Moore Threads on Saturday unveiled its next-generation GPU architecture, Huagang, at its inaugural MUSA developer conference, as the Chinese chipmaker seeks to strengthen its position in AI and high-performance computing. The Huagang architecture supports full-precision computing from FP4 to FP64 and delivers a 50% increase in compute density, while improving energy efficiency by tenfold, the company said. Powered by Moore Threads’ self-developed MTLink high-speed interconnect technology, the architecture can scale to intelligent computing clusters of up to 100,000 GPUs. Based on the Huagang architecture, Moore Threads also launched two new chips: Huashan, an integrated processor designed for AI training and inference, and Lushan, a graphics rendering chip. [Moore Threads, in Chinese]
