The industry’s first unified training system, built for reliable AI training at thousand-GPU scale and consistent operations across environments
SUNK (Slurm on Kubernetes) brings Slurm and Kubernetes together in one production-grade training system.
See how it enables:
- Up to 96% training goodput and 10× longer mean time to failure
- Unified Slurm and Kubernetes workflows on shared GPU infrastructure
- Deep observability through Mission Control and automated recovery with SUNK Auto Repair
- Faster deployment and consistent operations across environments with SUNK self-service and SUNK Anywhere
