What is MIG (Multi-Instance GPU)?
MIG partitions a single NVIDIA GPU into up to seven hardware-isolated instances, each with dedicated memory and compute.
MIG splits an A100/H100-class GPU into as many as seven instances, each with its own slice of memory and streaming multiprocessors. Because the split is in hardware, one instance can't consume another's memory or crash it — real isolation, unlike time-slicing.
This makes MIG ideal for multi-tenant sharing and many small/interactive workloads: you can pack notebooks, small inference services, and dev jobs onto one physical GPU with predictable performance.
The tradeoff is fixed profiles and a drain on reconfiguration. Reserve whole GPUs for large training jobs and MIG-split the rest — a mixed strategy the Compute planner will recommend.