NVIDIA announced it would assist Taiwan server OEMs in developing custom GPU server solutions.
Artificial intelligence has grown increasingly mature over recent years as developers have aggressively pushed the technology forward, enabling applications including autonomous driving, robots, and voice assistant platforms. NVIDIA has established itself as a pioneer in AI technology development, leveraging its expertise in GPU parallel computing to become one of the top suppliers of deep learning hardware. Since 1999, NVIDIA's GPUs have evolved from engines for creating virtual worlds in video games and films to sophisticated systems capable of simulating human intelligence, running deep learning algorithms, and serving as the brain for computers, robots, and self-driving cars that can perceive and understand the world.
According to NVIDIA general manager and vice president of Accelerated Computing business unit Ian Buck, AI should be understood as "a new way of doing computing." Rather than requiring humans to provide specific instructions, AI's neural network can figure out algorithms itself from data. "By giving a few million images of an object to an image network, it will be able to learn the most efficient way to detect the object," Buck explained. This represents a fundamental shift where people can let AI figure out the code for them rather than writing it manually.
At Computex 2017, NVIDIA announced partnerships with Taiwan-based server players including Foxconn Electronics (Hon Hai Precision Industry), Inventec, Quanta Computer, and Wistron to develop GPU-oriented datacenter products aimed at fulfilling market demand for AI cloud computing hardware. NVIDIA will provide these server players early access to its HGX reference architecture—based on Microsoft's Project Olympus initiative, Facebook's Big Basin systems, and NVIDIA's DGX-1 AI supercomputers—along with GPU computing technologies and design guidelines. Through this cooperation, server players can choose how to design their datacenter server system layouts and which components to adopt to create differentiation and add value for customers.
NVIDIA's engineers will maintain close communication with these partners to ensure they receive necessary support for hardware buildups, enabling the players to develop datacenter server systems in the least amount of time and quickly release their products to market. NVIDIA recognizes that clients require special customization such as form factor and cooling methods for their server products and is working with the ODMs to create products using NVIDIA's HGX-1 architecture. According to NVIDIA's vice president of Solutions Architecture and Engineering Marc Hamilton, the HGX reference design meets the high-performance, efficiency, and scaling requirements that cloud datacenter servers need and serves as the basic architecture for ODMs to design their cloud computing datacenter servers.
The standard HGX design architecture includes eight NVIDIA Tesla GPU accelerators in the SXM2 form factor connected in a cube mesh using NVIDIA NVLink high-speed interconnects and optimized PCIe topologies. With a modular design, HGX enclosures are suited for deployment in existing data center racks globally, using hyperscale CPU nodes as needed. This builds on NVIDIA's earlier March collaboration with Microsoft to push the HGX-1 hyperscale GPU accelerator, an open-source design designed to meet demand for AI computing in fields such as autonomous driving, personalized healthcare, human voice recognition, data and video analytics, and molecular simulations. The HGX-1 is powered by eight NVIDIA Tesla P100 GPUs in each chassis, along with NVIDIA NVLink interconnect technology and the PCIe standard, enabling a CPU to dynamically connect to any number of GPUs.