Multi-Instance GPU (MIG)

Partition NVIDIA A100 and H100 GPUs into hardware-isolated instances with guaranteed compute, memory bandwidth, and fault isolation for multi-tenant Kubernetes clusters.

Understanding MIG

Multi-Instance GPU (MIG) is a hardware-level feature available on NVIDIA A100, A30, and H100 GPUs that allows a single GPU to be partitioned into up to seven independent instances. Each instance has dedicated compute units, memory, and memory bandwidth — providing true hardware isolation.

MIG Partition Profiles

The A100 80GB supports these MIG profiles:

Enabling MIG in Kubernetes

Configure MIG using the GPU Operator's MIG manager:

# Enable MIG mode on the GPU (requires node drain)
sudo nvidia-smi -i 0 -mig 1

# Create MIG instances (example: 3 x 2g.20gb + 1 x 1g.10gb)
sudo nvidia-smi mig -i 0 -cgi 9,9,9,19 -C

# Verify MIG instances
nvidia-smi mig -lgi

Kubernetes MIG Strategies

The NVIDIA device plugin supports three MIG strategies:

• none: MIG is disabled. GPUs are exposed as whole devices.
• single: All MIG instances on a GPU must be the same profile. Resources are named nvidia.com/gpu.
• mixed: Different MIG profiles can coexist. Resources are named by profile, e.g., nvidia.com/mig-2g.20gb.

# Pod requesting a specific MIG profile
apiVersion: v1
kind: Pod
metadata:
  name: inference-small
spec:
  containers:
  - name: model
    image: nvcr.io/nvidia/pytorch:24.01-py3
    resources:
      limits:
        nvidia.com/mig-1g.10gb: 1  # Request a 1g.10gb MIG slice