I have installed on my ESXi (VMware Horizon View 7) NVIDIA Tesla M60 graphic card and so fair it works all best. But now I will use the “CUDA” with the vGPU (grid_m60-8q) and if I try to install CUDA Toolkit “cuda_8.0.44_win10”, I receive the message:
“This graphic driver could not find compatible graphics hardware. You may continue installation, but you may not be able to run CUDA application with this driver. This may occur with graphic hardware that is never than this toolkit. In hats case, is suggested that you keep your existing driver and install remaining portions of the CUDA Toolkit.”
I use NVIDIA Tesla Driver:
ESXi: NVIDIA-vGPU-VMware_ESXi_6.0_Host_Driver 367.64-1OEM.600.0.0.2494585
Windows 10: 369.17_grid_win10_server2016_64bit_international.exe
Can someone help me here to solve the problem?
In general, CUDA is not supported with GRID vGPU usage at this time.
The exception would be the case where you have a single GRID vGPU running on a physical GPU (i.e. have chosen a vGPU profile which uses an entire physical GPU). In that case, the GPU is effectively in pass-through mode, and CUDA should be usable.
Oh, but this is not good, or rather a fraud from Nvidia side. We bought the Tesla cards extra for this CUDA in vGPU, because Nvidia has so published and now that and still wanted to use several VM’s with CUDA vGPU, now that is not so or very limited.
How can I understand this here: In general, CUDA is not supported with GRID vGPU usage at this time?
Will it be coming soon or there is a way that I can use CUDA under vGPU? I mean, even if it is not official, I would appreciate your support because I have some customers who are waiting extra for vGPU CUDA and so I cannot offer enough resources.
Use the method I describe in my exception, ie. that is covered in the published link. 1:1 mapping between VM and physical GPU.
Yes, I’ve installed today and so it works. Since I have two Tesla cards, I can connect no more than 4 VM’s :-(.
Thanks for your help.