Provided it’s true, I’m really digging the specifications on the upcoming (2013?) nvidia GPUs.
- 2880 CUDA Cores
- 15 SMX Clusters
- 384-bit Memory Controller
- Up to 24GB of GDDR5 memory
- 2nd Gen ECC
- Hardware GPU Silicon Virtualization
- Hyper-Q (Slashes CPU idle time by allowing multiple CPU cores to simultaneously utilize a single Kepler GPU, dramatically advancing programmability and efficiency)
- Dynamic Parallelism (Simplifies GPU programming by allowing programmers to easily accelerate all parallel nested loops – resulting in a GPU dynamically spawning new threads on its own without going back to the CPU)
- 50-85% Double Precision Rate to Single Precision
- At least 1.5 TFLOPS DP FP64
- Target: 250 GB/s bandwidth
I get pretty good render times on the type of scenes I work on with the 1344 CUDA cores I have access to now (3 GPU’s x 448 cores each). I can only imagine what it would be like to render with 2880 cores x 3 GPUs = 8640 CUDA cores! That may be getting close to real time. Not to mention the memory footprint increase for more complex scenes.
Fingers crossed that it won’t require a second mortgage on a home to purchase!
25 comments
Trackback e pingback
-
New PC Required As Per This Usage - Page 3
[...] http://www.nyc3d.org/3dwp/PortfolioI...s-quadro-5000/ Maxforums - What's up with Quadro? - page 1 Tesla K20, 2013 GeForce and Quadro GPUs ...
If software develers will optimise their code based on cuda5 in order to get performance boost from Kepler cards, even the 690 might be a good deal as Octane is gong to offer instances in the end of this summer and by tlhe roadmap (if i remember right) iRay is goingto do this too..SO..2gb gtx 690 might bea good deal for the money if you need speed and 680 if You need 4gb of ram. As for teslas,..because of high server space asdoption i believe the top models are going to be highly priced and some of those cards will not be optimise to run windows for soke time too.. As you look to the official website presentation only g10 ( & g20) are not for desktop users =) windows driverns are missing =p only one card 2090Q is for workstation =) the other are smoly not for avarage consumer =p nleas in your house you have your own developers that could write You a bit of code to use them =p
Hi Jeff,
We are confident that Kepler cuda 1344 are comparable to cuda 1344 Fermi?
Any news on the timing of Iray updated for compatibility with Kepler?
As far as Kepler support goes, nvidia/mental images states that is should be beore the end of this year…
The pro cards are way out of my reach at the moment. I’m putting together a water-cooled rig with 4x GTX 580s with 3 gig each = 2048 cuda cores. That should see me through to when I can afford some pro hardware or Octane and iRay can properly exploit the power of the 680 Kepler architecture. Thanks to water-cooling, decent performance is within my grasp. In fact, for the price of the pro cards I could probably put together two rigs!
Hello Jeff,
nice Blog and Website, well done, go on with this please. Also the small tutorials, super easy explained, also a fine work.
The nVIDIA Quadro cards based on the upcoming Kepler GPU´s I wish prefer to see for the Autodesk 3ds Max,
Maya, Softimage, Rhino, C4D in combination with the Vray, Mental Ray + Iray, Arion or other Software and render engines that will works fine with the nVIDIA Maximus setup. But the GTX 590 and then the GTX 690 are a very good choice for the AutoCAD, Inventor and Revit group. To give them something what is powerful and with the good vapor chamber coolers on the brand new GTX 690 also a very good cooling choice for this action.
But this will be also directed and owed to the bad situation they are in now and at second that I also don´t want to miss my Maximus nVIDIA card setup. I´m working with two nVIDIA 6000 SLI and 2 Tesla C2075 cards and since
this times I don´t want to seddle back to 10 hours, 5 hours or 3 hours render time. The post production work is mostly done in Adobe Photoshop and After Effects, so let us have a look on their brand new Creation Suite CS6.
Whats new,some new engines and they are also starting to direct some actions to the GPU cores, thanks nVIDIA!
