Partitioning MIG GPUs

Partitioning GPUs (MIG)

1- Drain the node

2- Run sudo nvidia-smi -mig 1 to convert ALL GPUs to MIGs.

3- Install nvidia-mig-parted on the node.

4- Istall nvidia-mig-manager on the node (https://github.com/NVIDIA/mig-parted/tree/main/deployments/systemd).

4- Write a config file config.yaml for the node to /etc/nvidia-mig-manager/config.yam.

An example is:

version: v1
mig-configs:
  current:
  - devices: all
    mig-enabled: true
    mig-devices:
      1g.10gb: 7
  all-disabled:
  - devices: all
    mig-enabled: true
    mig-devices:
      1g.10gb: 7

You can also ssh into an exisiting MIG node and run sudo nvidia-mig-parted export.

5- Test persistence with reboots.

6- Don't forget to uncordon.