again
bought the gtx 580 3GB yestday for gpu render on my system already having a quadro fx580.
my intention is to use the quadro only for display and gtx only for rendering. Hdmi/dvi display cables connected only to quadro.
Quadro drivers were already installed since a while. Working flawless & smooth…
Installed the gtx drivers from the cd included.
All fine till I try to use it as an opencl device with vray RT gpu. It says device could not be assigned. Although its showing both devices present. Just cant use either of them…
Was working fine with the quadro gpu earlier.
Even max is giving errors on startup and showing blank white viewports.
Any tips shall be highly appreciated…..
Rgds.
further to above issue, just discovered that vray-RT GPU is simply unable to use the GTX 580 3GB!!
tried it with only the GTX 580 (removed the quadro). fresh & clean install of drivers. still vray standalone server simply crashes every time u initiate a render.
sys config:
core-i7 870
8 gigs ram
palit gtx 580 3 gb
quadro fx 580 (removed right now)
max2011, vray 2.0
any help guys??
I’d say your best bet is to check over at the VRay R/T forum. I remember reading a post over there where I think someone had problems with getting a 580 to work and I believe it was due to the VRay RT license server wasn’t running correctly as a service. So make sure you check that.
Hello Jeff,
I would like to buy a new graphic card GTX 590 for Maya.
Do you think if there is any problem between maya and this card? I can not find any information on autodesk’s site about this card.
Thank you!
Afraid I can’t provide much help on that one. Hopefully someone that uses Maya will chime in with some advice.
so, I render a lot of hi rez still images in maya & Vray.
if you had to get a new graphics card set up & money wasnt an issue what would you go for?
GTX’s
Quadro’s
Tesla’s
anyone?
Doesn’t really matter since VRay doesn’t use the GPU for calculations. If you meant VRay R/T using GPU mode then you’ll have to pick a GPU that has enough memory to hold your scenes & textures.
yeah sorry, i meant Vray RT. speed wise from reading here the GTX’s sound as fast ther teslas, but have less on board memory. If i am outputting big renders say 10k pixels does that use the gpu mem too? along with geometry & textures
Good question. I’m not 100% certain if the output size affects VRay R/T like it does (or did) with iray. In the latest version of iray released they say the output size doesn’t matter so that may have been something specific to iray 1.x and VRay R/T already handled it. I can’t find any info on how the frame buffer may or may not affect VRay R/T in GPU mode in their documentation. I’ve asked this question over at the VRay R/T forum and I’ll also post the response here once I receive one.
Answer: With VRay R/T (GPU) the output size does not matter to the GPU as the image is stored in the system memory instead.
Great post Jeff! Last week I tried to render some scenes in my boyfriend’s laptop (an Asus with i7 SB, 12 GB DDR3 ram and GTX 580 3 GB) but it disappointed me with the results:
Using 3ds Max 2011 + Vray 1.5 SP5 it worked just a bit faster than my current desktop pc (Core 2 Duo, 3 GB DDR3 and a 9400gt with 1GB)
And while using Vue xStream 9 it didn´t render the scene! Just crashed (while my pc did render it).
Is it because the GTX 580 is not certified for that kind of software or just because it was the mobile version of that video card?
I’m building a new desktop pc and I was considering to buy either Quadro 600 or Quadro 2000 (are the ones I can afford).
For a laptop… Which videocard should work?
Thanks in advance!
http://www.newegg.com/Product/Product.aspx?Item=N82E16814130733
This my friends is a water cooled gtx580 with 3072 MB of memory, and it only costs $750. Get two of those in SLI and you are sitting pretty.
I wonder if it is worth it to get a Quadro FX5800 w 4GB onboard for Solidworks. You can see them going for as low as 800US on ebay, I guess the new Tesla cards are driving old new stock prices down.
How would it compare to having 2 gaming cards in SLI if bought with the same budget?
You don’t SLI cards together for rendering. It only slows down the process (slightly) and it doesn’t combine the memory of the GPUs as many think it will. The Quadro FX5800 has 240 CUDA cores while something like a GTX 580 will have around 512 CUDA cores running at a much higher clock speed. So specific to rendering the GTX card will be much faster. However, as you know you can’t find a GTX card with more than 3GB of memory. So while the Quadro FX5800 would be slower to render it can hold more complex scenes.
This is the tradeoff that each of us has to figure out…speed vs. memory footprint….which is more important to you. I usually lean towards the memory side of things because if the card can’t hold the scene it’s useless no matter how fast it is. If you decide to go with a GTX based card because you can fit your scenes onto 3GB then just keep a close eye on the temps as they can get very hot while rendering if there’s not ample cooling.
Hi Jeff.
When rendering on a 560Ti, 570 or 580, what would you concider ok when it comes to load temps on the graphics card? With long render times in mind for animations, it would of course be good to keep things as cool as possible. I’ve noticed that I can reduce the temps by at least 10 degrees C by ducting cool room air directly into the fans on the card. I’ve been rendering for hours both on temps in the low seventies with normal ventilation, and at around 60 degrees C as load temps during renders with air ducted in.
60c to 70c is really good IMHO for a GPU under load. They are typically rated for higher temps than something like the CPU. I think nvidia cards can reach 100c to 120c for short periods without damage, but I certainly wouldn’t run them at that range for long. Bottom line I think if you keep them under 85c while rendering you’re doing pretty good.
Reassuring to read this. I have several rigs like that, running a small production studio for business in my house, and that equipment is actually part of the heating plans for my house this winter. For good reasons I’m sure you agree, the continuous heat output is on the generous side.
