Grokking DevOps and SRE Interviews
0% completed
Running GPU and Model Workloads
A cluster has 8 GPU nodes with 8 GPUs each, so it holds 64 GPUs. A GPU (graphics processing unit) is a processor with thousands of small cores, and most model training and inference runs on it. The dashboard shows 40 GPUs in use and 24 free.
A training job asks for 4 GPUs on one node, and it has waited for 50 minutes. Every node runs 5 busy GPUs and has 3 free, so no single node can hold the job.
This optional lesson covers an emerging area whose tools change quickly. It explains how Kubernetes allocates GPUs, and how to scale, release, and upgrade model workloads without wasting them
.....
.....
.....
Like the course? Get enrolled and start learning!
Reading Progress
0%