GTC 2020: Inside NVIDIA's Multi-Instance GPU Feature

GTC 2020 S21975
Presenters: Jay Duluk,NVIDIA; Piotr Jaroszynski, NVIDIA
Abstract
NVIDIA’s latest GPUs have an important new feature: Multi-Instance GPU (MIG). MIG allows large GPUs to be effectively divided into multiple instances of smaller GPUs. The primary benefit of the MIG feature is increasing GPU utilization by enabling the GPU to be efficiently shared by unrelated parallel compute workloads on bare metal, GPU pass-through, or on multiple vGPUs.

Watch this session
Join in the conversation below.

I am wondering why do we have to disable “L2 residency” feature when enable MIG?