Artificial Intelligence Computing Leadership from NVIDIA
Hyper-V 2019/RDSH 2019/Tesla M10
Hello, We are working on a new infrastructure which will contain 2 clustered Hyper-V 2019 hosts, each with 1 Tesla M10 board. They will be hosting 4x RDSH 2019 VM's in a 50/50 workload. As I have understood, for RDSH we would need the DDA function of Hyper-V and you can empower a maximum of 4 VM's with a Tesla M10 (4 GPU per board) through DDA, which would suffice in case one host fails or is put in maintenance. The documentation is referring to vGPU types such as M10-1A and the maximum amount of vGPU's for each GPU of a M10 (8 in the case of M10-1A). Does this mean it can support merely 8 RDSH sessions on 1 RDSH VM? Or is vGPU in this case referring to VDI VM's? So the questions are as following: - Is Server 2019 supported as a guest OS by Nvidia? - With 10-15 users per RDSH VM (totalling 60 RDSH users), is this a supported scenario with 60 CCU's? - Any remarks and tips regarding to clustered Hyper-V and GRID? Thanks in advance.
Hello,

We are working on a new infrastructure which will contain 2 clustered Hyper-V 2019 hosts, each with 1 Tesla M10 board. They will be hosting 4x RDSH 2019 VM's in a 50/50 workload.
As I have understood, for RDSH we would need the DDA function of Hyper-V and you can empower a maximum of 4 VM's with a Tesla M10 (4 GPU per board) through DDA, which would suffice in case one host fails or is put in maintenance.

The documentation is referring to vGPU types such as M10-1A and the maximum amount of vGPU's for each GPU of a M10 (8 in the case of M10-1A). Does this mean it can support merely 8 RDSH sessions on 1 RDSH VM? Or is vGPU in this case referring to VDI VM's? So the questions are as following:

- Is Server 2019 supported as a guest OS by Nvidia?
- With 10-15 users per RDSH VM (totalling 60 RDSH users), is this a supported scenario with 60 CCU's?
- Any remarks and tips regarding to clustered Hyper-V and GRID?

Thanks in advance.

#1
Posted 04/18/2019 05:44 PM   
Hi s3, you are on the right track. You need to run DDA and you can use 4xRDSH with 1xM10. So you have 8GB of FB per VM and this should be sufficient for up to 30 CCUs depending on the use case and load. 60 CCUs should be easily doable with 4x RDSH VMs per physical host. In general I see 100 - 120 CCUs per M10 in other customer deployments. vGPU 8.0 does support Server 2019! Hope this helps. regards Simon
Hi s3,

you are on the right track. You need to run DDA and you can use 4xRDSH with 1xM10. So you have 8GB of FB per VM and this should be sufficient for up to 30 CCUs depending on the use case and load. 60 CCUs should be easily doable with 4x RDSH VMs per physical host. In general I see 100 - 120 CCUs per M10 in other customer deployments.
vGPU 8.0 does support Server 2019!

Hope this helps.

regards
Simon

#2
Posted 04/18/2019 09:59 PM   
I have a question, where does one find the drivers to install an M10 when passed through using DDA. Or do we need to get an enterprise license/trial to download these.
I have a question, where does one find the drivers to install an M10 when passed through using DDA. Or do we need to get an enterprise license/trial to download these.

#3
Posted 11/22/2019 07:57 AM   
M10 requires licensing and therefore you can only get the drivers in the licensing portal
M10 requires licensing and therefore you can only get the drivers in the licensing portal

#4
Posted 11/24/2019 10:53 AM   
Scroll To Top

Add Reply