Tesla M10, Hyper-v 2022

Virtual GPU Forums More vGPU Forums NVIDIA Virtual GPU Drivers
1

Hello. I have the following test server:
Nvidia Tesla M10 32Gb / 2xIntel E5-2667v2 / 512Gb / Supermicro X9DR3-F
server 2022 with Hyper-V role
nvidia driver version 474.82
I have received a temporary license for 90 days on the portal https://ui.licensing.nvidia.com/

How do I provide video processor resources to 10 virtual machines running Windows 10 x64 22H2? I am not considering DDA, because it will be broadcast only on 4 vWS .

2

Using an other hypervisor. No way with Server 2022 Hyper-V.

3

Can I use MS SRV 2025 or do I need to deploy this on vmware/proxmox?

4

Server 2025 supports GPU-P, so yes.

5

Changed the test server. Now the following characteristics:
Nvidia Tesla M10 32Gb / 2xIntel E5-2667v2 / 256Gb / Supermicro X9DRi-LN4+
server 2025 24H2 Std with Hyper-V role
nvidia driver version 474.82
the server has received 16 licenses

Windows Administration Center does not see the GPUs

how to properly transfer a partition to a VM?

6

I have previously found the reason. there is no SR-IOV support. I’ll try to update the BIOS.

7

Checking the operation of the SR-IOV function

PS C:\> (get-vmhost).IovSupport
True

Find GPU (There are 4 of them)

PS C:\> Get-VMHostPartitionableGpu


Name                          : \\?\PCI#VEN_10DE&DEV_13BD&SUBSYS_116010DE&REV_A2#6&17f903&0&00400010#{064092b3-625e-43bf-9eb5-dc845897dd59}\GPUPARAV
ValidPartitionCounts          : {32}
PartitionCount                : 32
TotalVRAM                     : 1000000000
AvailableVRAM                 : 1000000000
SupportsIncomingLiveMigration : False
MinPartitionVRAM              : 0
MaxPartitionVRAM              : 1000000000
OptimalPartitionVRAM          : 1000000000
TotalEncode                   : 18446744073709551615
AvailableEncode               : 18446744073709551615
MinPartitionEncode            : 0
MaxPartitionEncode            : 18446744073709551615
OptimalPartitionEncode        : 18446744073709551615
TotalDecode                   : 1000000000
AvailableDecode               : 1000000000
MinPartitionDecode            : 0
MaxPartitionDecode            : 1000000000
OptimalPartitionDecode        : 1000000000
TotalCompute                  : 1000000000
AvailableCompute              : 1000000000
MinPartitionCompute           : 0
MaxPartitionCompute           : 1000000000
OptimalPartitionCompute       : 1000000000
CimSession                    : CimSession: .
ComputerName                  : FRAME06
IsDeleted                     : False

Adding GPU to VM

$vm = "test_vgpu"

Add-VMGpuPartitionAdapter -VMName $vm -MaxPartitionCompute 1000000000 -MaxPartitionDecode 1000000000 -MaxPartitionEncode 18446744073709551615 -MaxPartitionVRAM 1000000000 -MinPartitionCompute 0 -MinPartitionDecode 0 -MinPartitionEncode 0 -MinPartitionVRAM 0 -OptimalPartitionCompute 1000000000 -OptimalPartitionDecode 1000000000 -OptimalPartitionEncode 18446744073709551614 -OptimalPartitionVRAM 1000000000 -Verbose -InstancePath "\\?\PCI#VEN_10DE&DEV_13BD&SUBSYS_116010DE&REV_A2#6&17f903&0&00400010#{064092b3-625e-43bf-9eb5-dc845897dd59}\GPUPARAV"

Set-VM -GuestControlledCacheTypes $true -VMName $vm
Set-VM -LowMemoryMappedIoSpace 1Gb -VMName $vm
Set-VM -HighMemoryMappedIoSpace 32GB -VMName $vm
Start-VM -Name $vm

After the powershell script is running, the VM stops running.
What am I doing wrong?

8

I continue the dialogue with myself.
the VM startup solution uses two new registry entries

[HKEY_LOCAL_MACHINE\SOFTWARE\Policies\Microsoft\Windows\HyperV]
"RequireSecureDeviceAssignment"=dword:00000000
"RequireSupportedDeviceAssignment"=dword:00000000

In the previous post, the addition to the virtual machine was not quite correctly indicated if there are several video processors (video adapters). By default, the system selects the first one. The names of the adapters can be obtained via PS

Get-VMHostPartitionableGpu | FL Name,ValidPartitionCounts

in the construction need add name

Add-VMGpuPartitionAdapter -VMName $vm --InstancePath 'Name'

(InstancePath=Name)