AI-Focused Neural Processing Chips : KL730

The KL730 is a neural processing unit (NPU) chip designed by Kneron, a San Diego-based full stack AI company. The chip is auto-grade and has an integrated image signal processor (ISP), which enables it to connect various digital signals such as image, video, audio, and millimeter waves. The chip is built specifically for machine learning and AI applications, and supports transformer neural networks that power generative pre-trained transformer (GPT) models. The chip offers a base-level compute power starting from 0.35 to 4 effective tera operations per second, and can run large language models (LLMs) either partially or fully offline.

The KL730 has significant business implications for the AI industry, as it tackles one of the major challenges of AI adoption: the high energy costs of running AI models. The chip is 3 to 4 times more energy efficient than previous Kneron models, and 150 to 200 percent more energy efficient than major industry peers. This means that users can run AI models at a lower cost and with less environmental impact. The chip also enhances data privacy and security, as it allows users to run AI models on their edge devices without connecting to the cloud. This gives users more control over their data and reduces the risk of data breaches. The KL730 is expected to bring secure and low-energy AI capabilities to a wide range of applications, from enterprise-edge servers to smart home appliances and advanced driving assistance systems.

Image Credit: Kneron