Nvidia announced a new partnership with top server manufacturers such as Foxconn, Inventec, Quanta, and Wistron that aims to bring the HGX-1 servers faster to market. The servers target “AI cloud computing,” so the primary customers will be data center-owning companies.
The development of the HGX-1 server chassis was a collaboration between Microsoft, Nvidia, and Ingrasys. The companies optimized the enclosure for machine learning applications and open-sourced its design earlier this year. The HGX-1 boxes support up to eight GPUs in a single chassis, and up to 32 GPUs can work together when four HGX-1 boxes are interconnected.
The HGX-1 supports Nvidia’s latest compute GPUs, including the Pascal-based Tesla P100 and the latest Volta-based Tesla V100. However, HGX-1 also works with AMD Radeon GPUs and Intel machine learning accelerators.
Partnership With Top ODMs
Through the HGX Partner Program, Nvidia will provide top original design manufacturers (ODMs) such as Foxconn, Inventec, Quanta, and Wistron early access to its own HGX reference architecture, GPU technologies, and design guidelines. This partnership should allow the ODMs to build their own HGX boxes faster, thus reducing the time it takes to deploy them in their customers’ data centers.
Nvidia said that all of the top 10 hyperscale businesses use its GPUs to accelerate their machine learning projects. According to the company, its new Volta-based GPUs should be three times faster than the Pascal GPUs. This is in part due to the machine learning-optimized Tensor Cores, which are now included its new compute architecture.
“Accelerated computing is evolving rapidly — in just one year we tripled the deep learning performance in our Tesla GPUs — and this is having a significant impact on the way systems are designed,” said an Nvidia representative.“Through our HGX partner program, device makers can ensure they’re offering the latest AI technologies to the growing community of cloud computing providers,” he added.