On September 17, the Shengteng AI Artificial Intelligence Industry Summit was held, and the Shengteng 950 supernode was officially launched on site. At the same time, Shenzhou Kuntai, a subsidiary of Shenzhou Digital, officially released a next-generation AI computing power product matrix based on the Shengteng 950 technology route, covering KunTai A989 I5 standard servers, KunTai A989 I5 air-cooled supernodes, and KunTai PoD2000 A5 liquid-cooled supernodes to accurately adapt to core AI business scenarios such as large model training and high concurrent intelligent reasoning, building a high-performance computing power base for AI intelligent development in various industries. Among them, for large-scale data center construction, trillion-level model training, and central high concurrent inference scenario computing power, the KunTai PoD2000 A5 integrates 1024 cards in a single supernode, supports 256 TB unified memory addressing, and a kilocalories MTBF of up to 300 hours, leading the industry by more than 25%. Relying on the ultra-high bandwidth of TB-level NPU connectivity and ultra-low RTT latency of 3 μs, it breaks through communication bottlenecks for various resources such as computing and storage within the cluster.

On September 17, the Shengteng AI Artificial Intelligence Industry Summit was held, and the Shengteng 950 supernode was officially launched on site. At the same time, Shenzhou Kuntai, a subsidiary of Shenzhou Digital, officially released a next-generation AI computing power product matrix based on the Shengteng 950 technology route, covering KunTai A989 I5 standard servers, KunTai A989 I5 air-cooled supernodes, and KunTai PoD2000 A5 liquid-cooled supernodes to accurately adapt to core AI business scenarios such as large model training and high concurrent intelligent reasoning, building a high-performance computing power base for AI intelligent development in various industries. Among them, for large-scale data center construction, trillion-level model training, and central high concurrent inference scenario computing power, the KunTai PoD2000 A5 integrates 1024 cards in a single supernode, supports 256 TB unified memory addressing, and a kilocalories MTBF of up to 300 hours, leading the industry by more than 25%. Relying on the ultra-high bandwidth of TB-level NPU connectivity and ultra-low RTT latency of 3 μs, it breaks through communication bottlenecks for various resources such as computing and storage within the cluster.