At last but not least, a outdated machine (rig) from 2010 is also pretty good if the right cards are inserted, of course for a lot of money but, this cards will be even enough for me for the next to or perhaps also 3 years and in older times, with a looking eye towards the new suites from Autodesk, Adobe, C4d, and many other programs were you have to settle up all two years, pointed to the cpu and ram set up, it was a better investment I´ve ever done. I consider to jeff that the most things but not for all, the nVIDIA Maximus setup is pretty a fine way and the most interesting render engine that goes a better and fair way is the arion render engine. You cant easily tell the engine, what you want to use CPU + ram, GPU or nodes over the LAN, this is perhaps the best choice for all of us (the way or ability) I mean. It was in older days not so easily to sell a older rig, but this cards you will ever be able to sell away for a good price and with the new E5, the new Kepler based cards, the new Suites from Adobe and Autodesk and also faster and cheaper ram it seems to be the year for the creative peoples.
Best regards from Germany
Carsten Wartung
Hello Jeff,
I came across your website/blog several times over the past two or three years as I tried to find good tutorials and advice about 3ds max. Your work is inspiring, at least for me because I recently started 3d animation and there are many programs I want to try (For example RealFlow) but can’t cause I’m sixteen. Anyway i just wanted to write somewhere that your work is inspiring and I respect it very much.
Also 8640 CUDA Cores?! That’s a LOT. I only have, what 636 cores. Anyway keep up the brilliant work. Constantly coming here for tips on iray rendering.
Cheers!
Ekin
Thanks for your message. Look into some of the educational licenses or demos of the programs you’re interested in. I know money can definitely be an issue for teenagers so if the educational prices are still out of the question then try contacting them directly. Who knows, you may catch the right person and be able to score a educational or Not For Retail version of the software.
Also have you tried Blender? It’s free and has a lot of features now, like GPU rendering and fluid dynamics. I think some people complain about it’s UI being unfriendly, but I also know people that use it for fluids (export them to programs 3ds Max, etc.). Also, Houdini has a free apprentice edition as well as a $99 USD version with no watermarks and larger output.
Looks its true http://www.nvidia.com/content/tesla/pdf/NV_DS_TeslaK_Family_May_2012_LR.pdf
Hopefully we’ll see them released ‘into the wild’ soon for testing.
Time to slam through a ton of client work so I can afford the ridiculous price tag the K20 is sure to launch with…
I mean, come on. 2880 cuda cores and up to 24GB? Shut up and take my money.
New Quadro K5000, based on Kepler with 1536 “new” Cuda cores is nothing to get excited about. I played with it at Siggraph with latest version of V-Ray, using RT in GPU mode, testing it with several scenes, including a car model. First of all, it does not feel any faster then current generation Quadro 6000. It has 4GB RAM, I hoped for at least 6GB RAM and hopefully 8GB or more in K6000 whenever it comes out. The only benefit is that it requires less power over current generation.
In viewport performance it does beat the hell out of GTX 670 and 680 but for GPU rendering it offer no speed advantage over new GTX cards which you can also get with 4GB for a fraction of the price.
If anyone really only cares about GPU rendering then you can get 3 GTX 670 cards for less then one K5000 will cost.
So for now I think it is better to go for a Quadro 6000 which you can find on ebay, just hunt down a decent price. That’s just my personal opinion based on what I have seen in person.
Thanks for the info Charles. I wonder if they are gearing the new Quadro’s more towards the ‘maximus’ protocol where the it’s primarily used for viewport performance while the tesla’s are used for rendering? If so then perhaps we will see the new tesla’s have better rendering performance and memory options than the Quadros.
Hi Jeff,
Yes, you are correct as this seems to be the case. They want users to use Quadro K for display and Tesla’s K20 for GPU processing.
Quadro K5000 is being priced at $2,249 MSRP and Tesla K20 at $3,199 MSRP. There is no word about K6000.
It appears that K5000 has definitely huge advantage in viewport performance over GTX 670/680 cards which is is expected based on Quadro’s history. It will also be much more responsive then current Quadro 5000 but I’m not sure if it will have an advantage over Quadro 6000. Definitely not as far as RAM goes.
But in GPU rendering, K5000 would be about the same as GTX680. GTX 670 is roughly 5% slower, but considering the $460 price for GTX 670 with 4GB, a great performance to price ratio deal. This is strictly based on V-RAY.
