Artificial Intelligence Computing Leadership from NVIDIA
vGPU Software for direct GPU Passthrough
Hi everyone! Surfing the web and the official vGPU software documentation I became a bit confused. I would like to clarify, do I really need vGPU software and vDWS or vCS license to just Passthrough the Tesla P100 card to one VM for computational purposes? Is it depend on the hypervisor vendor? Thanks in advance
Hi everyone!
Surfing the web and the official vGPU software documentation I became a bit confused. I would like to clarify, do I really need vGPU software and vDWS or vCS license to just Passthrough the Tesla P100 card to one VM for computational purposes? Is it depend on the hypervisor vendor?

Thanks in advance

#1
Posted 01/22/2020 08:31 AM   
Hi If you want to use vGPU (licensed under these versions: vApps, vPC, QvDWS, vComputeServer), then yes, you need a license and you need a specific version of the software which is only available from the Licensing Portal. If you want to use Passthrough, and only intend to use it for Computational workloads (without graphical acceleration, because there won't be any as it's the wrong type of driver) then you don't need a license, and you can download this driver from the normal NVIDIA Drivers site. Regards MG
Hi

If you want to use vGPU (licensed under these versions: vApps, vPC, QvDWS, vComputeServer), then yes, you need a license and you need a specific version of the software which is only available from the Licensing Portal.

If you want to use Passthrough, and only intend to use it for Computational workloads (without graphical acceleration, because there won't be any as it's the wrong type of driver) then you don't need a license, and you can download this driver from the normal NVIDIA Drivers site.

Regards

MG

#2
Posted 01/22/2020 09:13 AM   
Thank you for your reply! But now I am curious, if direct GPU Passthrough does not need vGPU Software (and license), why vGPU take 1 vComputeServer license per GPU in Passthrough mode? Is there some advanteges of using Passthrough with vGPU Software over Passthrough without vGPU (I meen in respect to computational workload, because for graphics mode you have explained the difference)?
Thank you for your reply!
But now I am curious, if direct GPU Passthrough does not need vGPU Software (and license), why vGPU take 1 vComputeServer license per GPU in Passthrough mode? Is there some advanteges of using Passthrough with vGPU Software over Passthrough without vGPU (I meen in respect to computational workload, because for graphics mode you have explained the difference)?

#3
Posted 01/22/2020 10:52 AM   
If you want to use multi-vGPU or a fraction of a GPU then you are good to go with vGPU instead of Passthrough. In addition, as soon as you run vGPU you are eligable for support. Passthrough instead (with default Tesla driver) doesn't include support. regards Simon
If you want to use multi-vGPU or a fraction of a GPU then you are good to go with vGPU instead of Passthrough. In addition, as soon as you run vGPU you are eligable for support. Passthrough instead (with default Tesla driver) doesn't include support.

regards
Simon

#4
Posted 01/22/2020 12:56 PM   
Hi vGPU allows you to share the GPUs resources between multiple concurrent VMs so you can run multiple workloads simultaneously. For example, instead of running the entire GPU inside 1 VM and doing nothing else until it's finished its task, you could run 2 concurrent VM workloads with an 8C Profile, or 4 concurrent VM workloads with a 4C Profile etc etc. It really depends how intensive your workloads are as to how you share the GPUs resources. Also, as Simon mentions, you also get the benefits of NVIDIA support as well as having a much more flexible platform. Regards MG
Hi

vGPU allows you to share the GPUs resources between multiple concurrent VMs so you can run multiple workloads simultaneously. For example, instead of running the entire GPU inside 1 VM and doing nothing else until it's finished its task, you could run 2 concurrent VM workloads with an 8C Profile, or 4 concurrent VM workloads with a 4C Profile etc etc. It really depends how intensive your workloads are as to how you share the GPUs resources. Also, as Simon mentions, you also get the benefits of NVIDIA support as well as having a much more flexible platform.

Regards

MG

#5
Posted 01/22/2020 01:41 PM   
Thank you all for your responses! I will consider different benefits that brings Virtual GPU Software. Perhaps Passthrough is not an optimal choice for my tasks.
Thank you all for your responses!
I will consider different benefits that brings Virtual GPU Software. Perhaps Passthrough is not an optimal choice for my tasks.

#6
Posted 01/22/2020 02:45 PM   
Scroll To Top

Add Reply