Where to set VMGpuPartitionAdapter Params?

[eric-gitta-moore]

Generally speaking, if using PowerShell to configure GPU-PV, parameters such as OptimalPartitionCompute would be utilized. I would like to know if there is a way to set such parameters in the HCS API? For example, restricting a certain virtual machine to using only 50% of the GPU resources.

NVIDIA SOP of GPU-PV setup: https://docs.nvidia.com/vgpu/latest/pdf/grid-vgpu-user-guide.pdf (page 29)

# VM name
$vmName = "GuestVmWithGpu"

$hostGpus = Get-VMHostPartitionableGpu
$targetGpuIndex = 0
$targetGpu = $hostGpus[$targetGpuIndex]

# From nvidia docs https://docs.nvidia.com/vgpu/latest/pdf/grid-vgpu-user-guide.pdf
Remove-VMGpuPartitionAdapter -VMName $vmName
Add-VMGpuPartitionAdapter -VMName $vmName `
-MinPartitionVRAM $targetGpu.MinPartitionVRAM `
-MaxPartitionVRAM $targetGpu.MaxPartitionVRAM `
-OptimalPartitionVRAM $targetGpu.OptimalPartitionVRAM `
-MinPartitionEncode $targetGpu.MinPartitionEncode `
-MaxPartitionEncode $targetGpu.MaxPartitionEncode `
-OptimalPartitionEncode $targetGpu.OptimalPartitionEncode `
-MinPartitionDecode $targetGpu.MinPartitionDecode `
-MaxPartitionDecode $targetGpu.MaxPartitionDecode `
-OptimalPartitionDecode $targetGpu.OptimalPartitionDecode `
-MinPartitionCompute $targetGpu.MinPartitionCompute `
-MaxPartitionCompute $targetGpu.MaxPartitionCompute `
-OptimalPartitionCompute $targetGpu.OptimalPartitionCompute

# Allow the VM to control cache types for MMIO access.
Set-VM -GuestControlledCacheTypes $true -VMName $vmName

# Set the lower MMIO space to 1 GB to allow sufficient MMIO space to be mapped.
# This amount is twice the amount that the device must allow for alignment. Lower
# MMIO space is the address space below 4 GB and is required for any device that has
# 32-bit BAR memory
Set-VM -LowMemoryMappedIoSpace 1Gb -VMName $vmName

# Set the upper MMIO space to 32 GB to allow sufficient MMIO space to be mapped.
# This amount is twice the amount that the device must allow for alignment. Upper
# MMIO space is the address space above approximately 64 GB
Set-VM -HighMemoryMappedIoSpace 32GB -VMName $vmName

[MouriNaruto]

First, your document only mentions the GPU-P (Partition), not GPU-PV (Paravirtualization, just passthrough Direct3D syscall to guest, it's also the base of WSL2 GPU acceleration). It's two different features, lol. In the HCS schema document https://learn.microsoft.com/en-us/virtualization/api/hcs/schemareference#gpuconfiguration, you will know these are two different things.

HCS API can set MMIO space and is mentioned in the document https://learn.microsoft.com/en-us/virtualization/api/hcs/schemareference#virtualmachine_memory, but NanaBox has not exposed it because NanaBox officially only supports GPU-PV (actually HCS API only documents the GPU-PV things), not GPU-P (although some people in the community add some GPU-P support to NanaBox main branch before, but I have no opportunities to test.)

Some GPU-P parameters seem can be set in the undocumented references https://github.com/MouriNaruto/MouriDocs/tree/main/docs/4#json-schema-references, but I have not found the way to use that, maybe https://github.com/microsoft/hcsshim implementation snippet with helpful to achieve the goal.

Kenji Mouri

[eric-gitta-moore]

@MouriNaruto I have just carefully gone through these materials, but I haven't found any clue yet. Thank you for your efforts.