Kevin Klues builds cloud-native infrastructure for GPUs at NVIDIA. He focuses on the Kubernetes components that manage heavy machine learning workloads, designing core tools like the Topology Manager and Dynamic Resource Allocation (DRA).
His recent talks break down the exact mechanics of making GPUs available to containers. He covers how to share resources efficiently—from basic time-slicing to using DRA for high-speed, multi-node training.
When he steps away from managing clusters, you can usually find him on a snowboard or somewhere up in the mountains.