Hi Jeff, first of all I’m a big fun of UR work, and where I live (Poland) even press is writing of Ur project for Mercedes
So my point is – I’m acually buying a new workstation. I have’t tried Iray (I’m working with mr only) so my choice is 2x Xeon 5650 @ 4,2 Mhz, 48 Gb of Ram and I have problem with GPU, I want to buy 2 or 3 GTX 580 3Gb just to have a choice between CPU and GPU rendering, but U said that buying a GPU it’s acually determineted by a RAM usage in Max ? I’m doing ONLY arch vis. My typical scene is about 4-5 millions polys. Mayby I should by 6Gb Quadro? It will be great of U to answer.
I just have few questions after reading this post. I know that when i combine 2 or 3 GTX type cards I won’t have their combine on board memory, but will i have dubble/or higher numers of CUDA cores? 4 example if i get 3 GTX 580 3GB i would have 1536 CUDA cores to go with and work togeather ? And next question – How/or where can I find information about the consumtion of RAM needed for textures and overall max scens. Sorry 4 my english, it’s hard to tell exacly what U fink if U don’t speek english from birth
Keep up this great work Ur doing.
Cheers
“will i have dubble/or higher numers of CUDA cores? 4 example if i get 3 GTX 580 3GB i would have 1536 CUDA cores to go with and work togeather?” – Yes, that’s basically correct. Each new GPU added to your system is like adding a new render node to a render farm. Each GPU you add to your machine will give you more processing power for rendering with applications like iray, thus speeding up the render process.
“How/or where can I find information about the consumtion of RAM needed for textures and overall max scens.” – Shane Griffith provided a guideline on that in his iray FAQ. He states: “For estimating memory usage, budget about 1 GB per 8 million triangles, to which you must also add 3 bytes/pixel for any referenced bitmaps.”
Thx very much, I’m really greatfull 4 UR answer, it helps me a lot
When I buy my work station I’ll post some time test comparing the same scene, rendered with Iray (3xGTX 580 3GB) and with mr on 2x Xeon 5650 @ 4,2 
This should be interesting 
Thanks once again Jeff.
That setup should be crazy fast!!
Thank you very much for the useful website.
Considering the high number of Cuda cores and 1792mb memory, I feel like a GTX 295 may be a good choice for Iray. But I don’t know much about “Fermi” thing, neither do I about the several frequency specs… Can you compare a 470 and 295 in terms of iray performance?
Thank you
Looks like the GTX 295 is a split GPU unit (240 CUDA cores x 2), similar to the newer 590 model GPUs. The problem with that is the memory is shared between the GPUs so instead of having 1792mb of memory you have 898mb x 2. Therefore the 3ds Max scenes you render with GPU based applications like iray will need to fit within that 898mb range, which isn’t a lot of space. With that in mind, if I had to choose between a GTX 295 vs. GTX 470 for use with GPU rendering, I would pick the 470 because it has more memory available for scenes (1280mb) than the 295 (898 mb x2).
Ultimately though I believe you may be better served by locating a GPU with at least 2048mb because I think being restricted to 1gb or less for rendering will prove to be difficult.
I gathered this info from the nvidia website here:
GTX 295 (refer to the Specifications tab)
GTX 470
Various nvidia geforce GPU models data
FWIW info on the FERMI architecture can be found HERE. I read somewhere the other day that a newer GPU architecture may be right around the corner though…and you know how that goes, newer = faster, etc..
Greetings Jeff , I’ve read your blog teraly and I like it mucho. Answer me if you could, how much gpu memory takes to render 03 restaurant and 03 interior on your portfolio. Thx much
Sure, it uses 908mb of GPU memory.
Scene info: 3,093,471 poly / 1,781,655 verts
THX much, and do you have some similar project in video and would you say that 3gb would be enough of gpu memory for lets say walk-through from entrance hall to living room exiting to yard with 3ds max in that kind of finish as those projects i mentioned earlier?…just to tell you that I have really powerful gpu Ati V7900, however can not use gpu processing on that so i am looking into changing it to gtx580 3gb.(I have researched that gtx 580 is possible x7 on SR-2 motherboard from EVGA with water cooling setup that would have to be double at least cause looping from one to another would rise temp. a lot so you could two or three cards at a time.) I really admire your dedication to finding the time and answering all of our questions here.
“would you say that 3gb would be enough of gpu memory for lets say walk-through from entrance hall to living room exiting to yard with 3ds max in that kind of finish as those projects i mentioned earlier?” – Hard to say. Most of the answer would depend on how much geometry is used, how detailed the geometry is, how many texture maps are used and how large they are. I just added some memory calculating guidelines/info into the FAQ here (point #14) that may help you figure out what may work or not.
You can also of course split the scene up into parts and render separately as I mentioned in my exterior scene articles HERE and HERE. Whether or not that ends up being more efficient or less efficient than rendering it all with a CPU based rendering application is up for debate (mainly would depend on render times between the two processes).
Thank you so much and here is for you something interestig 7 gtx 580 on one system with pics; http://www.evga.com/forums/tm.aspx?high=&m=757729&mpage=1#757729
Greetings from Pakistan!
Great stuff being talked about here; and it so proves to be the only meaningful discussion/review/insight to a number of answers I was looking for.
I have recently come in this construction business, playing with 3dmax was a thing of the past for me, during the days of 3dlabs Wildcat/Wildfire cards! Now I have to set up a rig primarily to be used for architectural visualization purposes both interior and exterior. Typically my views will be that of a double storey house, with plenty of interior renderings going towards a portfolio for a prospective buyer to see how the house, construction looks like.
