GTC CHINA: H3C Accelerates AI Transformation with Efficient Computation

Recently, NVIDIA GTC China 2018 was held in Suzhou. The event brought together authoritative experts and practitioners of all industries to exchange views on topics in the fields of in-depth learning and AI. As leading digital solution provider, New H3C Group, under Tsinghua Unigroup (hereinafter as "H3C") attended the conference and showcased the GPU server products applicable for AI and in-depth learning workloads. At the conference, NVIDIA announced that the leading server manufacturers in China, including H3C, will launch a series of servers based on the newly released NVIDIA T4 GPU, which, supported by the new TuringTM architecture, uses the multi-precision Turning Tensor Core and the new RT Core to provide unprecedented performance, combined with an accelerated container software stack.

Breakthroughs in AI are opening up new market space and creating new opportunities for business innovation and reshaping of traditional industries. According to a report of IDC, an international authoritative consulting agency, the global spending on "cognitive and AI systems" is expected to exceed USD 24 billion in 2018 and USD 77.6 billion by 2020, with an annual compound growth rate exceeding 37.3%. GPUs with excellent parallel computing performance are becoming great platforms carrying AI applications. At the conference, H3C showed the latest computing systems applicable for AI workloads - the H3C UniServer R5200 G3 server and the HPE Apollo 6500 Gen10 system.

The H3C UniServer R5200G3 server is a new generation of GPU server developed for AI. As a standard 4U rack-mounted server, it provides a strong computing capacity with up to 10 dual-width GPUs or 20 single-width GPUs, and is designed with a CPU/GPU communication link optimized according to the GPUs' computing characteristics, providing users with best performance experience.

With 8 GPUs, the HPE Apollo 6500 Gen10 system provides single-precision computation of up to 125 TFLOPs2 for GPU-intensive workloads. Built for accelerated computing, the system can be adapted to various requirements with interconnected PCIe and NVLINK GPU.

Liu Hongcheng, general manager for industrial standard server products of H3C, said, "H3C is experienced in the design and R&D of servers, and has led the innovation and transformation of computing architecture. NVIDIA is also working to expand the boundary of AI computation with cutting-edge technologies and products. In the new era where AI is more mature for commercial use, H3C and NVIDIA will join hands to accelerate the innovations and breakthroughs in the underlying hardware architecture of AI and in-depth learning, optimize AI application experience by upgrading the computing capability, and accelerate the advent of a more intelligent world."