Hugging GPUs? Solving the Utilization Problem AI/ML Teams Run Into

11/30/2022
Hugging_GPUs

Last month at KubeCon, we caught up with the Co-founder and CTO at Run:ai, Ronen Dar, as we took a car ride (thank you Ford Motor Company for providing us with the Mustang Mach-E!) with KBE Insider host and Clinical Assistant Professor of Computing and Data Sciences at Boston University, Langdon White, on the roads of Detroit. Run:ai is solving the problem of GPU utilization, an expensive resource when it comes to running AI/ML workloads, by offering an orchestration platform allowing AI/ML teams full visibility and automated GPU scheduling capabilities.

Ronen talks about scheduling capabilities from the high-performance computing (HPC) world that was missing with Kubernetes and so Run:ai built it and now they’re running full speed to help AI/ML teams better utilize their GPU resources while benefiting from everything Kubernetes offers. 

Watch the full interview and hear more from Ronen as he explains more about the benefits of Kubernetes and the problem Run:ai solves.

Open Source Tools to Help Run Efficient AI/ML Workloads

Ronen mentions the importance of giving back to the Kubernetes community, a theme we heard from our interviews at KubeCon. 

If you find yourself “hugging GPUs” as Ronen references in the car interview, here are a couple of open source projects Run:ai offers that we think can help with your AI/ML transformation:

ICYMI, Run:ai had two engineers, Natasha Romm and Raz Rotenberg presenting at Kubernetes AI Day, one of the co-located events prior to KubeCon North America, on the GPU utilization challenge and the open source tools Run:ai provides. Check out the recording from KubeCon, “Are You Really Out of GPUs? How to Better Understand Your GPU Utilization”. 

Get started with AI/ML on Kubernetes

Want to learn more? Kube by Example (KBE) offers a free AI/ML with Jupyter on Kubernetes learning path. Shout-out to Kaitlyn Abdo and Josh Wood for contributing! At KBE we’re committed to keeping KBE content updated and relevant. If you would like to contribute or have suggestions, reach out to us, we’re always open to feedback!