vGPU Tesla P100 does not apear in Compute Offering

[DaniloMurbach]

Problem

While running some lab tests to evaluate the vGPU capabilities, I encountered an issue where the Tesla P100 vGPU types are correctly detected at the hypervisor level, but they do not appear as options when creating a compute/service offering in the CloudStack UI.

##Environment:

Host: XCP-ng 8.3

GPU: Tesla P100 16GB

NVIDIA vGPU Driver: 535.183.04

##Observations:

The nvidia-smi vgpu output confirms the vGPU is operational:

# nvidia-smi vgpu
Tue Jul 22 16:08:38 2025
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 535.183.04             Driver Version: 535.183.04                |
|---------------------------------+------------------------------+------------+
| GPU  Name                       | Bus-Id                       | GPU-Util   |
|      vGPU ID     Name           | VM ID     VM Name            | vGPU-Util  |
|=================================+==============================+============|
|   0  Tesla P100-PCIE-16GB       | 00000000:05:00.0             |   1%       |
+---------------------------------+------------------------------+------------+

And xe vgpu-type-list shows the supported types:

# xe vgpu-type-list
uuid ( RO)              : 72a940ff-c8f4-7783-0650-8e603c1ac202
       vendor-name ( RO): NVIDIA Corporation
        model-name ( RO): GRID P100-1B
         max-heads ( RO): 4
    max-resolution ( RO): 5120x2880


uuid ( RO)              : c4ba708b-f747-8296-ff27-c7eab4afa5cb
       vendor-name ( RO): NVIDIA Corporation
        model-name ( RO): GRID P100-2B
         max-heads ( RO): 4
    max-resolution ( RO): 5120x2880

However, when trying to create a service offering in CloudStack, these vGPU types do not appear in the list.
Troubleshooting Steps Taken:

Verified the Tesla P100 is recognized and functional on the hypervisor.

Checked the database, but the GRID P100 types do not appear in expected tables.

Assumed the offering list might be populated from a DB table, but that does not seem to be the case.

##Question:

Is there a way to make the Tesla P100 GRID vGPU profiles (e.g., GRID P100-1B, GRID P100-2B) available when creating a compute/service offering in CloudStack?

Any insights or steps needed to manually enable or register these profiles would be appreciated.

Regards

versions

The versions of ACS, hypervisors, storage, network etc..

The steps to reproduce the bug

...

What to do about it?

No response

[boring-cyborg]

Thanks for opening your first issue here! Be sure to follow the issue template!

[vishesh92]

@DaniloMurbach As of now the options are hardcoded in the UI. You can always create the compute offering using API.
After #11143 is merged, you will be able to set the Compute Offering via UI.

To create using cmk, you can use the below command.

cmk create serviceoffering name=test123 cpunumber=1 cpuspeed=500 memory=1024 serviceofferingdetails[1].key=pciDevice serviceofferingdetails[1].value="Group of NVIDIA Corporation GP100GL [Tesla P100 PCIe 16GB] GPUs" serviceofferingdetails[2].key=vgpuType serviceofferingdetails[2].value=passthrough diskofferingstrictness=false

You need to ensure that the value for serviceofferingdetails[1].value= matches with group name in xenserver. You can check the group name with command xe gpu-group-list. Since you want to use vGPUs, replace passthrough with the required profile e.g. GRID P100-1B or GRID P100-2B.