Multi-Instance GPU

Multi-Instance GPU (MIG) is a technology NVIDIA introduced in 2020 with the A100, its first Ampere-architecture data center GPU. It divides a GPU's compute units and memory into as many as seven instances, each of which behaves like a separate GPU.

Data centers and cloud platforms use it to share one GPU among multiple users or workloads. Because each instance receives only part of the GPU's resources, it can run the same task more slowly than the full GPU.

This entry is based on AIPOST articles and widely known facts. If something is wrong, please send us a correction request.