Tencent Cloud has unveiled a self-developed high-performance computing network called Xingmai, which boasts the industry’s highest communication bandwidth of 3.2T. This network significantly increases GPU utilization by 40%, resulting in a reduction of model training costs by 30% to 60%, and delivers an impressive 10-fold improvement in communication performance for large-scale AI models. Built on Tencent Cloud’s next-generation computing cluster, known as HCC (high-performance computing cluster), Xingmai supports computing scales of up to 100,000 GPU cards. [36Kr, in Chinese]