MIG concepts and profiles
GPU Instance (GI) — A slice of the physical GPU with dedicated SM and memory.
Compute Instance (CI) — A sub-division of a GPU Instance defining compute resources.
MIG Profile — Format: Xg.Ymb (X = GPU slices, Y = memory in GB)
A100 80GB profiles:
| Profile | SMs | Memory | Max Instances |
|---|---|---|---|
| 7g.40gb | 98 (all) | 40 GB | 1 |
| 4g.20gb | 56 | 20 GB | 2 |
| 3g.20gb | 42 | 20 GB | 2 |
| 2g.10gb | 28 | 10 GB | 4 |
| 1g.5gb | 14 | 5 GB | 7 (maximum) |
Maximum instances on A100 80GB = 7 (using 1g.5gb profile).