Huawei reveals more details of the 384 self-developed chip solution, leading NVIDIA and AMD by one generation
Huawei Reveals More Details of 384 Self-developed Chip Solution, One Generation Ahead of NVIDIA and AMD
via cnBeta.COM Chinese Industry Information Station - Telegram Channel
Telegraph
Huawei Reveals More Details of 384 Self-developed Chip Solution, One Generation Ahead of NVIDIA and AMDHuawei reveals more details of 384 self-developed chip solution, one generation ahead of NVIDIA and AMD. According to Huawei, CloudMatrix 384 is built based on 384 Ascend chips. Through a fully interconnected topology architecture, it realizes efficient collaboration among chips and can provide up to 300 PFLOPs of dense BF16 computing power, nearly twice that of NVIDIA's GB200 NVL72 system. In addition, CM384 also has advantages in memory capacity and bandwidth. The total memory capacity is 3.6 times that of NVIDIA's solution, and the memory bandwidth also reaches 2.1 times, providing more efficient hardware support for large-scale AI training and inference. Although the performance of a single Ascend chip is about one-third that of NVIDIA's Blackwell architecture GPU, Huawei achieves... through scale...
