Recogni, an AI chip and software startup supported by major players like BMW, Bosch, and venture capital firm Mayfield, revealed a revolutionary computing method on Tuesday that promises to make AI chips smaller, faster, and more cost-effective. The company, known for developing specialized chips and software to power AI inferencing, introduced a new patented system called Pareto.
Pareto leverages a logarithmic approach that significantly outperforms traditional methods when running large AI models. This innovation marks a major leap in key performance indicators (KPIs) crucial to the design of AI computing hardware, according to Gilles Backhus, Recogni's co-founder and VP of AI.
AI models like OpenAI's GPT-4 and Google's Gemini typically rely on extensive, power-hungry mathematical operations to process even simple prompts in applications such as chatbots. Recogni's Pareto system innovatively converts these multiplication operations into additions, thereby dramatically reducing power consumption without compromising accuracy.
Recogni has already tested Pareto on AI models developed by companies like Meta Platforms and Stability AI, showcasing its potential to transform AI computing efficiency. The company's first chip, produced using Taiwan Semiconductor Manufacturing Co's seven-nanometer process, underscores their focus on cutting-edge technology.
Looking ahead, Recogni is collaborating with an unnamed partner to bring Pareto to a broader market. The company is exploring deployment routes that include providing hardware to data centers, potentially making it available on a rental basis for global users. This partnership is expected to be formally announced in the coming months.
With Pareto, Recogni is positioning itself as a leader in the next generation of AI hardware, aiming to address the growing demand for more efficient and scalable AI solutions.
