Inspur NF5488A5 Server Sets AI Performance Record, Ranks First in Global MLPerf Test
On July 30, the world’s authoritative AI benchmark test MLPerf announced the latest results. Inspur’s NF5488A5 server set an AI performance record in the most regarded Resnet50 training task and ranked first among all submitted single-server performance results.
MLPerf is currently the most influential artificial intelligence industry benchmarking organization in the world. It was established in May 2018 and has garnered the support and participation of Amazon, Baidu, Facebook, Google, Harvard University, Intel, Nvidia, Microsoft, Alibaba, Inspur, Stanford, and other institutions and universities. This MLPerf 0.7 training benchmark test has 8 tasks involving typical deep learning business scenarios such as image classification, target detection, reinforcement learning, recommendation, and language translation. There are 9 organizations participating in this training benchmark test and submitting results, including Google, Nvidia, Intel, Alibaba, Tencent, Inspur, Dell, Fujitsu, SIAT, etc.
Resnet50 is the world’s most widely used “gold standard” for performance evaluation of artificial intelligence computing systems and AI chips. In the Resnet50 training task of this benchmark test, Inspur’s NF5488A5 server was configured with 8 NVIDIA A100 GPUs and 2 EPYC 7742 processors. It took just 33.37 minutes to complete the ResNet50 model training, and its single-server performance score topped the list. The server also performed well and ranked among the best in other tasks on the benchmark test.
Inspur AI server NF5488A5
Inspur’s NF5488A5, an AI server released in May this year, supports 8 third-generation NVlink fully-connected NVIDIA A100 GPUs, and is equipped with 2 AMD CPUs that support PCIe 4.0. It can provide users with the ultimate AI performance and extreme speed. Its high bandwidth provides powerful computing power for various AI scenarios such as automated customer service, financial analysis, smart cities, and intelligent language processing.