Understanding the Role of Virtual GPU Software and Licensing

[JT-Trend]

I’m in the process of setting up a server with a GPU for use in virtualized environments, specifically for running 3D modeling software. My setup involves partitioning the GPU into four separate instances and assigning each one to a different VM. These VMs will be running 3D modeling software that requires GPU acceleration.

I’ve already installed the necessary drivers, partitioned the GPU into four virtual instances, and confirmed that each VM is able to access its allocated portion of the GPU without any issues. Essentially, each of the four VMs can use its respective GPU partition for the software, and the setup seems to be working perfectly so far.

With this configuration, I’m wondering about the role of Virtual GPU (vGPU) software and licensing. Since I have successfully partitioned and assigned the GPU to my VMs, do I still need to purchase or install additional software or licenses? If so, what specific vGPU licensing model would apply to my scenario? I’m assuming the software is necessary for managing these GPU partitions, but I would appreciate more clarity on how it fits into the overall setup.

Any insights or advice would be greatly appreciated!

Thanks!

[github-actions[bot]]

This issue is stale because it has been open 90 days with no activity. This issue will be closed in 30 days unless new comments are made or the stale label is removed. To skip these checks, apply the "lifecycle/frozen" label.

[rajathagasthya]

@francisguillier Do you have any thoughts on this?

[empovit]

@JT-Trend I would be interested to know how you did that, because the NVIDIA software for vGPU (virtual GPU) requires a license. Meaning you need a valid license even to download it.

[aishsnair]

@JT-Trend the primary vGPU based product for the Gfx use case you're describing maps to NVIDIA's Virtual workstation (vWS). Are you deploying the 3D modeling software on a VM with all GPUs passed through? Or are you using MIG-based partitioning (Multi Instance GPU) with vGPU? The exact mechanism matters for licensing and support.

In general:

If you’re using MIG on bare metal (no VMs): You can use the NVIDIA open-source data center drivers. No vGPU license is required.

If you’re doing GPU passthrough (1 physical GPU --> 1 VM): You can also use the data center drivers without any vGPU license. But you also don’t get support for GPU sharing across multiple VMs on a single GPU.

If you’re using time-sliced or MIG backed vGPU to share one GPU across multiple VMs ("4 partitions --> 4 VMs"): For a supported prod setup you need NVIDIA vGPU software and licenses (virtual Workstation) so you can download the host + guest drivers and be entitled to enterprise support.

Details: vGPU SW licensing guide