Kneron, a San Diego-based company focused on edge AI computing solutions, announced today (August 2015) the launch of its new AI chip, the KL8, in Taiwan. The chip will power lightweight edge computing and GPT-level application solutions.
Prior to this, Kneron launched the KL2021, its first edge AI chip supporting the Transformer AI language framework, in 530. The newly released KL730 combines an automotive-grade NPU with an image processor, delivering 0.35-4 TOPS of computing performance and supporting lightweight GPT large-scale natural language model computation. It can be deployed in applications such as edge servers, smart homes, and automotive driver assistance systems. The KL730 boasts a 3-4x increase in computing performance compared to Kneron's previous chips, and a 1.5-2x improvement in performance compared to similar products on the market.
The KL730 is built on the third-generation KDP series reconfigurable NPU architecture, using the Arm Cortex-A55 CPU design. It can accelerate voice processing and other tasks through a built-in DSP. It is manufactured using TSMC's 12nm process and is compatible with Cafee, Tensorflow, Tensorflowlite, Pytorch, Keras, and ONNX computing frameworks, as well as artificial intelligence computing models including CNN, Transformer, and RNN Hybrid. It can be applied to computing scenarios such as smart security, autonomous driving, video conferencing, industrial cameras, and commercial robots.
The KL730 enables deployed applications to run GPT models on endpoints in a partially or fully offline manner. Combined with KNEO, Kneron's private, secure edge AI network, it enables AI to run more securely on edge devices, thereby achieving privacy and security protections. The five operating modes are suitable for enterprise server solutions, smart vehicles, and AI-powered medical devices, enhancing inter-device collaboration while increasing security.
"Running AI functions requires specialized chips with an architecture fundamentally different from anything we've seen before," said Juncheng Liu, Founder and CEO of Kneron. "Simply reusing adjacent technologies, such as dedicated graphics processing GPUs, won't do the job well. The KL730 will be a game-changer for edge AI computing. With its unprecedented efficiency and support for AI language frameworks like Transformer, it will deliver powerful AI computing capabilities to a wide range of industries, unleashing the full potential of AI while maintaining high security and protecting data privacy."
The KL730 is expected to begin providing samples to equipment manufacturers for testing and deployment in related edge computing devices.
