Information on V-Ray RT GPU beta program

[peter]

Hello everyone,

We are pleased to announce the start of the V-RayRT GPU beta testing program.

In general all the information is available here.


Feel free to post any questions here and we will be glad to answer those.

Stay tuned for more announcements today on the upcoming V-Ray 2.0

Cheers

[Donald2B]

Just want to publicly thank Peter, Vlado and all of the great people @ Chaos for putting together a great set of tools for us to use on a daily basis! Thanks all! I look forward to many more years of great features/updates!

[Dariusz Makowski (Dadal)]

Hi

GPU pefrofmance wot is it ? Is it based on Cuda core? Core mghz? Memory Mghz? What we need to look at when we buy CGI cards?

[vlado]

There are several things to look for:
1) Fermi architecture - it's strongly recommended to get a card based on the Fermi core, as it is much faster than the previous generation
2) Number of CUDA cores - more cores give you faster render speeds
3) Video RAM - more memory allows you to render more complex scenes and in general will work more reliably
4) Core and memory Mhz - obviously, the more the better; not as important as number of cores though.

Best regards,
Vlado

[devin]

Vlado which cards are you using in house for testing, wouldn't those be the best ones to buy?

[vlado]

We are using various cards here, but most of the development happens on GeForce 480 and 295.

Best regards,
Vlado

[Dariusz Makowski (Dadal)]

Hi

Quick question about licenses... Is there any confirmation email that should come to my mailbox at some point saying that my request has been proceed bla bla bla? I have bad feeling I'm w8ing for nothing whhh..

[vlado]

Yes, you will get a .rtu file to reprogram the dongle with the additional license. However it's almost 11pm here, my guess is it will come tomorrow when the support guys get to the office. If you still don't get a reply, let me know and we'll sort it out.

Best regards,
Vlado

[robman321]

Vlado,

I'm I running a Dell Precision m6500 with i7 CPU x920 @ 2.00GHz 2.00 GHz 12.0 GB RAM and a Quadro FX 2800M # NVIDIA Quadro GPUs 1, CUDA Cores 96, GPU Memory Specs:Memory Size Total 1 GB GDDR3 Memory Interface 256-bit, Memory Bandwidth (GB/sec) 64

As it goes, I can not tell the difference in speed between the CPU & Single Kernel Open CL (GPU), as well my Kpaths arent any higher per this scene at 110 Kpath/s... Is the graphics card not good enought?

thanks, rob

[vlado]

First if all check if the RT render server is really running on the GPU. It if cannot initialize the OpenCL stuff (f.e. because of old drivers), it will fall silently back to the CPU code.

Best regards,
Vlado

[Stebbing3D]

How much faster can we expect the multi kernel version of RT to be (on a GTX 480)?

Is the single kernel somehow limited in terms of using all CUDA cores?

[vlado]

The single kernel is not limited in any way as far as performance is concerned. The difference is in the approach - the single kernel version is the entire renderer stuffed into one huge OpenCL program. It does not support procedural textures or 3rd party code, but allows the GPU to render at full speed.

The multi-kernel version allows you to program shaders for the GPU, and wire shaders in complex networks, but the downside is that at present it is somewhat slower and slightly inconvenient to work - everytime you change a material, it must be recompiled for the GPU. Also, the latest nVidia drivers crash with it. For these reasons, it will probably not be included initially.

Best regards,
Vlado

[Stebbing3D]

Thanks for explaining that..