Connectivity Issue for nvidia 595 drivers with NVads_A10_v5 sku's

Question

Currently having connectivity issues with nvidia 595 drivers which are required for application that is being deployed on VM with the sku NVads_A10_v5. This sku is supposed to compatible with these drivers. Are there some recommended steps that the community would recommend to be followed?

Answer 1

Hey Michael — sorry you’re running into connectivity issues after installing the NVIDIA R595 (595) drivers on NVads_A10_v5. In general, when GPUs “should be compatible” but connectivity still fails, the most common root cause is that the installed driver branch isn’t actually supported by that VM SKU’s vGPU/driver compatibility matrix.

Recommended next steps

1. Confirm what driver is actually installed
   • Check the NVIDIA driver version on the VM (e.g., via nvidia-smi).
   • Make sure it’s truly the 595 (R595) branch and not an older/newer variant.
2. Compare against the supported driver branch for NVads_A10_v5
   • The NVads A10 v5 series is expected to support specific vGPU driver branches.
   • A practical workaround is to test with a supported vGPU driver branch (for example, the commonly referenced supported branch such as vGPU 18.x, rather than R595/20.x), to verify whether the issue is compatibility-related.
3. Reproduce and capture the exact failure
   • Identify whether the issue is:
     • GPU not reachable / GPU communication failing
     • application service failing to start
     • networking/ports from inside the VM failing
     • errors in system logs or NVIDIA components
4. Review how the driver was installed
   • The installation method can affect what’s actually deployed (for example: Azure VM extension vs manual install).
   • If you installed the driver manually, it’s worth double-checking you installed the right package for the VM’s vGPU/GPU model.

Quick questions we need to narrow it down

Could you share:

1. What exact connectivity issue are you seeing?
   • For example: application can’t communicate with the GPU, timeouts, device enumeration failures, etc.
2. Did the problem start immediately after upgrading/ installing the R595 driver?
3. What OS is the VM running, and what is the exact NVIDIA driver version now installed?
4. Please paste the output of nvidia-smi (and any relevant error messages).
5. How was the driver installed (Azure extension / NVIDIA vGPU package / manual)?
6. Does the application specify any required driver/CUDA/vGPU versions?

Once we have those details, it’ll be much easier to determine whether this is a supported driver mismatch vs. something else (like installation packaging, missing components, or application requirements).

---

References

• https://learn.microsoft.com/en-us/azure/virtual-machines/windows/n-series-driver-setup
• https://docs.azure.cn/en-us/virtual-machines/linux/n-series-driver-setup

Note: This content was drafted with the help of an AI system. Please verify the information before relying on it for decision-making.

Answer 2

I think Azure supports NVads_A10_v5 when you use the Azure‑published image with the GPU Driver Extension, and Azure gives no info on installing the 595 driver by hand.

Answer 3

For NVadsA10_v5, only specific NVIDIA vGPU driver branches are supported and must be used; using an unsupported branch such as vGPU 20 (595.xx) can cause issues.

Key points from the supported matrix:

• NVadsA10_v5 supports GRID/vGPU 17.x or higher, but:
  • vGPU20 (595.xx Linux/Windows drivers) is not yet supported on NVadsA10_v5.
  • vGPU18 is the latest explicitly supported vGPU branch for NVadsA10_v5 and is available in Public, China, and Azure Government regions only.
• vGPU20 is currently supported only on NCv6 RTX PRO 6000 BSE and NCasT4_v3 series, not on NVadsA10_v5.

Given this, recommended steps are:

1. Verify driver branch and SKU support
   • If using the 595.xx (vGPU20) driver on NVadsA10_v5, roll back to the supported vGPU18 driver for this SKU:
     • On Linux: use the vGPU18 GRID driver for NVadsA10_v5 from the NVIDIA GRID/vGPU section for Linux N‑series VMs.
     • On Windows: use the vGPU18 GRID driver for NVadsA10_v5 from the NVIDIA GRID/vGPU section for Windows N‑series VMs.
2. Ensure correct OS and distribution support
   • On Linux, ensure the guest OS is one of the supported distributions for the provided GRID/vGPU driver (for example, Ubuntu 20.04/22.04/24.04 LTS, RHEL 8.x/9.x, SLES 12/15, Rocky 8.4) as listed in the supported distributions table.
   • Mismatched OS/distribution and driver versions can cause instability or connectivity issues.
3. Use the Azure-provided vGPU drivers where possible
   • For NVadsA10_v5, use the Azure‑redistributed GRID/vGPU drivers (vGPU18) rather than generic NVIDIA packages, because these are validated for Azure hosts and include the correct licensing and compatibility.
4. Avoid mixing unsupported branches
   • The latest major driver branch (n) is only backward compatible with n‑1. Running n‑2 or older on hosts or guests when Azure rolls out a newer branch can cause failures.
   • Ensure the guest driver branch matches the supported branch for NVadsA10_v5 (currently vGPU18, not vGPU20).
5. If issues persist on a supported driver
   • Confirm that Secure Boot and vTPM are disabled on Linux VMs if using CUDA/vGPU drivers, as installation can hang when they are enabled.
   • For Windows, consider using the NVIDIA GPU Driver Extension for Windows N‑series VMs, which installs the appropriate GRID driver for the SKU, then verify extension status and logs if there are issues.

If the application strictly requires the 595 (vGPU20) branch, it is currently incompatible with NVadsA10_v5 according to the support matrix; consider using a SKU that supports vGPU20 (NCv6 RTX PRO 6000 BSE or NCasT4_v3) or working with the application vendor to validate against the supported vGPU18 branch on NVadsA10_v5.

---

References:

• Install NVIDIA GPU drivers on N-series VMs running Linux
• Install NVIDIA GPU drivers on N-series VMs running Windows
• NVIDIA GPU Driver Extension for Windows