Question Regarding A40 vs A6000 for VMWare vGPU Server

Hello,

My company is interested in building (via SuperMicro) a VM server utilizing vGPU with a VMWare solution to create 6 to 12 VM’s (with each VM being either Windows 11 Pro or some Linux distribution) and utilizing between 4 to 8 GB of vGPU RAM (host OS will likely be Ubuntu Server Edition).

We’re having SuperMicro build our system, and we specifically requested that they create the system with two Quadro A6000 with NVLink.

They came back to us and stated the following:

"I have finished discussing your GPU server configuration with our VMware Product Management concerning the AS -4124GS-TNR. It has been qualified onto the VMware Compatibility Guide, but this was performed with the NVIDIA A100, A40 and A30 GPUs. To make sure you receive complete support from VMware for the server in case of any issue, they are suggesting to use one of those GPUs (Qty. 2) in your configuration instead of the NVIDIA RTX A6000s.

The server with VMware does provide support for vGPU and DirectPath I/O functionality."

Though I’m familiar with the A6000, I am not with familiar at all with the A40.

My company needed the A6000 to provide multiple VM’s with at least 8 GB of vGPU memory (for 4K and 8K rendering). I’m worried however that moving to the A40 could be a problem, as the A40 seems more like a compute/CUDA card.

Another problem with the A40 is that it:

  1. Uses slower video memory.
  2. Has 1 less displayport.
  3. Uses a passive cooling rather than active.
  4. May not be as easily able to handle 8K rendering (at most the card would need to handle 3, 4K renders or 1 or 2 (at the most) 8K renders).

If anyone could help, I would greatly appreciate it (as my company is having me build a virtualized system, and I know very little about virtualized systems, let alone vGPU systems).

I greatly appreciate any help anyone could provide.

Thank you,
Nelson

Hi Nelson,

In general you could use A6000 or A40. Both GPUs have the same GA102 chip on it. If you want to use virtualization with Horizon, why do you think the display ports are relevant? No matter if you use A6000 or A40, both GPUs need to be in DC mode to support vGPU which also means that the GPUs run in headless mode.
Now the main difference: A40 is a DC board and the default mode is already capable to run vGPU. A6000 instead is a workstation GPU and therefore needs to be modified to run vGPU at all (mode selector tool).
For a performance perspective, the A6000 has a slightly increased memory clock which results in maybe 5% better performance compared to A40 on specific use cases.
Having a passive cooling is not a disadvantage for server systems. Actively cooled boards quite often cause air flow irritations in servers and therefore the best option here should always be to run passive cooled GPUs.

So the recommendation in your use case would be to use the A40 to have a fully supported stack.

Best regards
Simon