5

I have an ESXi server and am having issues passing a Radeon RX 5700 through to a Windows 10 VM. Windows sees the GPU, but reports that it has stopped the device because it has reported problems (Code 43).

I deleted the first VM I created for this and created a new one but it produced the same error. I also made a Debian VM which also recognized the GPU, but it was unable to use it.

The server is based on a Supermicro X9SRL-F, a Xeon E5-2650v2, and 128GB of DDR3 ECC memory.

The VM has 8 GB of RAM (all reserved) and 2 cores (1 socket). IOMMU is not exposed to the VM. I have tried with and without the "hypervisor.cpuid.v0" parameter (set to false) in the VM's configuration.

I have tried with and without adding the Vendor/Device ID for the GPU and associated HDMI audio channel to /etc/vmware/passthru.map with resetMethod set to default and fptShareable set to false.

"Above 4G Decoding" is enabled in BIOS.

There is an LSI00301 passed through to another VM (FreeNAS) on the same server and it has been working flawlessly for years.

The Radeon RX 5700 works when connected to a physical computer.

The GPU is connected to the server through a 1x to 16x PCIe Riser (v.009S) which has been confirmed to work with another computer.

I have tried different PCIe ports on the server.

The Windows 10 VM is fully updated (build 19041.508) and has Radeon Software Adrenalin 2020 Edition (v20.9.1) and VMware Tools (v10.3.10.12406962) installed.

There is a monitor connected to the GPU via HDMI.

The intended purpose of connecting the GPU to the server is for crypto-mining.

Improve this question

1 Answer 1

Reset to default
0

You need to adapt you VMX file a little:

  • shutdown the guest

  • edit the VMX

    hypervisor.cpuid.v0 = "FALSE"

    pciHole.start = "2048"

You can either connect via SSH and edit it using VI or download it from the datastore, update it with any (basic) editor you want, delete the original and upload the updated one.

Do not use a GUI editor to edit it; you can break it.

  • once the VMX is up to date, start the guest

You can also update the Virtual GPU Manager from ESX CLI:

  • enter maintenance mode
  • use software vib install -v to install the latest VGPUM
  • reboot
  • exit maintenance mode
Improve this answer
2
  • I un-registered the VM, added pciHole.start = "2048" to the vmx file (hypervisor.cpuid.v0 = "FALSE" was already there), and re-registered the VM to no avail. I also tried a number of other values for pciHole.start, and tried including pciHole.end, despite the fact that this version of ESXi should be handling that automatically. Regarding the vGPU package, I didn't do this as it is specific to Nvidia's GRID GPUs which are for shared and virtual 3D gpu features which don't apply to my configuration. I am only trying to achieve full PCIe passthru of an AMD gpu to a single VM.
    – 534N
    Commented Oct 2, 2020 at 21:10
  • Are your MxGPU VIB drivers up to date ?
    – Overmind
    Commented Oct 5, 2020 at 8:02

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .