Can I use any NVIDIA GPU for configuring GPU partitioning?

Question

Hello.

I want to assign GPU partition to three VMs.

I'm looking to buy a new GPU, but the link below only shows 8 GPU models that support GPU partition.

https://learn.microsoft.com/en-us/windows-server/virtualization/hyper-v/gpu-partitioning

However, when I googled, I found a few articles that created GPU Partition with NVIDIA GPUs that are not included in the link.

Can I use GPU Partition even if I don't have 8 GPUs in the link?

I think we can't use 8 GPUs due to cost.

Any advice would be appreciated.

Answer 1

hi, I would like to clarify some things and offer some indications on your situation. GPU partitioning with NVIDIA GPU While the official documentation you referred to from Microsoft lists only specific GPU models (such as NVIDIA Tesla T4, A100, A40, etc.)

For GPU partitioning with Hyper-V, NVIDIA vGPU technology (which allows GPU partitioning) has wider compatibility with some additional GPUs, although not always officially supported in all contexts or use cases. Key points to consider: Officially supported GPUs: the list provided by Microsoft mainly includes corporate-level GPUs that officially support GPU partitioning via NVIDIA vGPU or vComputeServer.

These GPUs are designed for virtualization and server workloads and are equipped with the software and license options needed to support multiple virtual machines (VMs) that share a GPU.

Unofficial GPUs:

You may have found articles where users have configured GPU partitioning on GPUs not listed in the official documentation.

This is sometimes possible with older or consumer-grade GPUs, but typically involves alternative solutions or hacks that may not offer the same level of performance, stability or official support.

GPUs that are not officially supported often lack the complete suite of software tools and the optimizations necessary for GPU partitioning and their use in this context can cause problems.

License requirements:

even if you manage to operate GPU partitioning with an unlisted GPU, you should still have the appropriate NVIDIA vGPU licenses.

For some GPUs (such as GeForce or Consumer Level Framework), this may not be available or require specialized licenses, which can be expensive.

Cost Concerns: Given your mention of the costs, corporate GPUs that officially support GPU partitioning can really be expensive.

However, if cost is an important factor, consider the following options: NVIDIA Framework GPUs:

some Quadro GPUs (such as the Quadro RTX series) are often cheaper than Tesla / Datacenter cards and may still support GPU partitioning in some cases, particularly if you are using a professional configuration or workstation.

GeForce CPU with CUDA:

while consumer-level GeForce GPUs do not officially support partitioning via the vGPU solution, CUDA can still be used for some GPU sharing in less demanding use cases, although this is not as robust or flexible as real partitioning.

VMware or Hyper-V:

keep in mind that the software environment in which you are working (e.g. Hyper-V, VMware) is important. Official support for vGPU partitioning in Hyper-V is mainly for NVIDIA Tesla or Data Center GPUs.

However, other virtualization platforms such as VMware vSphere can offer greater flexibility in configuring GPU sharing or even partitioning with different levels of support.

and for your question "Can I use GPU Partition even if I don't have 8 GPUs in the link?"

Yes, you can potentially use GPU partitioning even if you don't have one of the 8 GPUs listed in the official Microsoft documentation. However, the ability to do this depends on several factors, including the specific GPU, virtualization platform and license constraints.

I hope I could have helped you with this answer.