Slurm on Top, Kubernetes Underneath.
Slurm scheduling on Kubernetes-operated infrastructure. Researchers keep the queues, partitions, and batch workflows they already use. Radiant operates provisioning, recovery, and lifecycle underneath.
The scheduler your team knows, the cluster they don't have to run
Slurm Interface
Familiar queues, partitions, priorities, and batch workflows, unchanged
Kubernetes Operations
Automated provisioning, scaling, and node recovery across the GPU fleet
One Control Plane
Compute, networking, storage, and lifecycle managed together
Pre-wired clusters
A cluster stands up with scheduler, node images, accounting, shared storage, GPU drivers, and recovery loops pre-configured.
Built for the Longest Runs
Direct-to-GPU scheduling
Jobs run on bare-metal capacity, no virtualization layer
Self-healing
Failed nodes are isolated, replaced, and reintegrated mid-run
Dynamic queues
Shift capacity across teams, projects, and training phases as priorities change
Unified GPU pool
Slurm and Kubernetes-native workloads draw from one capacity pool, no stranded compute between schedulers
