Linux vGPU and grid 2.0 questions

We have an OpenGL application which uses NVIFR for h264 encoding. it saves result in a file.
We are using xenserver + windows vms with attached vGPU now. But we have similar app for linux. It is interesting what state of linux support grid 2.0 platform have.
We want to use similar solution: linux vm + attached vGPU with new card tesla m60. I’ve got some questions concerning this:
is NVIFR supported for Linux vGPU?
is hardware encoder available for vGPU, will it work with 4k videos using vGPU tesla m60 card?

Yes. NVIFR and NVFBC are both supported in vGPU.

Yes it is available and supported for vGPU.

There are some license requirements for the use of 4K displays (a feature of Workstation Extended), however this is separate to NVENC.

Hello Jason_Southern,

Thank you for your answer, it is really great news!

But I had conversation with our developers. They have doubts about Linux NVENC support:
We are using CUDA to access NVENC encoder for Linux. But CUDA is not supported for Linux vGPU… (I see it supported for 1 vGPU = 1 GPU mode only, so it is similar to gpu passthrough.) But I am not sure about that.

As result I have bunch of questions:

  1. Is CUDA enabled for vGPU linux vms? Which vGPU types have CUDA support?

  2. Is it possible/How to access hardware h264 encoder on linux vGPU vm without CUDA?

  3. May be there are way to provide OpenGL buffers to h264 encoder without CUDA?

Thank you in advance.

CUDA is currently only available in Passthrough, M60-8Q and M6-8Q configurations. It will require the next generation GPU architecture for CUDA in other vGPU profiles.

NVENC is supported in all vGPU instances.

For queries on the NVENC SDK they need to go here

Other ISV’s are using NVENC in vGPU sessions, so yes it is possible. You may want to refer to the GRID SDK as well for possible examples.

Hello Jason_Southern

I see someone else already posted topic on devtalk forum, but nobody answer. May be you may recommend person for those questions?

It seems to me GRID SDK have dependency on NVENC issue too. (it is not possible to use without CUDA)

Is it possible to forward this question directly to NVENC developers please?