VMM api not supported on Nvidia A10

Accelerated Computing CUDA CUDA Programming and Performance
1

I am using the Cuda VMM driver api ( CUDA Driver API :: CUDA Toolkit Documentation) to implement a circular memory buffer on GPU. This works perfectly on my workstation.

However, when I try to run this in the cloud I get a not supported error from cuMemGetAllocationGranularity. I tried to search around but could find a solution or explanation why this is not supported.

From the GPU details

Device 0: “NVIDIA A10-8Q”
CUDA Driver Version: 12.2
CUDA Capability Major/Minor version number: 8.6

I would expect VMM api to be supported.

Can you help me?

2

A10-8Q is not a baremetal machine/instance. That is a VM, ie. vGPU. I suspect that VMM API is not supported in vGPU perhaps unless the vGPU profile corresponds to a full GPU instance (8Q profile is not a full GPU instance for A10).

I would expect that is the source of the “not supported” message. I’m not able to quickly find a doc link for that. You may get better help by asking about vGPU behavior on the vGPU forum.

3

Thank you, further information is here: VMM api support with vGPU - More vGPU Forums / Application Optimization - NVIDIA Developer Forums