GTX 690 Tweet

[gabrielefx]

I don't think that Nvidia optimizes the Cuda code for iray only. Cuda is used in many applications.
For sure the realtime raytracing will be used in the new videogames then the new GTXs will need more power to enable well this feature.
The GTX6xx lacks this feature, next GTX7xx will have more computational power (for rt videogames) but not equal to the Teslas.
In the past the M2090 was equal to the GXT580. Who own here an M2090?
Because the gpu based render is an incredible business for Nvidia then we will see implemented more hw/sw features only for the Tesla line.
Do you remember the 386 co-processors? I do, Autocad did run at the speed of light...

[abstrax]

But even the Nvida iray team guys say there will not be same kepler performance like on fermi.
And the new Kepler pro-boards are for double-precision only.

the bad thing is that it seems you guys have to re-code half your renderer if nvidia decides to change something (to their advantage).

I wouldn't say that. The recent work to have good support of Keplers and Fermi in one build was a pain, but it was worth it. It gives us more freedom when it comes to building CUDA code for different architectures and devices plus a few other benefits. We didn't change anything in the actual algorithms. It was more a build issue, but it should be sorted now

Obviously what I said about coherency on Keplers still applies, which is a bit more complicated to explain. Here is my personal view of all this:

If you look at Fermi you can see that it was leaning very much towards GPGPU computing. Actually it was fairly revolutional, but brought NVIDIA lots of delays, manufacturing and heat problems. You can see that with the GTX 480 which had 16 multi-processors, but only 15 enabled. It also earned strong criticism from the gaming community, since the performance didn't really match the price. And the gaming community is much larger than the GPGPU community, thus a lot more important, and it doesn't care how fast Octane renders as long as they can't play <pick your favourite game> smoothly in HD with full details.

So they brought out the second generation of Fermi, which solved most of NVIDIA's technical problems, but having basically the same architecture as the first generation, there wasn't much improvement in the field of gaming. All the while ATI was happily selling lots of Radeons which were a lot less suitable for GPGPU than Fermis, but worked great in games at reasonable prices.

Now, NVIDIA had to do something about that and the answer is Kepler. It's leading in most of the gaming benchmarks. Unfortunately they achieved this by chopping the number of multiprocessors in half and increasing the number of cores per multiprocessor by a factor of 6, i.e. having 3x as many cores in total. This is great for games that use rasterizing (i.e. all games), which is probably the most coherent algorithm you can think of, but it's not so great for GPGPU computing applications which often run fairly incoherent stuff.

Where does that leave us? We will see. The fact that the vanilla Octane render algorithms run on a 680 roughly as fast as one 590 GPU and with a much lower energy consumption gives me hope. I think, we also have to accept that the area of GPGPU computing is under heavy construction on the software as well as on the hardware side. This means there will be alot of progress in the coming years, but it also means that we very likely will have to adapt part of our rendering code to new architectures and new concepts. That can be tedious, but also fun

Cheers,
Marcus

[t_3]

thanks marcus, interesting read!

shouldn't the upcoming gk110 - with it's slightly different architecture - be nvidias answer for gpgpu? apparently the gk110 will again feature 15 smx units (at max.) while the current gk104 only provides a 8 or less of them. this should result in notable better performance. the question is, if they market this new chip only for computing or in cheaper gaming cards also - i hope for the latter

[abstrax]

thanks marcus, interesting read!

shouldn't the upcoming gk110 - with it's slightly different architecture - be nvidias answer for gpgpu? apparently the gk110 will again feature 15 smx units (at max.) while the current gk104 only provides a 8 or less of them. this should result in notable better performance. the question is, if they market this new chip only for computing or in cheaper gaming cards also - i hope for the latter

Yes, it's definitely aimed at GPGPU, but for now it looks like it's only aimed at the high-end, which makes it uninteresting to Octane users as it's waaay too expensive. Fermi was the try to cover the full range of applications from gaming to GPGPU with one GPU series (even the Quadros and Teslas had pretty much the same GPUs as the GeForce cards). And I would say it didn't work too awesome. So what NVIDIA seems to be trying now is that they are basically offering two different GPU series, which are based on the same building blocks, but are structured differently, for different markets. -> I don't expect a gaming card based on the GK110 soon or perhaps at all.

