On NVIDIA H100, creation of multiple compute instances after deletion of existing compute instances fails#

Description#

After compute instances are created and deleted on an NVIDIA H100 GPU, creation of multiple instances in a single nvidia-smi command fails. For example, the command nvidia-smi mig -cci 0,1,2 fails with the following error message:

Unable to create a compute instance on GPU  0 GPU instance ID  0 using profile 0: Invalid Argument
Failed to create compute instances: Invalid Argument

Workaround#

Create each compute instance in a separate nvidia-smi command, for example:

$ nvidia-smi mig -cci 0
$ nvidia-smi mig -cci 1
$ nvidia-smi mig cci 2

Status#

Open

Ref. ##

3829786