Groq, a company distinct from Elon Musk’s similarly named Grok chatbot, has developed custom hardware aimed at significantly enhancing the speed of AI applications. Their primary innovation is the creation of Language Processing Units (LPUs), which are tailor-made for handling Large Language Models (LLMs) like GPT-3, used in AI chatbots. This approach differs from the traditional use of Graphics Processing Units (GPUs), which are more common in AI but optimized for parallel graphics processing rather than language tasks.
Groq’s LPUs excel in processing sequences of data, such as natural language, making them more efficient than GPUs for tasks involving LLMs. The company claims that its technology can run LLMs up to 10 times faster than GPU-based alternatives. This speed enhancement is crucial for real-time interactions with AI, like chatbot conversations or composing emails.
Located in Mountain View, Groq focuses on developing high-performance processors and software for AI, machine learning, and high-performance computing applications. Although the company does not train its own AI models, it enables faster operation of existing models developed by others.
Groq currently supports several LLMs, including Llama 2 from Meta, Mixtral-8x7b, and Mistral 7B. Users can test these models for free using text prompts without the need for software installation. Groq’s unique approach of designing hardware specifically for the software they run sets them apart in the AI hardware landscape.
Visit Groq’s site HERE