A Shanghai-based firm has successfully built a 10,000-GPU computing cluster capable of staying online and available 99.99 percent of the time.

The achievement helps maintain China's globally leading progress in training multimodal models for the growth of artificial intelligence.

The company, under Shanghai's INESA, realized precise coordination of the GPUs, which range in architectures, generations and technical parameters, thus forming a highly complex and advanced computing system.

As one of China's first batch of 10,000-GPU intelligent computing clusters, the system is able to significantly enhance the efficiency of multimodal-model trainings and provide low-cost computing power services for AI enterprises and research institutions in Shanghai.