On-wafer fabric per engine, 214 Pb/s; cluster networking via dedicated NET cabinets
A sixty-four-node Cerebras inference cluster, provisioned as thirty-eight cabinets: thirty-two WSE cabinets holding two CS-3 systems each, plus six NET cabinets. Provisioned heat load is approximately 1.86 MW, about 1.75 MW on water and 0.11 MW on air, Cerebras's own provisioned total including networking. The cluster holds 2.816 TB of on-chip SRAM and no DRAM. Each engine is rated at 125 petaFLOPS, footnoted by Cerebras as sparse; the sparsity is unstructured, so no dense figure is derived.