An Azure service that is used to provision Windows and Linux virtual machines.
Hello Rince,
Thank you for reaching out and for providing the details of your current setup.
If a VM running on NVadsA10_v5 is still using NVIDIA vGPU driver version 17.x (550.x) after 15 May 2026, the VM itself will continue to stay up and running as normal. Azure will not automatically shut down, deallocate, or force reboot the virtual machine purely because of the driver version. From an infrastructure perspective, there is no direct action that stops the compute instance from running.
However, the key impact will be on GPU functionality once Azure completes the rollout of the updated host-side driver stack (vGPU 20.x / R595.x). Since vGPU 20.x is only compatible with guest drivers 18.x (570.x) and above, any VM still on 17.x will no longer have a supported GPU communication path with the host. In practical terms, this means the GPU will likely fail to initialize inside the guest operating system, and any workload relying on CUDA, rendering, or GPU acceleration will stop functioning or start failing with driver-related errors. The operating system and non-GPU applications will continue to work normally, but without GPU capability.
Another important point is that even if the VM appears stable after the rollout, GPU-related instability can surface during host maintenance events, redeployments, or service healing operations. In those situations, the GPU device may fail to attach correctly, which can lead to inconsistent behavior for GPU-dependent workloads.
So in summary, there is no automatic shutdown or forced VM termination expected due to the older driver version, but continuing with vGPU 17.x beyond the enforcement date will result in loss of GPU functionality and potential service disruption for any GPU-based applications.
To avoid this situation, the recommended approach is to upgrade all NVadsA10_v5 VMs to vGPU 18.x (570.x) or later before 15 May 2026 and validate the workloads in a controlled environment before rolling it out across production systems.
Hope this helps! Please let me know if you have any queries in comments.