Compute OS

Compute OS is the operational substrate for scientific software, not a hardware catalog.

Subsystems

• Cluster: Slurm, Kubernetes, queues, and quota policy.
• Network: IB, RDMA, and NCCL communication tuning.
• Accelerator: governance across GPU/NPU/DPU pools.
• Compiler: Triton, TVM, and graph-level optimization.
• Observability: reliability and platform visibility.

Objective

Provide a stable, scalable, and measurable compute operating layer for the Scientific Software Factory.