Kneron Inc., a San Diego-based full-stack artificial intelligence company that builds neural processing units, today announced the launch of its latest Edge AI server and a new AI-embedded PC chip for the low-power market.

The KNEO 330 represents the company’s second private “Edge GPT” server capable of handling AI inference with 48 TOPs of AI computing power, up to eight concurrent connections, supports large language models and stable diffusion.

Kneron’s first unit, the KNEO 300, launched in 2023, already has enterprise customers in manufacturing, financial services and universities. These machines, which Chief Executive Albert Liu called “GPT in a box,” allow these organizations to use AI models within the privacy of their firewalls using enterprise data without sending it out into the cloud.

“Privacy in many ways for applications is quite important,” Liu told SiliconANGLE in an interview. He emphasized how low-power embedded AI chips could enhance both security and privacy by running AI models locally. “Especially for medical, financial institutions and law firms,” he added. “For example, a big tech giant, they’re not that willing to send their information to OpenAI.”

This device is currently being used at Stanford University and the University of California at Los Angeles by professors, which have banned their staff from uploading data to ChatGPT because of data privacy restrictions. The “GPT in a box,” as Liu affectionately called the unit, can be configured on a professor’s desk with classwork information and data. It can then be privately accessed without ever leaving the campus and act as a digital assistant and tutor to answer questions about the curriculum without violating the rules.

Similarly, Liu said, the KNEO 330 can run at extremely low power. Compared with Nvidia Corp.’s H100 workhorse, the KNEO 330 consumes only about 20 watts, whereas the H100 has a peak draw of around 700 watts.

In addition to the KNEO 330, Kneron released a new chip, the KL830, the company’s third-generation NPU chip for AI-embedded personal computers. It can work alongside the central processing unit and graphical processing unit to run AI applications at reduced power and cost and allow the production of low-cost AI PCs. “With our chip inside, the PC is more like a personal GPT,” said Liu.

According to Liu, when combined with a leading GPU, it saved 30% on energy use and extended product lifetimes. The KL380 provides a peak power consumption of two watts. These chips will pave the way for more accessible AI PCs by providing a way to embed a low-power NPU alongside the CPU and GPU.

The same chip can also be embedded in industrial applications for AI internet of things devices and it is available as a USB dongle that enables any device, such as a broadband router, a camera or a classic PC, to become edge AI-enabled.

Image: Pixabay