Repository to demo GPU Partitioning with Time Slicing, MPS, MIG and others with Red Hat OpenShift AI and NVIDIA GPU Operator .
Check also the OpenShift GPU Partitioning Methods Docs if you want to know more.
Table of Contents
GPU Partitioning Overview Time-Slicing
Multi-Instance GPU (MIG)
Multi-Process Service (MPS) - Not yet available
Pros and Cons of GPU Partitioning Methods
Requirements
Usage Time-Slicing
MIG-Single
MIG-Mixed
MPS
NO GPU Partitioning - Default
Validate and Check GPU Partitioning
Testing with LLMs
Install Nvidia GPU Operator from Staging / Development
Other Interesting Links
0. Why GPU Partitioning?
GPU partitioning can help optimize GPU resource utilization across your clusters, though it might not be necessary if you have a single GPU with sufficient memory for your specific needs.
By partitioning GPUs, we can allocate the right-sized GPU resources to each workload, ensuring that each application gets exactly what it needs to perform efficiently.…
