FRANKFURT, Germany, June 20, 2017 — Inspur and Intel made a joint announcement at ISC 2017 that they would comprehensively enhance the Knights Landing Evaluation and Escalation Program (KEEP Program) to build the Knight’s Mill (KNM) Infrastructure Experience Platform by the end of the year 2017. Both companies also revealed that the application channels towards worldwide deep-learning and HPC users would open at the first quarter of next year.
KNM is the new Xeon Phi processor family that Intel plan to release at the fourth quarter of this year. KNM will use the same infrastructure as KNL, but adopts new instruction set and carries out a number of node optimization for the widely-used deep learning structure. It will achieve both single and double precision variation and efficient expansion, saving a lot of training time for the deep learning model.
Leijun Hu, Vice President of Inspur Group, said “Inspur and Intel had released the KEEP Program one year ago. We worked together to build the world’s first high-performance computing system based on KNL and provided application testing, optimization and migration platform for the HPC and Deep Learning users. After the program enhancement, the KEEP platform will simultaneously support both KNL and KNM, where the KNL specializes in HPC, and the KNM primarily focuses on AI deep learning. With the support of the leading 100G OPA network, the enhanced KEEP Program will keep up with the trend of HPC and AI integration, and provide satisfying experience for HPC and AI users across the world.”
Barry Davis, Intel’s General Manager of Accelerated Workload Group, stated that AI has already become an exceedingly important computing, and Intel’s KNM has been greatly optimized for deep learning and delivered better performance on the training for the deep learning model. The KNM platform of the enhanced KEEP Program will become an important part in Intel’s global AI computing strategy, and will bring more efficiency to the worldwide users in AI computing.
Apart from the KEEP program enhancement, the two companies also published the Caffe testing report for the world’s first 32 node KNL system. The report shows that a 30X speedup has been achieved on 32 KNL nodes compared with 1 node, and the real performance is 16,516 images per second.
As a leading total solution provider in cloud computing, big data and AI products and solutions, Inspur continues to develop intelligent AI computing. Intel is a world leading processor design and manufacture, and has been leading the way in the entire computing industry. The collaboration between the two companies not only focuses on AI computing processor and HPC acceleration equipment, but also on large-scale AI application research with the establishment of Inspur-Intel calculation collaborative laboratories in China. Intel and Inspur are committed to the application of AI technology in the energy, manufacturing, financial, medical, and insurance industries, and so forth. This will remarkably accelerate the establishment and improvement of global AI computing.