Being in pakistan, at best I have ATI HD6990 (4gb) and Nvidia GTX580 (1.5gb) available in the market.
Software I would mainly be using is 3ds Max 2012 design, autocad 2012.
Hardware : supermicro workstation mobo, fitted with 2×5560 xeon, 8gb fb ram, 2x147gb SCSI
I need advice on which GPU to purchase. If I go for gtx580, I am limited with memory, but then iray will support it. If I go ahead with ATI, from what I read here vray r/t will support it, but iray will not. After googling, it says that 6990 is a dual GPU card, so in essence its memory would also be split like that of gtx590 – but it provides for 2gb on each GPU.
Question will vray r/t take advantage of both the GPUs on the HD6990? Normal consensus is to go along with the CUDA cores on the nvidia, and ideally if I could I would get try and get the gtx580 with 3gb – but import and customs is a hassle, not to mention warranty issues!
I am not expecting to do renders larger than 1600×1200 at worst case. Goal is to do some clean realistic renderings to give a feel of the space, form and shadows.
P.S. we do not have access to quadros or firepro here
Thanks,
Raza
Hi Jeff,
Maximum respect for your patience and thanks for your dedication.
- How will Nvidia Maximus implementation be affecting your workflow?
Does it mean that the max render size will now be the Tesla’s amount of DDR5? Leaving the viewport task to Quadro?
If that is the case, say a Quadro 2000 + Tesla C2075 would be a decent set up, “affordable” yet efficient…I hope…Tell me I’m right…please
http://www.nvidia.com/object/maximus.html
- For those willing to use GTX frying pans (just kidding, I’m doing that myself…Getting priorities right!…Right?:).
GTX 580 Liquid Cooled Graphics Card are available for $580 for limited time at:
http://www3.pny.com/GTX-580-Liquid-Cooled-Graphics-Card-font-colordc0431LIMITED-TIME-BUNDLEfont-P3015C470.aspx
And hopefully soon, DIY Liquid Cooled GPU kits will be available.
Wicked work Jeff !!!
Best I can tell the “Maximus” thing is mainly a workflow for using Quadro/Tesla combinations where a user assigns the test renderings to the Tesla and can continue to work since the primary display GPU (quadro) is not tied up with rendering. Then for a final render a user can activate all GPUs in the workstation if desired. If that’s what “Maximus” is then that’s basically been my workflow for the past year with iray + iray manager script so I don’t see anything changing for me personally. I was hoping for an announcement about some new GPUs with more on-board memory!
“If that is the case, say a Quadro 2000 + Tesla C2075 would be a decent set up, “affordable” yet efficient…I hope…Tell me I’m right…please
” – Correct. You could use a less expensive quadro as your primary display device and then the Tesla c2075 (6gb model), or a few of them, specific for rendering since the Quadro 2000 and tesla cards are far less expensive than the Quadro 6000.
“Merci” for your swift answer Jeff.
Hey, I Have a couple of questions I was wondering if you could help me out. I will be doing a workstation build soon which will be used mainly for AutoCAD, Revit (architecture), 3D modeling software, and Redering software, May also do some Video editing and gaming.
1.) I’m trying to decide between a Quadro 4000 and GTX590 or depending on when I can do the build possibly those cards’ Kepler successors. For the Tasks listed above which would you suggest? They both are available for ~$700 and the 590 specs completely KILL the Quadro’s Except in onboard memory. the Quadro is 2GB, and the 590 is 1.5GB (well actually 3GB, but its like 2 “GPU’s” on 1 physical CARD, so taking what you said into consideration about on board memory not being cumulative, I’m counting it as 1.5GB) Also I know the Quadro’s drivers play a role so I’m at a loss in deciding which one is better.
2.) You mentioned you had a GTX/Quadro/Tesla all in the same machine running together. How did you accomplish that? Upon reading all the features for the Quadro 4000 on the Nvidia webpage, Its says that Quadro’s are capable of SLI, but only from “approved” platforms…..meaning only when you buy a pre built workstation form select manufactures that come with 2 quadro cards already. So I’m assuming Nvidia just somehow “locks” that capability in software. So I’m curious how you got a quadro in sli at all, let alone with a Geforce and tesla card. Did you have to use a hacked driver or something?
If you could help me out I’d greatly appreciate it, thanks.
I believe you’ll find answers in my iray/GPU FAQ here: http://jeffpatton.net/2011/10/iray-and-gpu-faq/
I have recently installed a gtx 460 I used the gpu with the iray but when i try to do a render of more than 1300×1000 the 3ds max no response, shoud I download any driver ? or what can I do to improve the performance.
If it renders at resolutions smaller than 1300×1000 then it sounds like you may be running out of GPU memory rather that it being a driver issue. This would be especially true if you’re using iray ver.1 instead of Ver.2 in the latest 2012 advantage pack for 3ds Max. I say that because in Ver.1 the rendered image size counted towards memory usage whereas it doesn’t in the latest version of iray.
yes I am using the 3ds max student version, perhaps the 2013 version will have iray 2.0, thanks
Hi Jeff
Keep up the great work, you’re a star.
Quick question about iray2. You mentioned something about the rendered image size no longer counts to the memory usage on your card. Does this mean that you can now render crazy big resolutions on 1gb cards so long as the model and textures fit within that 1gb?
Also, I found nitrous uses up a big chunk of memory. So much in fact that on my gtx470 with 1.25gb I had to dump nitrous and revert to direct3d to fit the 3500×2650 render on the card. Anyone else encounter that?
