VME and Nvidia vGPU and Windows Server and RDS question

[Albert Gostick]

Hi all,

I am in the process of recommending a new set of servers to a customer and they have asked about including an Nvidia card in each server just to speed up video rendering for their users.  I normally get this info from the distribution channel (TDSynnex Canada) but they seem short on this specific question.  The customer is also thinking of switching from VMware to VME but because it is just in the "thinking" stage, they have limited understanding of VME (as do I).

The question is: can VME take advantage of an Nvidia A2 card and use its vGPU manager to allocate vGPU slices to an Remote Desktop Services (RDS) solution that servers up RDS sessions to users.  To make this more concrete, if the Hypervisor was VME and the hardware (DL360 Proliant) had an Nvidia A2 card in it (16GB), can the Nvidia management software server up a portion of the A2 GPU as a vGPU to a RDS session that is sitting "inside" of a Windows Server 2025 RDS session host?

I did not know where else to go to post this - it is not a support issue (it is a pre-sales engineering question but so far not been able to get it via pre-sales at distribution).

Thanks...even if you can point me to a more appropriate place, that would be great.

Albert

[CalvinZito · Accepted solution]

Hey Albert - are you an HPE partner? If you are, there's HPE TechPro, a great pre-sales resource for partners. There are Slack channels where you can ask questions like this. So if you're a partner and you aren't leveraging HPE TechPro, I highly recommend you do. https://techpro.hpe.com/

That said, VM Essentials today supports 1:1 GPU mapping, not slicing. It is on the list of things the team is working on. 

[Albert Gostick]

Thanks.  I will check out the site.

I assume "GPU slicing" means the ability to aportion a section of the GPU to desktop sessions within Microsoft RDS?

And if so, is there a timeline for this (I don't know if HPE publishes a roadmap for development or not - if so, please share link

Albert

[Albert Gostick]

Just talked to an HPE pre-sales tech rep (and yes, I am a partner) and he said that current VME does not allow 1:1 mapping between the GPU and individual VMs within VME - but that it is coming "soon".  And he said that from his understanding, there is no roadmap to partition the GPU to individual desktop sessions within one RDS server (vm) - that it just cannot be handled within the architecture of RDS.  He said if a GPU slice was allocated to an RDS VM, that potentially 1 person within the RDS VM (i.e. one user session) could tie up the entire GPU slice with an application (say like video rendering) and the other users within the RDS VM (say, if there were 9 others), might be locked out of using the GPU.  Does that sound right to you?  I can understand what he is suggesting.

Albert