1. Set the VM configuration (MMIO limits).
   • Formula: 2 x (card memory) x (number of cards)
   • Example with 16 GB A2s (1 per server) -> 2 x 16 x 1 = 32 GB ~ 32768 MB
   • Example with 24 GB A30s (2 per server) -> 2x24x2=96 GB ~ 98304 MB
2. Locate the correct PCI location for the card.
3. Disable the GPU.
4. Dismount the GPU.
5. This removes the PCI device from the host server.
6. Assign PCI resource (GPU) to VM.
   Warning: Multiple cards mean multiple device locations. Be careful not mixing them.
   Set the VM configuration (MMIO limits).
7. Log in to the node through RDP or iDRAC.
8. Run the following script one line at a time.

   #Configure the VM for a Discrete Device Assignment
   $vm = 	"ddatest1"
   #Set automatic stop action to TurnOff
   Set-VM -Name $vm -AutomaticStopAction TurnOff
   #Enable Write-Combining on the CPU
   Set-VM -GuestControlledCacheTypes $true -VMName $vm
   #Configure 32 bit MMIO space
   Set-VM -LowMemoryMappedIoSpace 3Gb -VMName $vm
   #Configure Greater than 32 bit MMIO space
   Set-VM -HighMemoryMappedIoSpace 33280Mb -VMName $vm
   
   #Find the Location Path and disable the Device
   #Enumerate all PNP Devices on the system
   $pnpdevs = Get-PnpDevice -presentOnly
   #Select only those devices that are Display devices manufactured by NVIDIA
   $gpudevs = $pnpdevs |where-object {$_.Class -like "Display" -and $_.Manufacturer -like "NVIDIA"}
   #Select the location path of the first device that's available to be dismounted by the host.
   $locationPath = ($gpudevs | Get-PnpDeviceProperty DEVPKEY_Device_LocationPaths).data[0]
   #Disable the PNP Device
   Disable-PnpDevice  -InstanceId $gpudevs[0].InstanceId
   
   #Dismount the Device from the Host
   Dismount-VMHostAssignableDevice -force -LocationPath $locationPath
   
   #Assign the device to the guest VM.
   Add-VMAssignableDevice -LocationPath $locationPath -VMName $vm

The Extensions page is displayed.

This installs the GPU controls at a cluster level.

1. Ensure that the VM is turned off.
2. Under Extensions, click the GPUs tab.

   
   

   

   
   

   Under the GPUs tab, all the GPUs are listed per node.

   Note: The Nvidia driver package on installed to each server does not include the '3D Video Controller' INF driver. After the GPU is dismounted from the host, the Nvidia driver package that is installed on the VM will properly recognize the full Nvidia device ID.
   
   

   
   

   

   
   

3. Click the GPU Pools tab to create the GPU pools. Enter the following details:
   • Servers – Enter the server details.
   • GPU pool name – Enter the name of the GPU pool.
   • Select GPUs - Only one GPU per node per pool is allowed.
   • Select Assign without mitigation driver (not recommended) - This is because the mitigation driver is not available in the current release.

     
     

     

     
     

4. Select the Assign VM to GPU Pool to assign the GPU to the VM.
5. Select the server, pool and virtual machine.
6. Click advance and enter input the memory requirements.
   • Low memory mapped I/O space (in MB)
   • High memory mapped I/O space (in MB) – Adjust the maximum memory mapped I/O space to match your particular GPU. The formula is 2 x GPU RAM per attached GPU. So, a VM with two 16 GB A2s will need ((2x16GB) x 2) = 64 GB memory.

     
     

     

     
     

     Note: This can be calculated by known specifications, or you can run Survey DDA.ps1 provided my Microsoft. For more information, see Virtualization-Documentation and Discrete Device Assignment
     
     
   • MMIO - Set on per VM basis

     
     

     

     
     

   • Install the drivers inside the VM.
   • Transfer the executable files to the VM and run.

     
     

     

     
     

     
     

     

     
     

     
     

     

     
     

     
     

     

     
     

     
     

     

     
     

1. Turn off the VM.
2. Remove the GPU pool assignment inside WAC.
3. Quick/Live Migrate VM to the new node.
4. Assign the GPU pool on the destination node to the VM (if none exists, create one).
5. Power on the VM.
6. Log in to ensure that the VM is created successfully.

1. Turn off the VM.
2. Remove the GPU pool assignment inside WAC.

3. Quick/Live Migration is possible, and the GPU assignment must change (automatic assignment of GPU on the correct node).

   This triggers a driver failure in the VM.

4. Unassign the VM, and then reassign the VM to the correct GPU in the pool.
5. Power on the VM.
6. Log in to ensure that the VM is created successfully.

Run the command “lspci” to reveal the pci device that is attached to the VM.

There are different driver versions.