Skip to main content

Groq LPU (Language Processing Unit)

Overview

Groq LPU (Language Processing Unit) is Groq's proprietary deterministic LLM inference processor, renowned for its extremely low inference latency (LLM token generation measured in milliseconds). It employs a compiler-defined hardware architecture: no cache misses, no out-of-order execution, all latency predictable. GroqCloud provides API services.

Core Specifications (GroqChip v1)

ItemSpec
ArchitectureTensor Streaming Processor (TSP)
ProcessTSMC 14nm
On-chip SRAM228 MB
INT8 Compute (est.)1,000 TOPS
FP16188 TFLOPS (official)
InterconnectGroqSync (deterministic interconnect)
TDP300 W (system-level)
Form FactorGroqRack (8 GroqChips)

Vendor Information

ItemDetails
ManufacturerGroq Inc.
Websitehttps://groq.com
Product Pagehttps://groq.com/products/
API Consolehttps://console.groq.com
SDKhttps://github.com/groqcloud
Documentationhttps://docs.groq.com

Key Features

  • Deterministic architecture: All execution predictable
  • Ultra-low latency: LLM inference 5-10x faster than GPU
  • Compiler-driven: Software-defined hardware
  • Streaming tensor architecture

Use Cases

  • Real-time LLM inference (Chatbot, Code assistant)
  • Low-latency AI services
  • GroqCloud API (per-token billing)