Groq

Groq's LPU hardware runs Llama, Gemma and other open models at speeds up to 800 tokens per second, enabling low-latency AI applications requiring real-time response times.