4 ms·

Hyper-V machines do not have access to GPU, previous implementation is legacy and cannot be selected anymore. No one knows when the new implementation is going

• mfatica 7y ago
  So it sounds like maybe-yes? My use case is machine learning. I have an AMD GPU and tensorflow is supported through ROCm, but ROCm only runs under ubuntu. Currently I dual boot Windows/Ubuntu but would love to be able to stay in Windows.
  • vetinari 7y ago
    Direct GPU access requires way more than just virtual machine. You would need PCIe pass through, so the VM could control the GPU. For that, you would need either separate GPU, or a GPU that supports SR-IOV, and a board that supports IOMMU. ROCm specifically is not only userland, but also a kernel driver (amdkfd/amdgpu). To use it, you are going to dual boot into native Linux for a while.
    • mfatica 7y ago
      Ah bummer. Well hopefully someday soon the Windows software catches up
• withinrafael 7y ago
  Not true. Hyper-V supports Discrete Device Assignment [1] and therefore GPUs, USB hosts, etc. Where you'll run into issues is on the NVIDIA side. They deliberately cripple their consumer drivers to prevent use in virtual environments to extract licensing dollars. (Haven't tried AMD but last I remember, their cards don't support being pushed around by the IOMMU. Could be wrong.) [1] https://docs.microsoft.com/en-us/windows-server/virtualization/hyper-v/deploy/deploying-graphics-devices-using-dda https://docs.microsoft.com/en-us/windows-server/virtualizati...