I am using a Tesla M60 card as pass-through by using the m60-8q profile, and I understand that this should allow me to use OpenCL on a Windows 10 VM running on the VSphere 6.7 host, but when I run GPU-Z it shows no OpenCL or CUDA support.
Driver version is as below and the Windows 10 client VM is running the 441.66 that came with the GRID software ZIP:
[root@stc-vdi-01:~] nvidia-smi
Mon Jan 11 17:04:26 2021
±----------------------------------------------------------------------------+
| NVIDIA-SMI 440.43 Driver Version: 440.43 CUDA Version: N/A |
|-------------------------------±---------------------±---------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 Tesla M60 On | 00000000:05:00.0 Off | Off |
| N/A 33C P8 27W / 150W | 8143MiB / 8191MiB | 0% Default |
±------------------------------±---------------------±---------------------+
| 1 Tesla M60 On | 00000000:06:00.0 Off | Off |
| N/A 30C P8 25W / 150W | 8143MiB / 8191MiB | 0% Default |
±------------------------------±---------------------±---------------------+
I can see the M60 GPU in task manager and it is working for graphical operations, just no OpenCL support.
The server is a Dell PowerEdge R730 and I have configured the Graphics Cards in VSphere like this:
CUDA doesn’t get installed inside the Hypervisor, which is why the vGPU Manager doesn’t list it as per your image above. CUDA gets installed with the vGPU Driver inside the VM.
For reference, if you were running Pascal or newer architecture, you could use any Q Profile and be able to run CUDA workloads. It’s just Maxwell that requires the full 8Q Profile be used for CUDA due to the age of the architecture. But it’s a good place to start with, as it’s one of the few GPUs that allows you to run mixed vGPU Profiles concurrently on the same board.
Within the VM, Compute Mode also shows as “default” which I think means that CUDA/OpenCL should be available in the VM, but GPU-Z (and also my compute app) both say that OpenCL is not available in the VM:
You will note that I have used the m60_8q profile, which should allocate an entire GPU to the VM, which should allow OpenCL/CUDA (as it is not a shared GPU without pre-emption)
I am only performance testing to evaluate various hardware setups at this time (I may not virtualise and just run on baremetal) and have not setup a flexlm license server yet. I was under the impression that it should work (albeit with a pop-up warning) for testing. Certainly, the vGPU for graphics is working OK, is the OpenCL/CUDA computation mode different - does it definitely need the vGPU license before it will enable?
Yes, please make sure it’s licensed. Functionality will reduce very quickly if the VM is not licensed.
You can register for a vGPU evaluation here GRID Evaluation | NVIDIA However as you already have the vGPU software, I’m assuming you already have access to the licenses and an evaluation. Setting up a license server is very quick, so you should be up and running in no time!
