Relying on the AI Software and Hardware Collaborative Innovation and Adaptation Verification Center, the Chinese Academy of Information and Communications Technology officially launched the “AishPerf Large Model Training Engineering Practice Performance Benchmark Research Project” based on the AishPerf AI Software and Hardware Benchmark System. The work takes “building benchmarks, gathering data, and promoting optimization” as the core direction. It focuses on the core performance bottlenecks of kilocalories to 10,000 kilocalories in the training and inference process. The system sorts out the definition caliber and empirical range of key indicators such as computational power utilization, video memory utilization, linear expansion efficiency, and ratio of effective training time, first term element latency, throughput, and video memory bandwidth utilization. Combined with the industry's public technical reports and measured data, it provides the industry with reference benchmarks for reliable engineering performance that can be referenced, benchmarked, and traceable Standardized development.

Relying on the AI Software and Hardware Collaborative Innovation and Adaptation Verification Center, the Chinese Academy of Information and Communications Technology officially launched the “AishPerf Large Model Training Engineering Practice Performance Benchmark Research Project” based on the AishPerf AI Software and Hardware Benchmark System. The work takes “building benchmarks, gathering data, and promoting optimization” as the core direction. It focuses on the core performance bottlenecks of kilocalories to 10,000 kilocalories in the training and inference process. The system sorts out the definition caliber and empirical range of key indicators such as computational power utilization, video memory utilization, linear expansion efficiency, and ratio of effective training time, first term element latency, throughput, and video memory bandwidth utilization. Combined with the industry's public technical reports and measured data, it provides the industry with reference benchmarks for reliable engineering performance that can be referenced, benchmarked, and traceable Standardized development.