Unable to start VMs with VGPU

Hello,
I have a vm in windows server 2019 on which I have an NVIDIA GRID vGPU grid-rtx8000-12q Ppofil.
When this vm is running, and I want to start another vm with the same NVIDIA GRID vGPU grid-rtx8000-12q profile, I get an error message saying "Insufficient resources. One or more devices (pciPassthru0) required by the TEST VM are not available on the host "

the ecc is indeed deactivated.
I am in vsphere client 7.0.2 and the vcenter is in the same version.
I have walked a lot of tracks, in vain …
Thanks for your help

Have a idea ?
Thanks

Please open a support ticket with NVES ti look into this. Could be a hardware defect or other reasons.

Regards Simon

Can you please show us how we open a ticket?

I assume you have valid SUMS if you are using vGPU. Then you should know how to create a support ticket. If you are evaluating the product then there is no chance to open a support ticket. You should contact a partner to assist in troubleshooting.

regards
Simon

We have a few A100 cards attached to our severs. We plan to split them into MIGs and attach the MIG slices into KVM virtual machines (VMs). Do we need to use vGPU license to attach MIG slices to the VMs or we can use PCI pass-through without licenses. Thank you for your support in advance.

When using a virtualization layer, adding vGPU capability to VM(s), vGPU licenses will be required.
https://docs.nvidia.com/grid/13.0/grid-vgpu-user-guide/index.html#features-grid-vgpu

Please contact your NVIDIA vGPU Partner for more details with regard to acquiring NVIDIA vGPU Software.

Thanks.

Hello,
i have the nvidia support and my problem is solve.
If you have a vm with a nvidia profil grid with 12q, you can’t power on another vm with a different nvidia profil grid.
My second VMs has a nvidia profil grid 4q, that the reason why she doesn’t start.
i put nvidia profil grid with 12q and she start.
Thank you for help

Ok, good to hear. But in your previous problem description you mentioned you are unable to start another 12q profile.
So this is different situation and we only support homogenous vGPU profiles at this stage. Anyways, good example how fast our NVES can help to solve such issues.

Best regards
Simon

Yeah strange, I also noticed he listed the same profile twice, so according to the “solution” it should’ve worked from the beginning. But is that actually true? I thought this is the whole purpose of vGPU, to be able to split it (and that includes different sizings in my opinion)?!

I have the same problem with VMs with profile “NVIDIA GRID vGPU grid_p40-4q”. Four identical hosts with a P40 each, four VMs with that profile, and I cannot migrate two VMs to one host…?