Cerebras
#10 en Matériel d'inférence IACerebras · depuis März 2024 (CS-3 / WSE-3) · 2× · vu le 01 nov. 2026
35
Momentum
Cerebras is a product line of the AI hardware company Cerebras, based on the Wafer-Scale Engine 3 (WSE-3) and the CS-3 system, offered as a cloud inference API (pay-per-token) and as on-premises systems. According to the vendor, the successor CS-4 uses WSE-3 Turbo processors.
Historique du momentum
03.08.01.11.
Fonctionnalités
| Manufacturing Process (nm) | TSMC 5 nm |
| Platform | Cerebras Cloud (OpenAI-compatible API) and on-premises CS-3 systems |
| Price | GPT-OSS-120B: $0.35 input / $0.75 output per 1M tokens (pay-per-token) |
| Compute Performance (FLOPS/TOPS) | 125 PFLOPS (FP16, vendor figure stated as sparse) |
| Release Date | CS-3 / WSE-3: March 2024 |
| Memory | 44 GB on-chip SRAM, 21 PB/s memory bandwidth |
| Availability | CS-3 broadly available (on-prem and shared/private cloud); AWS deployment announced (March 13, 2026) |