AI性能：业界领先的服务器和基准测试.pdf

1、AI Performance:Industry-leading servers and benchmarksJoshua Bassey HPE HPC&AI Performance Team5thMay 2026.1AI performance at HPE means the demonstration of AI workloads under extreme circumstances to prove the quality of HPE systems and solutions.HPE strives to show the world,our customers,and ours

2、elves what is possible when you focus on excellence,then show others how to do the same.It isnt enough to understand that a technology just works,instead ask how well does it work?What is AI performance to HPE?2 HPC benchmarks designed for massive scale-out and high-bandwidth performance Thousands o

3、f nodes,Millions of cores,MegaWatts of power,ExaFLOPs of performance LINPACK:Reflects system performance by solving a standardized dense system of linear equations Double-precision FP64 HPE systems dominate Top500(#1-3,6 in top 10,13 in top 25 clusters)1 HPE Cray EX(x86_64 CPUs,AMD GPUs,HPE Slingsho

4、t)#1-3 top clusters HPE Cray XD670(x84_64 CPUs,H200,IB)#16(2nd APJ cluster),#78(top Canadian cluster)What does performance tell you?-Top500 310,000+Nodes11,000,000+Cores1.8 ExaFLOPsPerformance29 MegaWattsPower400+Nodes479,000+Cores145 PetaFLOPsPerformance3.6 MegaWattsPowerRank#1#161 Based on Top500-

5、November 2025 MLPerf is a set of open industry-standard benchmarks,results,and reference architectures MLPerf measures quality and performance to prove and improve:AI accelerators,server hardware,software,and AI models through Benchmarking for HPC,Storage,AI Training,AI Inference,AI Safety,&more Dat

6、asets that are open,large-scale,and diverse Research open collaboration and support with the research community helps accelerate and democratize scientific discovery,derive new insights for new breakthroughs in AI with reference code and systems Industry-wide consortium to compar