NVIDIA, at their GTC Taiwan event, introduced the NVIDIA HGX-2, a unified computing platform for both AI and high-performance computing equipped with 16 NVIDIA Tesla V100 graphics processing units. The 16 GPUs collectively provide half a terabyte of GPU memory and two petaflops of computing power.
The NVIDIA HGX-2 is a cloud-server platform with multi-precision computing capabilities and provide high flexibility to future-proof its computing power. It has achieved record AI training speeds. According to a statement by NVIDIA, the GPU server can process 15,500 images per second on the ResNet-50 training benchmark. It is capable of replacing up to 300 CPU-only servers. These high-precision calculations are done using FP64 and FP32 for scientific calculations and simulations and FP16 and Int8 for AI training and inference.
Many leading computer makers like Lenovo, QCT, Supermicro, Wiwynn, and manufacturers like Foxconn, Inventec, Quanta, and Wistron are planning to bring systems based on NVIDIA HGX-2 platform for large cloud data centers. HGX-2 will be used by manufacturers to create advanced systems to be used for AI and HPC. The first system built using HGX-2 was the NVIDIA DGX-2 which was recently announced.
The 16 GPU units that HGX-2 consists are able to work collectively using NVIDIA NVSwitch interconnect fabric. It links the 16 NVIDIA Tesla V100 Tensor Core GPUs to act as a single, giant GPU.
Jensen Huang, founder and chief executive officer of NVIDIA, speaking at the GPU Technology Conference in Taiwan, said: “The world of computing has changed.” He further added, “CPU scaling has slowed at a time when computing demand is skyrocketing. NVIDIA’s HGX-2 with Tensor Core GPUs gives the industry a powerful, versatile computing platform that fuses HPC and AI to solve the world’s grand challenges.”
HGX-2 follows the launch of the original NVIDIA HGX-1, at Computex 2017. The HGX-1 was powered by eight GPUs. The HGX-1 reference architecture was adopted worldwide by leading server makers and companies operating data centers, including Amazon Web Services, Facebook’s Big Basin and Microsoft Project Olympus.