NVIDIA Multi-Instance GPU
AssessTechniques
A GPU partitioning technology that splits one GPU into isolated slices with dedicated compute and memory.
Why it's here
Placed in Assess: 1 article(s) of evidence from 1 source(s), led by framework updates, with 1 in the last 30 days. Confidence 24%. Low accumulated evidence, so it defaults conservatively pending more signal.
Evidence (1)
- 7The New Stack·8/6/2026framework_updateKubernetes 1.34 DRA aims to fix GPU scheduling pain
The article explains how Kubernetes historically treated GPUs as identical units, causing inefficient scheduling, out-of-memory failures, and wasted MIG capacity across mixed hardware clusters. It highlights Dynamic Resource Allocation in Kubernetes 1.34 as a new way to express GPU requirements more precisely, including memory, GPU generation, MIG profile preferences, and NVLink-connected devices.