Cheers
“Does this mean that you can now render crazy big resolutions on 1gb cards so long as the model and textures fit within that 1gb?” – I don’t have a 1gb card here to verify, but I would imagine it will help a lot by of offloading the frame buffer size. However, I don’t know if it will be enough to allow for crazy big renders from 1gb GPUs. Maybe someone will chime in that has tried it with a 1GB GPU.
On nitrous – Yeah it does consume more memory. You may want to try disabling the progressive refinement to see if that helps. But the improved look/functionality of the viewports do require more from the GPU.
So with GTX and Quadro cards what is the difference for viewport display? i thought it was a openGL and direct3D thing? I thought quadro’s are based off openGL and GTX off direct3D? does this even matter anymore?
2nd… am mainly a FX artist and dont deal with geometry that much. Does particles work the same way in viewport as my scenes can get pretty heavy with millions of particle counts. Any info would be appreciated.
Thanks!
One more thing so i been comparing the quadro 4000 vs the GTX 560 TI and they both have 2gb of memory but the gtx has almost double hte cudas and speed… so why is the quadro almost 4 times more expensive? the gtx is 230 and the quadro 4000 is almost 800!
THanks again!
There’s a document floating around the nvidia site somwhere that explains the difference between Quadro/Tesla/GTX GPUs, but I can’t seem to locate it at the moment. Check around the nvidia forums for more info, like THIS.
“So with GTX and Quadro cards what is the difference for viewport display?” – Afraid I don’t have a lot of info on how particles & the viewports play together these days with the various GPUs. You may want to ask around at CGtalk in the 3ds Max forum as I’m sure there are people there that can give you a more solid answer on particle viewport performance with various GPUs.
Hi nice blog thanks
I order Gainward GTX 580 3gb Phantom. Hope it fit big scene.
I just want ask if this is problem with todays GPUs or 3d max. Im not a programmer but for me it sounds like 3d max handler render wrong maybe shuld made a better soft to render useing of GPU memmory. No matter if GPUs have 3gb or is splitted up to 2 x 1500MB.
I think 3d max shuld do something about this. I dont think problem is GPUs as they today is verry powerful.
Sorry for bad english.
Great blog jeff
Jeff.. thank you for this article. I have a special problem that I am seeking an answer to.
My workstations are older 8 core Intel Mac Pros running win7 64. They are PCI-e V1.0
I really need to upgrade from the consumer card I have now and I am not sure what my options are. Can I get something like a GTX card and run it? albeit.. slower than it should run?
I think the PCI-e V2 cards will run on PCI-e V1 slots, but slower as you mentioned. To be sure, I would contact/ask the vendor that you’re considering. Most, like EVGA have a forum where I feel you’ll get a more solid answer from to help with your purchase decision.
Thanks Jeff.. And thanks for all of your instructional works throughout the years.
Hi all
Just want to say that evga 3 gb copper 2 have some problem with running mutiple screens.
“EVGA GeForce GTX 580 3072MB Hydro Copper 2″
I planned order this card from beginning but changed my mind after reading some topic on evga site about this problem with running multiple screens. I ordered Gainward GTX 580 3gb Phantom instead, its without water cooling but you can add any GTX 580 watercoler block later. Gainward is also lilte cheaper.
I admire your patience and keeping up with so many comments!. I’ll try to make my questions few and short. We have a small production film company. We are looking for ways to increase our production performance. We were thinking about upgrading our workstation for rendering purposes. Our current production gig require us to use a lot of high def VFX b-roll shots and 400-1500 frame CGI animations.
We’ve used the traditional way to render animations thus far, and it works fine, but depending on the amount of details you put in your animations it can quite slow at times. We are starting to see that we need a faster way to output frames while keeping costs down. GPU rendering seems to fit the bill. We use a lot of composition, however it’s always nice to have a faster way to render our comp frames..
I am personally trying to understand the limits of these cards and their programs. So excuse my ignorance on the topic.
I wanted to ask you. How does Iray handle visual effects like this simple test animation we have done:
http://www.youtube.com/watch?v=zlrWQUaJaoU&list=WL10C2E3CA5042A72C&feature=mh_lolz
The animation was rendered in LW9.6, and we used a free model from sci-fi meshes. The model only has a diffuse map, bump map, and a specular map applied to it. We put a texture image for the background, and used a bit of LW’s hypervoxel for the gas cloud. For the camera effects. Only motion blur. Rendered at 720HD. Lots of antialiasing!
Is it possible to do “outer space” animations in Iray, or Octane render for that matter, like the one above?. What would be the limits or problems we may encounter?
Thanks for your helpful blog!
You could probably use something like iray or octane for space scene. However I think it would be more efficient to use a render engine without as many restrictions as applications like iray and octane present.
Hi , Great information here thank you.
My current system i7-2600k with 2x Radeon 4870 modded to fireGL 8750. I want to get into Vray RT my scenes are 7million poly, complete car renders.
So I am thinking 2x 580 3gb cards. My question is one of my pcie slots is X4 and the other X16, is this going to kill the performance of one of the cards. My Radeons dont seem to suffer on the X4 slot. Is this a bad idea?
Thanks
another question sorry regarding my post, is could i render with two Geforce 590s? for VRAY RT if one of my slots is X4 versus X16 ? or would that be a bad idea?
