On November 13, Arm China, a prominent domestic supplier of chip IP design and related services, organized a product launch in Shanghai to formally unveil its next-generation 'Zhouyi' X3 NPU IP. As the flagship product under Arm China's 'All in AI' initiative, the 'Zhouyi' X3 leverages the cutting-edge DSP+DSA architecture, tailored for large-scale AI models, with the ambition to establish a new standard for AI computing efficiency at the edge.

The product offers substantial enhancements in terms of performance, functionality, and ease of use. It supports configurations with up to 4 cores in a single cluster, delivering a flexible performance range from 8 to 80 FP8 TFLOPS, along with a single-core bandwidth of up to 256GB/s. In comparison to its forerunner, the 'Zhouyi' X2, the X3 model demonstrates a 30% to 50% performance boost in CNN models, with multi-core performance linearity reaching between 70% and 80%. Given the same performance benchmarks, its capacity for handling AIGC large models has surged tenfold.

Equipped with features like integrated self-developed decompression hardware WDC, innovative W4A8/W4A16 computational acceleration modes, and compatibility with multi-precision fused computing, it effectively tackles the complexities associated with deploying large AI models at the edge.

Furthermore, the 'Zhouyi' X3 is supported by the 'Zhouyi' Compass AI software platform. This platform streamlines development and deployment processes through a hardware-software co-design approach, offering comprehensive services that span from hardware and software solutions to after-sales support.

The 'Zhouyi' X3 is strategically positioned to cater to four primary sectors: infrastructure, intelligent vehicles, mobile devices, and smart IoT. It promises to deliver unparalleled AI computing experiences across a range of devices, including acceleration cards, smart cockpits, ADAS systems, embodied AI solutions, AI PCs, AI smartphones, smart gateways, and smart IP cameras.