GPU catalogue · Groq
Groq LPU
Deterministic inference silicon built around SRAM, not HBM.
GPU catalogue · Groq
Deterministic inference silicon built around SRAM, not HBM.
Specifications
Good for: Very low latency token generation.
| Machine | Groq LPU |
|---|---|
| Memory | 230 MB SRAM per chip; models are spread across many |
| Architecture | LPU |