I don’t think you’ll see much difference in GPU render speed between the 4x slots vs. the 16x PCIe slots. I think the main area where you’ll see a difference is in translation time/GPU loading/unloading time. I imagine that process would take longer on a 4x slot than a 16x slot. Once the scene has been loaded onto the GPU(s) the render time should not be affected.
Hi Jeff, I hope you can help me, I have Ati firepro V7900 which is used as main card and i am planing to get a second computer with gtx580 3gb onboard to use as a render node. Do you know if 3ds max 2012 will utilise that gtx and is there gonna be any problems. I ask because when i add nvidia card on my computer with ati v7900 the whole system becomes very very slugish. I know that nvidia in their drivers does that so that you cant use ati and nvidia on one system however if there was as i wrote one system as workstation and second as a node would you know if there would be problems?
“i am planing to get a second computer with gtx580 3gb onboard to use as a render node. Do you know if 3ds max 2012 will utilise that gtx” – I can’t say for certain as I haven’t tried that process here. However, I would guess that as long as the scene is configured to use any available GPU it will render once sent to the render node that has the nvidia GPU.
Jeff Hi, nice article there… very informative. Knowing Nvidia and how they’ve been touting their new Maximus technology, I kind of suspected those results from the Tesla cards. I’ve been wondering if the Tesla c2075 cards really posts any real world advantages over say a GTX 580, in render times (aside from the 6GB ram vs 3GB on the GTX card). And based on your results with the 470 vs the 2050, I’m very much in doubt that the tesla, given its $2000 price point is going to be 4 times faster than a $500 GTX card to justify the price difference. In fact, I wouldn’t be surprised if they weren’t closer to equal in performance. And for the sake of this argument, I’m ignoring the difference in RAM.
I’m getting ready to build a render farm, and I’m stuck at a crossroads as far as which technology to go with. It’s clear that the digital world is exploring moving further towards GPU rendering over CPU. Even though GPU rendering technologies has limitations right now with shader support etc… But my thinking is that in 12-18 months, things will improve and continue to do so in the thereafter. And I’m willing to bite he bullet and hold out for better compatibility in the near future.
I priced 15 traditional render nodes at $300 each ($4500 total), with each box built of:
AMD Athlon X II 630’s 2.66GHz
8GB DDR3 1333 RAM
motherboard & harddrive and case
Then alternatively, I looked at a 2 node GPU based render farm solution. With each node valued at $2300 each ($4600 total):
Intel Core i-3 2120 3.3GHz
16GB DDR3 1600
(3) EVGA GTX 580’s, 512 cores, 3GB RAM
motherboard & harddrive and case
Ok, so from a price point, I’m looking to spend about the same. But from a performance point, how do you think I would stack up comparing these two set-ups. Of course, you could only speculate here as I don’t expect you to run out and pick up all this hardware to run tests, lol. To break things down a bit, each of the 6 total GTX580’s are costing me aprox $760 with their supporting hardware. That’s equivalent to 2.5 traditional render nodes. Ok, this is a more manageable scale of things. The big question is, does 1 GTX580 outperform 3 of the CPU nodes, or does it underperform 2 of the CPU nodes. Do you have any insight on how you think this battle would end, and do you know anywhere I could get some valuable feedback with this particular situation?.. Thanks for your comments in advance.
HI Jeff,
I was wondering, if i could use i7 Extreme edition, MSI Big Bang Xpower II Mother board with GTX 580 4 way sli for gpu rendering and 1 regular vga card for the physx. I’ll be using it for GPU rendering with 3ds MAX 2012 and vray RT. is it possible to use 6 EVGA GTX 580 hyro Copper 2 using sli?
Thank you.
For both iray & VRay RT you can assign whatever combination of application compatible GPUs you wish to use for rendering. Same goes for PhysX, which you’ll assign via the nvidia control panel application. As for SLI, I’ll direct you to my iray and GPU FAQ (question 1).
Oh it wasnt after all! Hi Jeff, great article and sorry for doubleposting. I bought many of ur lessons and were very helpful keep up the good work. I was wondering if the EVGA 560Ti 2win uses the full 3gb RAM for rendering and doesnt behave like the 590 card which split the ram in 1.5 and 1.5. Im about to order the 560 ti 2win. Does it worth it? thanx a lot
I can’t find any detailed specs on a 3GB 560 ti 2win card to say for sure.
What about the MSI gtx 580 Lighting Xtreme edition? Does it got 1024 CUDA + full 3gb RAM ? U think is fast enough for a single gpu card?
You’ve just answered a question I’ve been searching for the answer to for MONTHS.
Thank you
Hi Jeff,
I came across this article which could be of interest to you:
http://www.techpowerup.com/161879/Sapphire-Toxic-Radeon-HD-7970-with-6GB-Memory-Showcased.html
Also, I was wondering if you had any issues with GTX and Quadro drivers on the same machine, running at the same time?
If so, how do you solve this?
Thx and many regards.
Your work rocks!!!
I don’t recall encountering any specific issues when I had ran a Quadro & GTX card together. However, back at that point in time the whole system seemed rather unstable so that could have been part of the issue (the main issue was with the motherboard).
Hi there Jeff. I’ve been finding this blog quite useful so far. I’m planning on building a new workstation in the next few months but am uncertain about certain aspects of workstation graphics. I’ve read plenty of benchmarks that say how much better something like an entry-level Quadro is for viewport performance compared to a consumer level card. It seems almost unbelievable considering the hardware differences. If I were to get, say, a Quadro 600 would it be able to run games too? I really could do with being able to throw around millions of polygons in the viewport without lag (which I just cannot stand with my current out-dated system and a gaming-level card) but I would also like to be able to test out the models I make on the latest game-engines which would probably run a lot better on a gaming card. So is there some sort of compromise that can be made? Could a modern GTX560 not match a low or mid-range Quadro for viewport performance and quality? Thanks in advance for your help.
