Hi all
Is there a chance that one day we have an Vray GPU-based engine who can exploit the high level of computation of solutions like Tesla c1060?
thkx
From our latest tests, the sample code that we had ran about 4 times faster on a dedicated GPGPU than the equivalent CPU code on a single core. This still does not justify the effort… we would need it to be 10 or more times faster to make it worthwhile.
Further on, some algorithms (raytracing included) simply cannot exploit that particular type of computational parallelism. This is because they tend to be task-parallel rather than data parallel.
Best regards,
Vlado
Oh!!, only 4 times 
how can (RTSQUARE & Gelato) claim up to 10 to 50 times faster than MR/vray/… is it because of CUDA?
you say “4 times faster on a dedicated GPGPU than the equivalent CPU code on a single core” but I pointed out that 1 TESLA GPU = 240 cores then can we say 4 times x 240 cores ?!!
Best regards
Naroura
Vlado - what about Larrabee? Any tests yet? I read that it’s quite powerfull and easy to program.
What about throwing stuff like texture filtering to the video card using the anisotropic filtering present in most cards? I have no idea whats involved in passing things between the CPU and GPU but that seems like a particular thing that can be slow to calculate.
Don’t know about RTSQUARE, but Gelato?? As far as I know, even Gelato did its raytracing on the CPU, not the GPU (and it was done by nVidia, after all), and V-Ray was quite a bit faster in that department.
Best regards,
Vlado
Anyways, rest assured that we are looking into our options in that department 
Best regards,
Vlado
take a look :
RTsquare--> http://www.gputech.com/rtsquare.html
Gelato --> http://www.nvidia.com/object/io\_1186464580202.html
Best regards,
naroura
Yes, I know the Gelato page; however it is about relighting, not about re-rendering, let alone raytracing. Last time I checked RTSquare on my machine at work, it was quite a bit slower than V-Ray when doing GI and lots of raytracing.
Best regards,
Vlado
it’s depend on the hardware (Multi-GPU => faster )
anyway, the biggest advantage with Tesla is that you can bring the performance of a small cluster to a single workstation.
and my questions for you Vlado:
1- is it possible for Vray’s users to benefit from this technologie ?
2- Is it possible to build a kind of pilot (or any other soft) which can catch the instructions of the execution of Vray sent to the CPU, adapt them for GPU, and send them towards GPU instead of the CPU, or dispatch them betwin CPU & GPU?
thank’s
Best regards,
NAROURA
PS: sorry It’s a bit difficult for me to express me in English
Yes, and multi-core CPU => faster as well
I can’t yet comment on whether V-Ray will benefit from Tesla specifically; there are other GPGPU solutions as well and some are better suited for the task.
Best regards,
Vlado
You can probably do that, but there is no point in it. To get the full power of the GPU, the algorithms need to be redesigned from the ground up. It is not as simple as a “recompile” for the GPU, whatever some companies would have you believe.
Best regards,
Vlado
I’ve heard of companies using a bunch of Playstation3 as a renderfarm. (not with VRay obviously) - I think it was mentalray but can’t remember now. There was some demo vids on the subject but also can’t recall the location now. Just do a search in Youtube
Yep I know; don’t think it was mental ray though… the reason being it needs to be specially compiled and reworked for this, and obviously only mental images can do that. There was an announcement about mental images reworking mental ray to run on Cell processors, but that was before they were bought by nVidia. After the acquisition, don’t know if that’s still the plan.
Best regards,
Vlado
so vray will never run on ps3 ? :roll:
or is it realy not possible because of the leak of the amount of memory etc…
Even if they got it to run on the PS3 after months (possibly years) of development, it would only be useful for really small/test scenes. Consider that the average production VRay scene takes 1.5gb+ of memory. Alot of the tech demos released are these small scenes or a car with an HDR environment, they’re not actually practical when it comes to serious work, they’re just “cool” curiosities.