How to Fix NVL Errors on VMs Running NVIDIA vGPU Manager
Suspending a vGPU-enabled VM on a host with different main release branch versions of the NVIDIA vGPU manager fails. This issue can occur in automated test environments when VMs are not cleanly shut down.
The nvidia-smi command can’t retrieve the driver version, license status, and accounting mode of a vGPU device. This problem affects Windows guest VMs.
nvlddmkm.sys
Some of the most common causes of NVL errors are faulty memory and problems with your hard drive. These issues can be fixed by running the CHKDSK utility in Windows. To run the utility, open a Command Prompt window as an administrator and type chkdsk /f c:. After the CHKDSK utility has completed, restart your computer.
vGPU-enabled VMs that are deployed by using VMware Horizon instant clone technology may fail to start or show a black screen. This occurs because of a translation lookaside buffer (TLB) invalidation in the NVIDIA vGPU software graphics driver. To resolve this issue, reduce the amount of memory allocated to each VM.
A vGPU-enabled VCID-based VM can become unresponsive after a reboot or power cycle. This is caused by a shortage of frame buffer memory for the NVIDIA hardware-based H.264/HEVC video encoder. To address this issue, disable NVENC on profiles that provide 512 Mbytes or less of frame buffer.
nvlddmkm.exe
If you are having trouble with NVL errors, you can try running CHKDSK in Command Prompt to check and repair the problem. The issue could also be due to problems with your hard drive, so it’s important to rule out these possibilities by using the in-built Windows utility.
NVIDIA vGPU software graphics drivers fail to install in guest VMs in environments with Security Enhanced Linux (SELinux). The error message alternates between “Getting devices ready: 50%” and “Preparation in progress”. To resolve the problem, you need to set SELinux to permissive mode.
nvlddmkm.dll
When you assign a 2Q or 3Q vGPU to a VM running Linux and using the NVIDIA hardware-based H.264/HEVC encoder in NVENC mode, the NVIDIA vGPU fails to capture the interactive logon message displayed to users. This is caused by a shortage of frame buffer memory. Use a 4Q virtual GPU type to provide more frame buffer memory for the vGPU.
Suspending a VM deployed using VMware Horizon instant clone technology and resuming it on a host that is not running the same version of NVIDIA vGPU software causes the vGPU to fail to acquire a license from the license server. This can occur in environments where non-persistent VMs are forcibly powered off or not properly shut down.
The NVIDIA Windows vGPU software driver contains a vulnerability in the kernel mode layer handler for DxgkDdiEscape, where calls to functions that require lower IRQL can be made under raised IRQL, leading to a denial of service. This issue is resolved in the NVIDIA vGPU software driver 15.2.
nvlddmkm.inf
When a Linux VM configured with a Tesla V100 or Tesla T4 vGPU is migrated from a host that uses the legacy NVIDIA vGPU software license server to a host that uses NVIDIA vGPU Manager, the VM becomes unstable and the error message XID error 31 is written to the log files on the destination hypervisor host. To workaround this issue, migrate the licenses to NVIDIA License System as part of upgrading to a new vGPU Manager release.
When a VMware Horizon instant-cloned VM is deleted and shut down, the NVIDIA vGPU software driver is not cleaned up, which causes the NVIDIA vGPU Software License Server to run out of licenses. To workaround this issue, add the NVIDIA vGPU Software license Server address to the NO_PROXY environment variable for each affected licensed client.