“If I were to get, say, a Quadro 600 would it be able to run games too?” – I haven’t tried any really complex modern games with my Quadro so it’s difficult for me to say. I guess I need to bite the bullet and buy a modern game to try.
Hi Jeff, today i read this article about the new nvidia release. Check this and tell me what u think. http://www.evga.com/articles/00669/#GTX680FTW
For a single card i think worths its money; 1536 cuda, 4gb ram and watercooling! looks like a beast! what u say?
Having more memory is great. However, in the last few days some sites are reporting less than stellar computational performance from the 680′s. Like HERE, HERE, and
HERE(article was pulled for some reason).As with anything on the internets it’s probably best to have a wait and see outlook though. I’m sure someone will eventually get one or two of these cards and put it through a GPU rendering test and post some real-world rendering results. At least that’s what I’m personally waiting on before forming an opinion on the 680 series.
Hi Jeff,
First, congratulations for the great job you are doing. I´m plannig to buy one quadro and 2 tesla Gpus. I want to know how much memory will be disponible for me if i use this configuration:
quadro 6000 _____________ 6gb of ram
2 tesla c1060_____________ 5gb x 2 = 8gb of ram
I´m afraid to spend a lot of money and cant´t be able to use all i have.
Quadro 6000 = 6GB of ram.
Tesla c1060 = 4gb of ram.
GPU memory is not combined/cumulative (see my FAQ here). Therefore the hardware you mention is a bit of a mismatch IMHO. If your scenes fit within 4gb of memory then no need in purchasing the quadro 6000 with it’s 6gb. Likewise if you need 5gb or 6gb, no need in purchasing the tesla c1060′s because they won’t be used on scenes that require more than 4gb of memory.
Note: You may also want to wait and see what happens with the 680 gtx series. If you can get by with 4GB of memory then they may be a good solution. We don’t know yet though. I think we’re all waiting for someone to get their hands on one and test it first.
Thanks a lot Jeff,
I took a look on the FAQ before but considering the combination of only teslas and quadro i thought that it would increase the memory amount. Thats all clear now. Thanks again!
Thanks a lot for sharing such a valuable information. as far as i have learnt. some websites say GTX 680 is not that suitable for computing, so should we stick to 400 & 500 series cards?
I’m waiting to see actual numbers from a 680 used in GPU rendering before forming any conclusions on that subject. Apparently it’s going to be a little while before that’s possible since iray doesn’t run on the Kepler GPU’s at the moment.
Hi Jeff,
do you know whether you can mix Kepler’s with your existing setup of Quadro and Tesla’s cards? (Once Kepler is working with iray, of course).
Cheers,
Steve
Afraid I don’t know yet either.
Interesting article and great timing because I am at a crossroads myself. My system currently runs with a Quadro 2000 and here is my delima. My main focus is running Revit but I need the ability to render in MAX while I am working in Revit. I just ordered a GTX 580 because it is faster so I am wondering if I could assign the Quadro 2000 to run my display while telling Iray to render with the GTX while I continue to work in Revit?
I was wondering if it was possible for someone to write a script which would render the left hand side of the camera view and then the right hand view – which could then alleviate memory requirements on the gpu – these could then be combined in nuke / AE later. This might make small memory cards more useful in rendering. Any thoughts on this?
Thanks for the concise breakdown Jeff. Unbiased certainly cuts down on setup time for the scene and IMVHO produces more realistic results.
It’s a good thought, but I don’t think it’s possible with iray (or any GPU application) at the moment since geometry/textures/lights can’t be selectively assigned to a GPU at this point in time. I believe the only way something like this would be possible right now is if the original .max scene was physically split into two different .max files, like left.max & right.max. But of course if you physically split a scene in half then users will run into issues from the missing geometry/lights/etc. Missing reflections from one side to the other, incorrect illumination (missing lights or bounced/blocked light areas), etc. would all be issues.
Since I don’t foresee a 12+GB GPU on the immediate horizon I would hope the application developers are researching ways to make GPU rendering more efficient. Who knows, in time maybe they will find a way to do something similar to what you’ve outlined above. A big increase in memory performance would certainly open up GPU rendering for more types of scenes & users.
Hi Jeff,
what are your views on the future of the Maximus setup?
Nvidia were/are still pushing this setup, with cards based on Fermi technology.
I know it’s only speculation, but once a 6GB Kepler card is on the market, and working with iray, I wonder whether we’ll have a Maximus 2 setup.
Cheers,
Steve
If I remember correctly Maximus was mainly a keyword/buzzword/marketing term for using the primary GPU (quadro) to drive the displays and other GPUs (tesla) for rendering. As the 3ds Max viewports become more complex and taxing on the primary GPU I would imagine the ‘maximus’ configuration will be very helpful regardless whether it’s Kepler or Fermi based. This is especially true for people that want to use active shade render options.
ah, I see. I thought Maximus was a specific set up.
I’m keen to see what the new high end Kepler’s will be – I’m thinking of investing in a 6GB card and don’t know whether to hold off until the new cards come out, or go for a Tesla and add this to a GTX 560i I have, for now (as a kind of cheaper-ish GPU set up).
Use your best judgment but IMHO it might be a good idea to wait if at all possible. Like you said, let’s see what the Kepler’s do once they are working with iray. Especially if/when they announce a Kepler based tesla GPU. If there’s not a huge performance jump from FERMI to Kepler then maybe the “older” FERMI based tesla GPUs will be on sale for cheap.
