NVIDIA L4 GPU Cards on Windows Server Hyper-V VMs with DDA

[samadhan]

We are experiencing issues with our NVIDIA L4 GPU cards when attempting to use them with Discrete Device Assignment (DDA) on Windows Server Hyper-V VMs.

We followed the DDA setup steps as per the official documentation, and the process completed successfully without any errors.

1. However, the GPU is not being detected within the Hyper-V guest VM, which is running Debian 12. This has been verified using the lspci command, which does not show the GPU.

2. Notably, we have successfully used NVIDIA RTX GPUs in similar setups on Hyper-V VMs with DDA passthrough, without requiring any additional licenses or configurations.

Request for Clarification:

• Are there specific additional steps or configurations required to enable DDA passthrough for the NVIDIA L4 GPU cards?

• Does the L4 GPU require a vGPU license (GRID License) or any special driver/firmware for such use cases, even for single VM passthrough (non-vGPU setup)?

• What are the key differences between the RTX GPUs and the Tesla L4 GPUs that might explain this behavior?

[samadhan · Accepted solution]

Hi All,

I resolved the NVIDIA L4 GPU issue on Windows Server Hyper-V VMs with DDA by following NVIDIA's guidance: Bug #2812853: Microsoft DDA not working with some GPUs.

Solution:

The problem occurred due to insufficient MMIO (Memory-Mapped Input/Output) space for GPUs with more than 16 GB of memory. To fix this, allocate additional MMIO space based on the GPU's BAR1 memory and the number of assigned GPUs.

Formula:

MMIO space=2×gpu-bar1-memory×assigned-gpus 

Example:

For a VM with 1 NVIDIA L4 GPU (23 GB memory):

MMIO space=2×23 GB×1=46 GB

Run the PowerShell command:

Set-VM –HighMemoryMappedIoSpace 46GB –VMName <VM_Name>

Reboot the VM, and the GPU will be properly recognized.

Best Regards,
Samadhan

[Sunitha_Mod]

Hello @samadhan.

That's Excellent! 

We are delighted to hear you were able to find the solution and we appreciate you for keeping us posted.