China has made a significant breakthrough in artificial intelligence by training a single generative AI (GAI) model across multiple data centers and GPU architectures. According to Patrick Moorhead, Chief Analyst at Moor Insights & Strategy, this achievement is groundbreaking and showcases the industry’s ability to adapt in the face of challenges.
The development is crucial for China’s continued advancement in AI research, particularly with American sanctions limiting its access to high-end chips. To overcome this hurdle, Chinese researchers have been working on combining GPUs from different brands into a single training cluster. This technique allows them to utilize limited stocks of sanctioned high-performance chips while minimizing efficiency drops.
China’s success in developing a single GAI model across multiple data centers is a testament to the country’s determination to push forward with its AI ambitions, despite uncertainty and challenges. As Huawei stated, China will find ways to continue driving its AI development, even in the face of sanctions. This approach highlights the importance of innovation and adaptability in overcoming obstacles.
While using multiple data centers is not a new concept, the technical challenges involved have been significant. Electricity consumption was a major bottleneck, with some data centers consuming gigawatts of power, posing risks to the power grid. However, as electricity costs decrease, it’s likely that the industry will eventually adopt this approach as well.
The development demonstrates China’s commitment to advancing its AI capabilities and provides insight into the country’s efforts to overcome challenges in the face of sanctions.
Source: https://www.tomshardware.com/tech-industry/artificial-intelligence/china-makes-ai-breakthrough-reportedly-trains-generative-ai-model-across-multiple-data-centers-and-gpu-architectures