CPU · Central Processing Unit
The host processor in a GPU node: runs the OS, storage and data pipelines, and orchestration while the accelerators do the tensor math; the CPU:GPU ratio is a node-design choice.
Current numbers
100 requests/s; 0.20 CPU-seconds/request; CPU occupancy ceiling 70%; thin host 16 cores; pool 32 cores; 30 active requests × 2.0 GB state; 40 GB runtime reserve; 128 GB host memory; sensitivity 120 requests/s.Size a host pool from CPU service demand — input ledger
29 required cores; 100 GB host state. The 32-core pool passes the screen; its CPU ceiling is about 110 requests/s.Size a host pool from CPU service demand — result and flip threshold
~8:1 → ~2:1 and belowGPU:CPU norm rebalancing toward more CPU per node as agentic RL adds rollout/tool/env load