As far as Tesla K20, it’a hard to say how much faster it will be then C2075 or C2090. Their marketing talk was that it has 3 x times performance advantage per watt. So we know it will require less power but that does not mean it will be 3 times as fast as C2075. They did not want to talk about it and all I got was very vague.
You can only hope K20 it will be 1.5 – 1.8 times faster then 2075, and if it is twice as fast in GPU rendering that would be a miracle
They are definitely aiming for much lower power consumption, but not really aiming at huge GPU rendering gains over current generation.
Most of the guys in front of systems did not even know what card they are doing demo’s on. I had to ask every single one to call it up within WIN so I can see it for myself.
At one point I messed around with a setup with one K5000 and BOXX GPU render box with 5 Tesla’s 2075. Loaded it up car model that was setup at the demo PC, the model was light, only 1.5 mill polys. I switched the cards on and off to see a performance difference between 5 Tesla’s, 3 Tesla’s and 1 Tesla 2075 and how it compares to the K5000. I run some renders at 1K and 4K and only judged it by visual feedback I was seeing on screen so I don’t have hard numbers. But it did not look to me like the 1500+ Cores K5000 was much faster then single C2075 in pure GPU rendering (Latest V-RAY in CUDA mode, not Open CL), maybe 15% at best, but it’s quite hard to tell by just looking at the screen.
That said, I kind of expected it based on GTX670/680 tests.
Tesla’s K20 will be out sometime in December, at $3199 MSRP, it better be fast like hell and have a lot of RAM
Based on the price increase/difference over Quadro K, I Imagine there will be more CUDA cores and better GPU rendering performance. I don’t think it will be 3 times as fast as C2075 although that would be very nice and amazing. Based on how Kepler Cores scale up it would have to have somewhere around 4600 cores to achieve 3 times speed gain over C2075. But if it comes out with rumored 2880 Cores we may see around 1.8 times speed gain.
We won’t know until early next year once some tests are done.
Thanks for the detailed response. I think you’re spot on with your thoughts here.
Sure, no problem Jeff.
I also got my new PC with 3930K few days ago, it has one GTX 670 4GB card in it. The performance in V-Ray GPU is really good. On average 4 to 6 times faster then RT in CPU mode. I added a second GTX 670 4GB two days later and rendering speed scales up really well, pretty much 100%. I have been testing it with one of the production car scenes which is 7 million polygons car, 800MB 3ds Max file, doing 4K and 6K renders.
Now I’m looking forward for iRay becoming compatible with Kepler 6 Series cards. As a matter of fact I was just thinking, V-Ray was ready within 3 weeks and it’s been now over 5 months and updated Kepler ready iRay is not out yet.
“As a matter of fact I was just thinking, V-Ray was ready within 3 weeks and it’s been now over 5 months and updated Kepler ready iRay is not out yet.” – Indeed, this has been an odd feeling situation for a while now.
Hi Jeff.
Steve from ARC wrote ” There is a good chance that Autodesk will publish iray on Kepler support also for existing versions of 3dsMax.
Best, Steve. ”
I hope AD will deliver the updatefor Kepler this year.
Best regards Seeda.
Hi jeff,
do you know whether iray surpport the quadro k2000m ,i use 3ds max2013,but iray tell me that can’t find the cuda device. my computer is thinkpad w530
I’m not familiar with that card so I can’t say for sure. However, I will assume it’s a new GPU and with the K in front of it sounds like a Kepler based GPU to me which is not supported by iray in 3dsmax 2013 at this time.
Hi Jeff-Charles,
I’ve gone through the benchmarks of the Kepler cards and when it comes to compute performance the GPGPU benchmark – all nVidia cards including the Quadros are basically thrashed by the Radeons. Also note that the Fermis have better numbers than the Keplers.
http://www.tomshardware.com/reviews/geforce-gtx-660-geforce-gtx-650-benchmark,3297-19.html
Compare the AMD FirePro w9000 6Gb with the Quadro 6000 6GB. The Quadro stands nowhere and the HD 7970 is at par with the FirePro w9000. Now, I don’t know how this benchmark translates into realworld scenario but knowing that FirePro’s are recomended for CAD applications, I’m guessing the HD7970 should perform fabulously!
