Home / server / Questions / 904771
Accepted
cyberquarks
Asked: 2018-03-28 02:46:08 +0800 CST

Hyper-V PCIe Passthrough of GPU working (but not 100%)

  • 772

I have managed to be able to passthrough a PCIe GPU (AMD RX580) into a Windows 10 Guest on a Windows Server 2016 Standard host.

The problem is that whenever the GPU will be used this error throws up:

enter image description here

The driver was installed through Windows update

enter image description here enter image description here

The PCIe passthrough was done using Powershell (Administrator):

  1. Set-VM -Name "Parsec" -AutomaticStopAction TurnOff
  2. Dismount-VmHostAssignableDevice -LocationPath "PCIROOT(0)#PCI(0301)#PCI(0000)" -force
  3. Add-VMAssignableDevice -LocationPath "PCIROOT(0)#PCI(0301)#PCI(0000)" -VMName "Parsec"
  4. Set-VM "Parsec" -GuestControlledCacheTypes $True -LowMemoryMappedIoSpace 2000MB -HighMemoryMappedIoSpace 8000MB

What could be missing here?

windows

3 Answers

  1. please follow this tutorial here https://youtu.be/XLLcc29EZ_8?t=570

    the script i used is below

    $vm = "YOUR_VM_NAME_HERE"

Add-VMGpuPartitionAdapter -VMName $vm
Set-VMGpuPartitionAdapter -VMName $vm -MinPartitionVRAM 80000000 -MaxPartitionVRAM 100000000 -OptimalPartitionVRAM 100000000 -MinPartitionEncode 80000000 -MaxPartitionEncode 100000000 -OptimalPartitionEncode 100000000 -MinPartitionDecode 80000000 -MaxPartitionDecode 100000000 -OptimalPartitionDecode 100000000 -MinPartitionCompute 80000000 -MaxPartitionCompute 100000000 -OptimalPartitionCompute 100000000

Set-VM -GuestControlledCacheTypes $true -VMName $vm
Set-VM -LowMemoryMappedIoSpace 1Gb -VMName $vm
Set-VM –HighMemoryMappedIoSpace 32GB –VMName $vm
    • 1

  2. looks like you followed Microsofts TechCommunity Blog: https://techcommunity.microsoft.com/t5/virtualization/discrete-device-assignment-gpus/ba-p/382266

    The last paragraph states that the CPU can hold back the GPU and you should set

    Set-VM "Parsec" -GuestControlledCacheTypes $True
    • 0
  3. Best Answer

    As of today the best and most effective way to do this is through https://github.com/jamesstringerparsec/Easy-GPU-P

    This will "passthrough" (although not dedicated passthrough); the GPU will be usable inside the VM through some kind of shared driver (not sure of the internals yet--but I've tested and it works like a charm)

    • 0