NVIDIA Multi-Instance GPU
Type: Technology Tags: NVIDIA, MIG, Multi-Instance GPU, partitioning, data center, Kubernetes Related: NVIDIA-DGX, NVIDIA-GPU-Operator, NVIDIA-DCGM, NVIDIA-AI-Enterprise, NVIDIA-Hopper-Architecture, NVIDIA-Blackwell-Architecture, NVIDIA-Jetson-Thor Sources: https://docs.nvidia.com/datacenter/tesla/mig-user-guide/index.html Last Updated: 2026-04-29
Summary
NVIDIA Multi-Instance GPU (MIG) partitions supported NVIDIA data center GPUs into multiple isolated GPU instances with dedicated compute and memory resources. It improves utilization and isolation for multi-tenant inference, development, and smaller training workloads.
Detail
Purpose
Large data center GPUs can be underutilized by small or latency-sensitive jobs. MIG lets administrators split a supported GPU into isolated slices so multiple users or services can run with more predictable performance and stronger resource boundaries.
Key capabilities
- Partition supported GPUs into multiple GPU instances.
- Provide dedicated compute and memory resources per instance.
- Integrate with Docker, Kubernetes, NVIDIA-GPU-Operator, and monitoring tools.
- Improve fleet utilization for mixed-size AI workloads.
- Support operational visibility through NVIDIA-DCGM.
NVIDIA context
MIG is a critical AI factory capability for NVIDIA-DGX, Kubernetes clusters, and shared enterprise platforms running NVIDIA-NIM, notebooks, training jobs, or inference endpoints with different resource needs.
Connections
- NVIDIA-GPU-Operator - automates GPU stack components and can manage MIG in Kubernetes.
- NVIDIA-DCGM - exposes GPU telemetry and health information relevant to MIG partitions.
- NVIDIA-DGX - supported DGX systems can use MIG for multi-tenant utilization.
- NVIDIA-Hopper-Architecture - Hopper GPUs support MIG capabilities for data center partitioning.
- NVIDIA-Blackwell-Architecture - Blackwell systems continue the AI factory utilization story.
- NVIDIA-Jetson-Thor - NVIDIA public Jetson Thor material surfaces MIG as an embedded physical AI isolation feature.
Source Excerpts
- NVIDIA’s MIG guide covers partitioning supported GPUs into isolated instances with dedicated compute and memory resources.