AI startup Recogni introduces a new computing method to cut costs and reduce power usage. The Economic Times
Recogni, an AI chip and software startup, made waves on Tuesday by introducing a groundbreaking computing method that could transform the efficiency of AI systems. This new approach, known as Pareto, promises to make AI chips smaller, faster, and cheaper to operate.
Supported by major players like BMW, Bosch, and venture capital firm Mayfield, Recogni focuses on creating specialized chips and software for AI inferencing. Inferencing is the process where AI models, once trained, make predictions or decisions based on new data. Recogni’s new system could revolutionize this process.
The patented Pareto system employs a logarithmic method that outperforms current technologies, offering significant improvements in key performance indicators for AI hardware. Gilles Backhus, Recogni's co-founder and VP of AI highlighted that Pareto represents a major advancement in AI computing. "It is a huge leap in all of the KPIs that influence silicon hardware system design when it comes to AI computing," he explained.
Currently, AI models, such as OpenAI's GPT-4 and Google's Gemini, rely on numerous complex and power-intensive mathematical operations, even for simple tasks like responding to prompts in chatbots. Recogni’s Pareto system addresses this challenge by converting multiplication operations into simpler additions. This adjustment drastically cuts down on power consumption while still maintaining high accuracy.
Recogni has already put Pareto to the test with AI models from Meta Platforms, Stability AI, and other major tech firms, demonstrating its effectiveness across various platforms. The company's first chip was created using Taiwan Semiconductor Manufacturing Co's advanced seven-nanometer process, showcasing its commitment to cutting-edge technology.
The startup is currently collaborating with an undisclosed partner to make Pareto more accessible and will reveal more details about this partnership soon. Recogni is exploring various deployment options, including partnerships with companies that provide hardware for data centers, potentially making Pareto available to a broader audience.
In summary, Recogni’s Pareto system represents a significant leap in AI technology, offering a more efficient and cost-effective solution for powering AI models. By simplifying computational processes and reducing power needs, this innovation could pave the way for more sustainable and scalable AI applications.