Jeff, thank you for the excellent write up.
In my experience using Iray (which is vastly less than yours) – I found my GTX 580 w/3GB renders scenes nearly as fast or slightly faster than a machine sporting 2 tesla M2090′s. It both shocked and saddened me.
If the only benefit to buying a Quadro/Tesla is the that they handle the heat better, why wouldn’t someone just buy a gaming card and spend some money on a nice cooling setup?
See item #2 in my GPU Frequently Asked Questions list HERE.
thank you!
I would like to know if there is a utility that checks on current GPU VRAM utilization (edit: is it CPU-Z enough?), to know how much VRAM is actually available for the Renderer ?
and maybe consider switching to a low-end GPU just for display puproses, while using the GTX as dedicated computation device (ala Tesla) ?
Yes, most of the applications like CPU-Z or nvidiainspector will show you the available/used amount of memory on your GPU. Unless you’re using a Tesla 2075, they don’t show up.
thanks for reply
and Which you choose for iray ? gtx 680 4gb or 580 3gb ?
http://www.anandtech.com/show/5699/nvidia-geforce-gtx-680-review/17
You’ll have to use the 580 series until nvidia/autodesk supplies a recompiled version of iray that’s compatible with the 680/Kepler based GPUs.
thanks a lot
Hey jeff just stumbled onto you and your corner of the interwebz here. Thanks for the comparisons- I have a custom PC with the GTX 470 but rendering in Bunkspeed Shot. Thought I’d chime in with a comment on SLI and value. Even though the Quadro series states SLI support there are several types/modes of SLI and they don’t really share (the whole point) till you get into the Quadro 5000 and 6000. This is all according to PNY a licensed maker of GPU’s for Nvidia. The Quadro 2000 and 4000 SLI modes cover things like “multiple OS support”- useful only if you run, say, Linux and Windows on the same machine. I stuck with my GTX 470 and I’m satisfied. I got my whole rig new for $650 with an i7 2.9 chip, 8Gb (now 16Gb), Win64p, new case and 4 fans (loud but effective).
Thanks for your post. So does Bunkspeed Shot take advantage of SLI? If so, what aspect? Currently with iray SLI’ing GPUs should be ignored because there’s no benefit and it may actually slow down the rendering process.
Hii Jeff!
“You’ll have to use the 580 series until nvidia/autodesk supplies a recompiled version of iray that’s compatible with the 680/Kepler based GPUs.”
What about Vray Rt? Does gtx 680 work with Vray gpu?
Thanks!
I believe so (using OpenCL), but the results aren’t what many were expecting. I say that based off Vlado’s comment HERE: “Some people did preliminary tests with V-Ray RT GPU and the 680 seems to fall between 570 and 580 in performance, closer to 570 – similar to the AnandTech benchmarks. Also, most CUDA stuff doesn’t seem to work on the 680 yet and apparently CUDA programs need to be updated to work. So unfortunately almost no other way to compare for the moment except for OpenCL.
Best regards,
Vlado “
Thanks a lot !
Hi Jeff,
First let me tell you that this article was SUPER useful, really appreciate that you keep answering our questions too.
I am in the dilemma of building myself a new PC and comparing with the budget I have with the fact that it will be used for hardcore gaming too, I’ve kind of decided to go for GTX rather than Quadro. (Unless you suggest me otherwise).
Until now I’ve been working on a mobile workstation (Dell precision M6300 w/ Quadro FX1600M) Never had any problem in Maya so far and I’m guessing my renders were probably awfully slow, but I’ve always lived with it, so cant really tell. (working with Mental ray). Otherwise couldn’t go more than 8 billions polygons in MudBox, etc..
Now I am hesitating between going for 2x GTX 580 3/4gb or a single GTX 690.. I know there is compatibility issues with the Kepler card still and so it’s hard to tell, but what would you recommend ? Is the VRAM more important than CUDA core ? Will using 580 in SLI totally be better than a single Kepler ? Or should I really go with Quadro cards.. (But I fear for gaming as the Quadro 1600 was really poor..) Many thanks in advance !
Balancing a computer between GPU rendering and gaming is a tough call. I’d probably look into the 3gb 580gtx series and just make sure you keep it running as cool as you can when rendering with a GPU rendering application. THIS post may be helpful too.
Can you give us the other link for post?
When I click “This” I got massage “You do not have permission to preview drafts.”
Fixed link.
Hi Jeff,
I’ve got to admit, that your blog is really cool ^^
Could You tell me, is it possible to mount in one machine Quadro 600 (with two displays connected), and GTX 560ti/580 as a second card used for cuda rendering?
Motherboard will by based on AMD 99FX chipset.
Thanks in advance,
Greetings from Poland ;]
Hi Jeff, I’ve been doing a lot of research into buying a 3d workstation some time before my part 2 degree in architecture this year.
I Came across this from Andrew Lynn on CGArchitect.com:
http://www.amazon.com/gp/richpub/listmania/byauthor/A238V1XTSK9NFE?ie=UTF8&tag=andylynnnet-20&linkCode=ur2&camp=1789&creative=390957
In the high end build he suggests to use a a firepro V7900 to power your monitor, and the two gtx 580′s for your cuda rendering.
“Use this card to power your monitor. Because it’s ATI, and CUDA doesn’t run on ATI cards, the CUDA system won’t get confused and try to put display and CUDA on the same GPU.”
