Groq (Llama 4 Scout)

Groq delivers Llama 4 Scout at industry-leading inference speeds using its proprietary LPU (Language Processing Unit) architecture.

Groq provides the hardware backbone for Meta's Llama 4 Scout, utilizing its LPU (Language Processing Unit) to eliminate memory bandwidth bottlenecks. By achieving sub-second latency and high throughput (measured in hundreds of tokens per second), Groq allows developers to deploy Scout's reasoning capabilities for real-time applications. This stack bypasses traditional GPU clusters, offering a streamlined environment for low-latency inference and high-density compute tasks.

https://groq.com