28.4d Enabling MIG: nvidia-smi -i 0 --mig-mode=1 and rebooting¶

📦 Cluster Orchestration 📖 Advanced GPU Resource Management

🧠 Context Introduction¶

Multi-Instance GPU (MIG) is a hardware-level feature available on NVIDIA Ampere and newer GPU architectures (such as A100, A30, H100, and H200). It allows a single physical GPU to be partitioned into multiple smaller, isolated GPU instances. Each instance has its own dedicated memory, cache, and compute resources — making them behave like independent GPUs.

Before you can use MIG, you must enable MIG mode on the GPU. This is done using the nvidia-smi command with the --mig-mode flag, followed by a system reboot. This topic walks you through that process step by step.

⚙️ What Does Enabling MIG Mode Do?¶

Activates MIG capability on the specified GPU (identified by its index, e.g., 0).
Prepares the GPU driver to accept MIG partition configurations.
Requires a system reboot to take full effect.
Once enabled, the GPU can be partitioned into up to 7 instances (depending on the GPU model).

🛠️ Step-by-Step: Enabling MIG Mode¶

1. Identify the GPU¶

Use the command: nvidia-smi
Look for the GPU index (e.g., 0, 1, 2) of the GPU you want to enable MIG on.
Confirm the GPU supports MIG (A100, A30, H100, H200, etc.).

2. Enable MIG Mode¶

Run the following command on the target GPU (index 0 in this example): nvidia-smi -i 0 --mig-mode=1
The -i 0 flag specifies GPU index 0.
The --mig-mode=1 flag enables MIG mode (use 0 to disable).

3. Reboot the System¶

After the command completes, you must reboot the system for the change to take effect.
Use: sudo reboot or the appropriate reboot command for your OS.

4. Verify MIG Mode is Enabled¶

After reboot, run: nvidia-smi
Look for the line: MIG Mode: Enabled under the GPU information.
If you see Disabled, the process was not completed successfully.

📊 Visual Representation: nvidia-smi MIG Activation flow¶

This flowchart maps out enabling MIG mode: enabling the state, resetting the GPU, and verifying partition configs.

flowchart LR StopApps["1. Terminate GPU processes"] --> EnableMIG["2. nvidia-smi -i 0 -mig 1 (Enable)"] EnableMIG --> ResetGPU["3. nvidia-smi --gpu-reset"] ResetGPU --> Verify["4. nvidia-smi mig -lgip (List profiles)"] classDef cpu fill:#eafaf1,stroke:#76b900,stroke-width:2px,rx:6px,ry:6px; classDef memory fill:#f0f7ff,stroke:#3498db,stroke-width:1.5px,rx:4px,ry:4px; classDef system fill:#f1f5f9,stroke:#64748b,stroke-width:1.5px; class EnableMIG cpu; class Verify memory; class StopApps,ResetGPU system;

📊 MIG Mode States¶

State	Command	Description
Disabled	nvidia-smi -i 0 --mig-mode=0	MIG is off. GPU runs as a single unit.
Enabled	nvidia-smi -i 0 --mig-mode=1	MIG is on. GPU can be partitioned.
Pending	After command, before reboot	MIG mode is set but not active until reboot.

🕵️ Common Issues and Troubleshooting¶

"MIG mode is not supported" — The GPU does not support MIG. Check the GPU model.
"GPU is in use" — Stop any running processes on the GPU before enabling MIG.
MIG mode shows "Disabled" after reboot — The command may not have persisted. Re-run the enable command and reboot again.
System does not boot after enabling MIG — This is rare. Boot into recovery mode and disable MIG using nvidia-smi -i 0 --mig-mode=0.

✅ Key Takeaways¶

MIG mode must be enabled once per GPU and persists across reboots.
A system reboot is mandatory after enabling MIG mode.
After enabling, you can create MIG partitions using nvidia-smi mig -cgi commands.
MIG is ideal for workloads that require isolation, guaranteed performance, and multi-tenant GPU sharing.

📝 Quick Reference¶

For reference, the command to enable MIG mode on GPU index 0 is:

nvidia-smi -i 0 --mig-mode=1

📤 Output: The command will return no output if successful. You will see a message like "Reboot required" if the change is pending.

For reference, the command to verify MIG mode after reboot is:

nvidia-smi

📤 Output: Look for the line MIG Mode: Enabled under the GPU details. If you see Disabled, MIG is not active.