I wonder if this is a solution for the people complaining that they need the good viewport performance from the pro cards, and the cheap/fast gpu rendering from the gtx cards?
If thats the case its probably the route I will take, but really not sure at the moment.
Afraid I haven’t tried that combination so I can’t offer any real world data on how well it may or may not work. If you haven’t done so already I would ask around on the larger public forums to try and find someone with that setup AND that works with the same types of scenes as you.
Thanks for the quick response. I’ll give that a try. I’ve got until September, so dont want to commit to early anyway.
Hello!
Great post
I see you work with mental ray and iray.
Why not vray and vray rt?
As I know vary is more realistic
Is there some advantage to use mental/iray ?
Thanks in advance
I see you work with mental ray and iray. Why not vray and vray rt? – I haven’t used mental ray in a few years. I switched to GPU rendering because it’s optimal for what I do.
“Why not vray and vray rt? Is there some advantage to use mental/iray ?” – I’m not using VRay r/t at the moment because I encounter a lot of crashes with VRay r/t on my work scenes. Plus it lacks motion blur at the moment which I use frequently with iray.
Hi Jeff,
I’m sure you’re aware, but the latest v-ray service pack now has motion blur, WrapperMtl reflection and refraction and there’s also a CUDA engine option – which might resolve some of the issues with render errors etc.
Cheers,
Steve
Been downloading it for the last two hours (10KBs..ouch). Will keep my fingers crossed that it works smoothly now with my workstation.
Hi Jeff,
I’m on a fence about a new system that I need to get for product rendering work. It’s pretty much all stills for advertising and web. Car CAD models I get are usually 5 to 6 million polygons, that includes all trim on layers.
Right now I’m using Max with Vray, and Vray RT. I haven’t tried iRAY since at the moment my system does not have CUDA cards (it’s a MAC Tower running WIN7 64bit).
I’m curious about how in real production scenario two Tesla 2075 cards stack up versus i7 3930K or E5 Dual Xeons.
For CPU comparison I would go with Cinebench 11.5 scores.
Two configurations that I’m looking at are:
i7 3930K 6C (6 Core) that gets 10.15 CPU score or so, nicely over clocked will go up 13.3.
2 x Xeon 8C E5-2660 – 2.2Ghz that gets 20.27 score which is quite fast for Vray RT.
The dual Xeon would cost around $2200 more.
What I’m extremely curious about is how to measure speed of Tesla C2075 versus the CPU. I have found some information but it is so all over the place, and with such a huge discrepancies it is really impossible to determine what is true and what is not.
So my million dollars question is how fast is Tesla 2075 in relation to CPU’s Cinebench scores. And what the Real Time speed advantage would be over E5-2660 – 2.2Ghz, if there is any.
I hope you’d be able to answer.
Thanks a lot Jeff,
Charles
“So my million dollars question is how fast is Tesla 2075 in relation to CPU’s Cinebench scores. And what the Real Time speed advantage would be over E5-2660 – 2.2Ghz, if there is any.” – I don’t know how I can translate GPU performance into CPU Cinebench scores. I have spoken with people that have tested dual xeon’s against tesla GPUs and the tesla’s were faster. However, that was last year. Here’s one example from “cf-ntu” post HERE.
Quote:
“cf-ntu
2x 6core Xeon X5660 @ 2.8Gig (with 2x Tesla C2050 fitted) = 19seconds
(17 seconds rendering time, 2 seconds conversion time)
2x 6core Xeon X5660 @ 2.8Gig (No CUDA enabled) = 60 seconds
I wonder if its fair to say that one Tesla C2050 is worth around 2x 6core Hyper Threaded Xeons (24 cores) in terms of iRay rendering power?
Of course, they also offers a great way to pack a load of power into one box
”
I don’t know how that holds up to today’s fastest Xeon’s though.
I feel that I need to say this too though. You mentioned your scenes are typically in the 5-6 million polygon range. A 7 million polygon scene will weigh in around 1gb (depending on number of textures, output size, whether or not the GPU also drives a monitor, etc.) so IMHO the 6GB tesla is overkill for your scenes. You can save some cash and get a 2gb or 3GB GPU that should handle those scenes nicely.
Unfortunately the CAD vehicles I work with are now in the 16-22 million polygon range and with textures and 7k to 11k output resolution I’m using anywhere from 4gb to 5gb of GPU memory so that’s why I have to use the 6GB models.
Hope that helps some?
Thanks a lot for your response Jeff. Based on that I think I can do some math and figure out the actual performance of Tesla. I should post it back here later once I do it.
That’s pretty crazy that the files you get are that big. I specifically tell my vendors when they do conversion from CAD to Maya/Max to keep them within 5-6 million polys as there is really no reason for them to be bigger even for high res print output. The print res shots I do end up being around 8K to 13K tops, but that’s a size of an actual composed image from my photoshoots, the car within the frame could be 3K or 4K render as I tend to leave nice amount of room around the car, usually don’t do tight crops, especially for web where the images will be used in many places at difference aspect ratios. And it the past I did 4K car output with scanned data converted into Maya and Max that was only 2 million polys and that held up really well for advertising shots.
Good to know that 6mill poly model will fit within the 3GB or so. I was going to test the GPU with GTX 670 4GB or GTX 680 4GB version and then see how to proceed from there. And if one day I need Tesla I can always get that.
I also haven’t been experiencing any crashes with Max and Vray, very rarely. But I can really push it hard for days without crashing so I wonder if there is something in your hardware configuration that causes that.
Thanks a lot for your response!
Charles