NVIDIA T4/VMWare "Could not initialize plugin" error when powering up too many VMs at the

Hi there,

does anyone encountered this strange issue?

When powering on all VMs on a single host some of the VMs report “Could not initialize plugin ‘/usr/lib64/vmware/plugin/libnvidia-vgx.so’ for vGPU ‘t4-2q’.” error?

After a little wait time i can start some to all of these VMs fine.

The server is a Lenovo SR650 with 4xT4 GPU inside. We use the 2Q profile and the server hosts 32 VMs.
Almost all VMs power up fine, all GPUs are used (saw in nvidia-smi and in the vCenter).

Could this be a driver related issue? I currently don’t have a test server to verify but this is my only guess since we use an older driver. Newer drivers had the issue that they crashed our CAD application.

Thank you

Hi

Which version of VMware and vGPU driver are you using?

Regards

MG

Release 8 (418.66) if this helps you to identify.

Hi

Thanks. Although supported, that’s over a year old and should be updated.

All I can initially suggest is upgrading to the latest 10.2 branch and see if that solves your issue.

Which version of VMware are you running? 10.2 supports 7.0 …

Regards

MG