Today, Kneron Inc., a full-stack artificial intelligence business based in San Diego that manufactures neural processing units, announced the release of two new products: an AI-embedded PC chip for the low-power market and its most recent Edge AI server.
Featuring 48 TOPs of AI computing capacity, support for big language models, steady diffusion, and up to eight concurrent connections, the KNEO 330 is the company’s second private “Edge GPT” server that can do AI inference.
Launched in 2023, Kneron’s first device, the KNEO 300, already has enterprise clients in the manufacturing, financial services, and academic sectors. These devices, dubbed “GPT in a box” by Chief Executive Albert Liu, enable these companies to access business data and AI models inside the secure confines of their firewalls without transferring it to the cloud.
In a SiliconANGLE interview, Liu stated, “Privacy is quite important for applications in many ways.” It was his emphasis that by executing AI models locally, low-power embedded AI processors may improve privacy and security. He said, “particularly for law firms, medical, and financial institutions.” An enormous IT company, for instance, is reluctant to provide OpenAI with its data.
The academics at Stanford University and the University of California, Los Angeles are now using this device, but they have prohibited their staff members from uploading data to ChatGPT due to data privacy regulations. A lecturer can set up the unit, which Liu dubbed “GPT in a box,” with data and classwork material on their desk. It can then be used in private without ever leaving the campus, serving as a tutor and digital assistant to respond to inquiries concerning the course material without going against the rules.
Liu also mentioned that the KNEO 330 has a very low power consumption. When compared to the H100 workhorse from Nvidia Corp., the KNEO 330 uses only approximately 20 watts, while the H100 draws about 700 watts at its peak.
The KL830 is Kneron’s third-generation NPU chip designed for AI-integrated personal computers, and it was released alongside the KNEO 330. In order to execute AI applications at lower power and cost and to enable the creation of low-cost AI PCs, it can cooperate with the central processing unit and graphics processing unit. According to Liu, “the PC is more like a personal GPT with our chip inside.”
It saved thirty percent on energy use and increased product lifetimes, according to Liu, when paired with a top GPU. Two watts of power is the maximum that the KL380 can deliver. By enabling the integration of a low-power NPU with the CPU and GPU, these chips will open the door for more widely available AI PCs.
Additionally, the same chip may be used in industrial settings for AI internet of things devices. It can also be purchased as a USB dongle that adds edge AI capabilities to any device, including broadband routers, cameras, and traditional PCs.