Designed to meet AI computing demands in autonomous driving, healthcare, voice recognition, molecular simulation, etc.
The adoption of a switching design enables dynamic connection of the CPU to multiple GPUs.

NVIDIA, together with Microsoft, unveiled the HGX-1, a hyperscale GPU accelerator that powers artificial intelligence (AI) cloud computing.

The HGX-1 hyperscale GPU accelerator was released as an open-source design following Microsoft's Project Olympus, providing a fast and flexible AI path for hyperscale data centers.

The HGX-1 architecture is designed to meet the explosively growing demand for cloud-based AI computing in various fields, including autonomous driving, personal healthcare, superhuman speech recognition, data and video analytics, and molecular simulation.



The HGX-1 architecture is powered by eight NVIDIA Tesla P100 GPUs per chassis and adopts an innovative switching design based on NVIDIA NVLink interconnect technology and PCIe standards, enabling CPUs to be dynamically connected to multiple GPUs. This supports cloud service providers standardized on the HGX-1 infrastructure to offer customers a wide range of CPU and GPU machine instance configurations.

Jen-Hsun Huang, co-founder and CEO of NVIDIA, stated, “Artificial intelligence is a new computing model that requires a new architecture,” adding, “Just as the ATX standard brought about the popularization of PCs today, the HGX-1 hyperscale GPU accelerator will contribute to the spread of AI cloud computing. The HGX-1 will enable cloud service providers to easily adopt NVIDIA GPUs and respond to the surging demand for AI computing.”

Kushagra Vaid, Senior Engineer and General Manager of Microsoft’s Azure Hardware Infrastructure, stated in a related blog post, "The HGX-1 AI accelerator provides exceptional performance scalability to meet the demanding requirements of rapidly growing machine learning workloads, and its unique design enables easy deployment in existing data centers worldwide."