I understand that iray only supports cuda at this moment and that maybe the case with some other engines as well but what about vray? Vray can utilize OpenCL (and that anyway looks like the future – even Adobe since CS6 has changed to OpenCL over Cuda).
Have you tested Vray-RT with a 7xxx Radeon in OpenCL mode? I read somewhere that at this moment only Nvidia’s implementation of OpenCL is supported by Vray RT… is that true? Where are these render engines headed… OpenCL or CUDA?
I’m going to make my purchase decision based on this. Should I go for a Nvidia 6xx or a AMD 7xxx?
“Have you tested Vray-RT with a 7xxx Radeon in OpenCL mode?” – I have not. Since the ati GPUs do not work with my current production rendering application I can’t personally justify the purchase. If someone at ati is reading this and would like to send me a GPU for testing with VRay r/t, I’d be happy to do so.
“I understand that iray only supports cuda at this moment and that maybe the case with some other engines as well but what about vray? Vray can utilize OpenCL (and that anyway looks like the future – even Adobe since CS6 has changed to OpenCL over Cuda).” – I do know that Vlado with ChaosGroup (developers of VRay / VRay r/t, etc..) has tried to get the 7970 to work with VRay r/t. They kept encountering problems with the ati drivers and thus no tests have been successfully made. There’s a GPU benchmark thread on the official VRay/VRay r/t forum HERE that includes test with some other ati hardware (5970, 5870, FirePro, etc.). Also, HERE’s a link to their discussion on the 7970.
“I read somewhere that at this moment only Nvidia’s implementation of OpenCL is supported by Vray RT… is that true?” – They (VRay r/t forum) have some ati hardware results in their benchmark so it apparently works with some of the ati stuff. The results aren’t stellar from what I see listed though.
“Where are these render engines headed… OpenCL or CUDA?” – I don’t see it shifting one way or another at this point. If it ever does it will be farther down the road (IMHO) and probably time to buy more hardware anyway if/when that time comes.
“Should I go for a Nvidia 6xx or a AMD 7xxx?” – Judging from the results listed in the VRay r/t benchmark thread and assuming your primary concern is rendering with VRay r/t then I’d recommend going with nvidia hardware at this point in time. That may change later down the road if/when the issues with the ati drivers and vray r/t can be worked out. However, so far the results from the Kepler based 6xx GPUs have been on par with the 5xx GPUs. I don’t know if that will change with as drivers and software evolve or not. Also keep in mind the upcoming tesla GPUs that I think should hit the market later this year. Perhaps they will offer a larger performance boost. Hopefully we won’t have to wait too long to find out.
Hi Jeff,
I have a question. Is it possible, and if so, would you recommend using a Quadro K5000 for viewport and have 2 or 3 GTX 680 4GB cards for iRay rendering? Would there be a driver confict? I would liquid cool the GTX cards as well.
I’m pricing out my second system and trying to decide what would work best or give the best performance. Price is a factor, but not if it is going to cause problems. Running 3dsmax 2013 with update v6. My other option is to go full Maximus setup. Quadro K5000 + 1xTesla K20. Not sure if I can afford 2x Tesla K20′s at this time.
Thank you,
Jason
I can’t say for certain whether or not there would be driver conflicts as I haven’t personally tried that combo. However, FWIW, I did run a Quadro 5000 and a GTX470 (I think) together a year or two ago when I started experimenting with GPU rendering. I didn’t encounter any driver issues back then.
Maybe someone out there reading this has a similar setup they are using and will chime in with a more solid answer for you.
I really need some advice. I’ve just ramped up, my client / partner is wanting to upgrade me to super fast rendering. We aren’t doing any visualization as of yet, so mostly pretty cars/broadcast / post fx for cinema. I use R13 on an i7 with a GTX 560ti currently, and a render farm when needed. Will a Tesla work for me if I get iRay?
I need some consultation here if you have any time for a phone call in the near future.
You seem to be the most informed person I can find online so far.
Cheers,
Jeff (California)