Tutorial
AWS tutorial demonstrates NVIDIA Run:ai GPU Fractioning on Amazon EKS
AWS published a walkthrough for using NVIDIA Run:ai GPU Fractioning on Amazon EKS to share a single GPU across Pods with enforced per-Pod memory limits. The validated example uses one g5.xlarge instance with an A10G GPU, two Pods each with an 8 GiB contract that allocate 6 GiB, and verifies that a 9 GiB allocation is rejected. It covers prerequisites, a custom AL2023 image build, KAI Scheduler and GPU Fractioning Operator installation, per-Pod metrics, and cleanup. The post notes that memory limits do not reserve proportional compute and that the test does not measure throughput gains or general fault isolation.
Amazon EKSNVIDIA Run:aiGPU sharing