Configure Multi-Instance GPU (MIG): Quick Reference — Administration (NVIDIA-Certified Professional: AI Operations)

Configure Multi-Instance GPU (MIG) — Quick Reference

This quick reference provides essential facts and rules for configuring Multi-Instance GPU (MIG) on NVIDIA GPUs, a critical skill for the NVIDIA-Certified Professional: AI Operations exam under the Administration domain.

What is MIG?

• Multi-Instance GPU (MIG) enables partitioning a single NVIDIA A100 or later GPU into multiple isolated GPU instances.
• Each instance operates independently with dedicated memory, cache, and compute cores.
• Allows multiple AI workloads to run concurrently on one physical GPU, improving utilization and workload isolation.

Key Concepts

• GPU Instances: Logical partitions of the physical GPU created via MIG.
• Compute Instances: Subdivisions within GPU instances that run CUDA or other compute workloads.
• Profiles: Predefined configurations specifying the size and resources of each GPU instance.

Supported Hardware

• NVIDIA A100, A30, A40, and newer GPUs support MIG.
• Only specific GPU models support MIG; verify compatibility before configuration.

Enabling MIG

1. Ensure the GPU driver version supports MIG (typically NVIDIA driver 450 or newer).
2. Enable MIG mode on the GPU using nvidia-smi:

Enable MIG Mode

Command:

• sudo nvidia-smi -i GPU_ID -mig 1

Note: Enabling MIG mode requires a GPU reset or system reboot.

Creating GPU Instances

1. List available MIG profiles with:

nvidia-smi mig -i GPU_ID -lgip

1. Create a GPU instance with a selected profile:

nvidia-smi mig -i GPU_ID -cgi PROFILE_ID

1. Create compute instances within the GPU instance:

nvidia-smi mig -i GPU_ID -cci COMPUTE_PROFILE_ID -gi GPU_INSTANCE_ID

Managing MIG Instances

• List current MIG instances and compute instances:

nvidia-smi mig -i GPU_ID -L

• Delete instances when no longer needed:

nvidia-smi mig -i GPU_ID -dgi GPU_INSTANCE_ID

Best Practices

• Plan instance sizes based on workload resource requirements.
• Use smaller instances for lightweight AI inference tasks; larger instances for training.
• Monitor GPU utilization to optimize instance allocation.
• Automate MIG configuration in cluster environments for consistency.

Common Commands Summary

• sudo nvidia-smi -i GPU_ID -mig 1 — Enable MIG mode
• nvidia-smi mig -i GPU_ID -lgip — List GPU instance profiles
• nvidia-smi mig -i GPU_ID -cgi PROFILE_ID — Create GPU instance
• nvidia-smi mig -i GPU_ID -cci COMPUTE_PROFILE_ID -gi GPU_INSTANCE_ID — Create compute instance
• nvidia-smi mig -i GPU_ID -L — List MIG instances
• nvidia-smi mig -i GPU_ID -dgi GPU_INSTANCE_ID — Delete GPU instance

References

• NVIDIA MIG User Guide
• NVIDIA Developer Blog: Multi-Instance GPU (MIG)

More in this topic

Configure Multi-Instance GPU (MIG) — Administration (NVIDIA-Certified Professional: AI Operations)Describe datacenter architecture for AI workloads — Administration (NVIDIA-Certified Professional: AI Operations)Configure Multi-Instance GPU (MIG): Practice Questions — Administration (NVIDIA-Certified Professional: AI Operations)Administer Slurm clusters: Quick Reference — Administration (NVIDIA-Certified Professional: AI Operations)Administer Run:ai platforms — Administration (NVIDIA-Certified Professional: AI Operations)Configure Multi-Instance GPU (MIG): Worked Example — Administration (NVIDIA-Certified Professional: AI Operations)Administer Slurm clusters: Practice Questions — Administration (NVIDIA-Certified Professional: AI Operations)Administer Slurm clusters: Common Mistakes — Administration (NVIDIA-Certified Professional: AI Operations)Administer Slurm clusters — Administration (NVIDIA-Certified Professional: AI Operations)Administration — NVIDIA-Certified Professional: AI OperationsConfigure Multi-Instance GPU (MIG): Common Mistakes — Administration (NVIDIA-Certified Professional: AI Operations)Administer Slurm clusters: Worked Example — Administration (NVIDIA-Certified Professional: AI Operations)Administer Kubernetes environments — Administration (NVIDIA-Certified Professional: AI Operations)