Number of virtual machine (VM) instances that can be run on a single virtual GPU (vGPU)

Hello everyone,

I have a query about the number of virtual machine (VM) instances that can be run on a single virtual GPU (vGPU).

Referring to the “Virtual-GPU-Packaging-and-Licensing-Guide” PDF, specifically at the bottom of page 10, there’s a statement: Maximum 10 concurrent VMs per GPU.

This implies that, for example, it wouldn’t be feasible to operate 16 VMs on an A16 GPU. However, I’m curious about a specific scenario: if I were to run 10 VMs on an A16 GPU, would the remaining 6GB of VRAM be rendered unusable or redundant?

Any insights or clarifications on this matter would be greatly appreciated.

You seem to mix-up different licensing models. This is just for the “former” vCS licensing which is based on GPU. For sure this is not relevant for the typical vGPU licensing (vApps, vPC or vWS)

I see, would you mind sharing with me the correct virtual licensing gpu PDF? I need to understand how the composition of FB and profiles work.

Hi, this is the right document but you mixed up vCS licensing (based on GPU) with the CCU licensing for graphics workloads.