But, I wouldn't worry too much though. The GPU development in the last years has been fairly awesome and I'm pretty sure that we will see some more awesomeness in the future and the Keplers are not that bad.

Cheers,
Marcus

[gueoct]

so iRay and Arion are performing better with kepler?

What i heard, iray has kepler issues....

[justix]

But even the Nvida iray team guys say there will not be same kepler performance like on fermi.
And the new Kepler pro-boards are for double-precision only.

the bad thing is that it seems you guys have to re-code half your renderer if nvidia decides to change something (to their advantage).

I wouldn't say that. The recent work to have good support of Keplers and Fermi in one build was a pain, but it was worth it. It gives us more freedom when it comes to building CUDA code for different architectures and devices plus a few other benefits. We didn't change anything in the actual algorithms. It was more a build issue, but it should be sorted now

Obviously what I said about coherency on Keplers still applies, which is a bit more complicated to explain. Here is my personal view of all this:

If you look at Fermi you can see that it was leaning very much towards GPGPU computing. Actually it was fairly revolutional, but brought NVIDIA lots of delays, manufacturing and heat problems. You can see that with the GTX 480 which had 16 multi-processors, but only 15 enabled. It also earned strong criticism from the gaming community, since the performance didn't really match the price. And the gaming community is much larger than the GPGPU community, thus a lot more important, and it doesn't care how fast Octane renders as long as they can't play <pick your favourite game> smoothly in HD with full details.

So they brought out the second generation of Fermi, which solved most of NVIDIA's technical problems, but having basically the same architecture as the first generation, there wasn't much improvement in the field of gaming. All the while ATI was happily selling lots of Radeons which were a lot less suitable for GPGPU than Fermis, but worked great in games at reasonable prices.

Now, NVIDIA had to do something about that and the answer is Kepler. It's leading in most of the gaming benchmarks. Unfortunately they achieved this by chopping the number of multiprocessors in half and increasing the number of cores per multiprocessor by a factor of 6, i.e. having 3x as many cores in total. This is great for games that use rasterizing (i.e. all games), which is probably the most coherent algorithm you can think of, but it's not so great for GPGPU computing applications which often run fairly incoherent stuff.

Where does that leave us? We will see. The fact that the vanilla Octane render algorithms run on a 680 roughly as fast as one 590 GPU and with a much lower energy consumption gives me hope. I think, we also have to accept that the area of GPGPU computing is under heavy construction on the software as well as on the hardware side. This means there will be alot of progress in the coming years, but it also means that we very likely will have to adapt part of our rendering code to new architectures and new concepts. That can be tedious, but also fun

Cheers,
Marcus

which leaves me to the conclusion that right now a 670 is out of the question if I want to speed up my renderings from my Gtx470..or?

[glimpse]

SomeWhere on the web, I've read a qoute from one of nVidia heads, that they do look forward to sell more Teslas then ever before combined together.

For me it would be no surprise if they chose to segment their product line mor and cut computational power for gaming cards.

On the other end I do not see the problem to buy 4-5 time pricier card if it has way more of vRam, renders faster, draws less energy (and serves for longer period at the sametime) though I would not pay that if cheaper gamer card could do that - I think it's obvious and nVidia understands that.

With new architecture nVidia got to the top again..doubt well see gk110 based cards soon as there is little need of them from marketing perspective - who wants to cut their profits (6xx line do sell good enough as gamers can expect the same performance from 680 as it was only available only for 590 owners, but for way lower price, pluss availability is better too)

Do You guys plan to test some new tesla products?

P.S. Does the ECC help for stability at programs like Octane? I think Teslas have that, but is it useful in our case?..

[GeorgeR]

I really don't think that market segmentation is going to fly. For a while now, games have used Compute Shaders and CUDA acceleration for the heavier things (FFT, etc.), so the gaming community will demand their cards perform in this area.