how it will work
How it fits into GPU.ai
NVIDIA Run:ai is open source software for scheduling AI workloads on Kubernetes GPU clusters, including the KAI Scheduler and GPU fractioning. GPU.ai plans an integration so teams can run Run:ai on GPUs rented through GPU.ai and let it decide which workloads run where.
- Your workloadstraining, inference, notebooks
- Run:ai on Kubernetesqueues and GPU fractions
- GPU nodesrented on GPU.ai
Rent the GPUs on GPU.ai
The GPU machines for a Kubernetes cluster are launched through GPU.ai, like any other instance.
Run:ai schedules the work
Run:ai's queues, priorities, and GPU fractions decide which workloads run on those GPUs.
Grow with the queue
Add GPU nodes from GPU.ai when the queue grows, and release them when it empties.
NVIDIA Run:ai is a product of NVIDIA. This page describes a planned integration. GPU.ai is not affiliated with or endorsed by NVIDIA.