FMX follow up - V-Ray RT / GPU version

[GLASS-CANVAS]

V-Ray RT 2.0 with GPU support is expected to be released later this year, so expect more news soon.

This can only mean that you are rolling VRay and VRay RT into one package? Or will you still sell 2 somehow different versions of VRay? I'm hoping for just 1 version!

[sandykoufax]

This can only mean that you are rolling VRay and VRay RT into one package? Or will you still sell 2 somehow different versions of VRay? I'm hoping for just 1 version!

If you see the video in the first post, you can get the answer. (around 50sec)

[duke2]

If you see the video in the first post, you can get the answer. (around 50sec)

Boy would I be pissed if I bought VRay RT. If they are to be different products - who would ever buy VRay RT (non GPU) when the featureset is the same?

[jacksc02]

Vlado, the links to the videos seem to be dead.

[vlado]

Vlado, the links to the videos seem to be dead.

Seems to work for me (tried just now).

Best regards,
Vlado

[digitalx3d]

Vlado,
Will the GPU version have options for rendering displacement objects and Vray proxies? (a checkbox for these items or something like that) I think what many of us are hoping to see with the GPU renderers is a permanant replacement for CPU renderers so we can use the GPUs to render out high-res stills or animation frames. And, obviously, noise free. I'd be curious to compare render times between GPU and CPU for a final noise-free image.

[vlado]

Vlado,
Will the GPU version have options for rendering displacement objects and Vray proxies? (a checkbox for these items or something like that)

Nope, at least not initially.

I think what many of us are hoping to see with the GPU renderers is a permanant replacement for CPU renderers so we can use the GPUs to render out high-res stills or animation frames.

GPU renderers will not be there at least for a while more. Of course you can use them for final frames in many cases, but there will also be many other cases where they don't have the needed capabilities.

And, obviously, noise free. I'd be curious to compare render times between GPU and CPU for a final noise-free image.

On the CPU you have things like the light cache and the irradiance map; they give the CPU version quite an advantage, since these cannot be implemented easily on a GPU.

Best regards,
Vlado

[baumgore]

Hi,
first, impressive, very impressive videos, the speed increase is tremendous. I once called VrayRT a historic step in CG and hoped for GPU acceleration, but this is simply amazing. I am happy to be in the CGI business with such fundamental changes rising up.
Do you have some benchmarks with TESLA hardware? Are the new 5xxx ATI cards fully Open CL compatible?
Just saw the Tesla M2070 with 6 GB RAM and I am asking myself how must faster it is compared to a standard i7 6 core CPU ? Are we expect to get the same performance as a single GTX 480? Only with the very important plus of 6 GB VRAM? Does someone have seen price tags for the M2070?

[vlado]

Tesla M2070 has 448 CUDA cores compared to 480 in the GTX 480; additionally the clock speeds are slightly slower. So it can be expected that the card as a whole will be a bit slower than the GeForce. But of course you have more RAM.

Best regards,
Vlado

[GLASS-CANVAS]

But of course you have more RAM.

Yes, I was wondering about this - it stikes me that the onboard RAM will be the single limiting factor of what you can and can't render. At the moment the only Fermi Tesla card that's available (I think) is the 3GB 2050. So what happens when you hit the 3GB limit? Does VRay just crash? Or are you looking into implimenting a paging system? 3GB to contain all the polygons and textures of one of my typical scenes is going to be used up in no time at all.

[Dariusz Makowski (Dadal)]

curiously cant Chaosgroup release some kind sdk or wot ever is called to general public so we can compile our own plugins for gpu/rt version? It would save u guys lots of time I guess and allow u lead on marked over other gpu renders that dont support plugins.

[vlado]

curiously cant Chaosgroup release some kind sdk or wot ever is called to general public so we can compile our own plugins for gpu/rt version? It would save u guys lots of time I guess and allow u lead on marked over other gpu renders that dont support plugins.

Hm, I'm not sure how much time it will save. Coding for the regular V-Ray is quite difficult already; coding for the GPU - even more so. Further on, while you can write custom shaders for the V-Ray RT/GPU, this is not the case for lights and cameras, which for performance reasons are implemented in the core GPU code.

Best regards,
Vlado

[GatesB]

deleted post.

[duke2]

In terms of GPU vs CPU, I bought Octane (which is purely CUDA) because it was so damned cheap, and damn is it fast on my lowly 8800GT. I can spit out a 4000x1000 clay renderer for early design review in, quite seriously, ~5 seconds.

Definitely looking forward to VRay's implementation