HGX is a platform designed by NVIDIA specifically for high-performance computing (HPC) and artificial intelligence (AI) workloads in data centers. HGX is not a specific product but rather a scalable hardware design that serves as a blueprint for building GPU-accelerated servers and systems.

Key features of NVIDIA HGX:

• Modular design: HGX platforms are modular, allowing data center operators and system integrators to build and configure servers tailored to their specific needs.

• GPU acceleration: Designed to accommodate NVIDIA GPUs, including Tesla and Ampere GPUs, which are optimized for AI, machine learning, deep learning and other compute-intensive tasks.

• Scalability: HGX platforms support scalable configurations, enabling organizations to scale up or scale out their computing infrastructure based on workload demands.

• Interconnectivity: Includes support for high-speed interconnect technologies like NVIDIA NVLink and PCI Express (PCIe), facilitating efficient communication between GPUs and other components.

• Compatibility: HGX platforms are designed to work with NVIDIA's software ecosystem, including the Compute Unified Device Architecture (CUDA) and NVIDIA GPU Cloud (NGC), providing optimized AI and HPC software containers.

Applications of NVIDIA HGX:

• AI and machine learning: Accelerating training and inference tasks for deep learning models in AI applications.

• High-performance computing (HPC): Performing complex simulations, scientific computations and data analytics in research, academia and industry.

• Data centers: Supporting the deployment of GPU-accelerated servers for cloud computing, data processing and enterprise applications.

NVIDIA HGX serves as a foundational platform that empowers data centers to leverage NVIDIA's GPU technology effectively for advanced computing tasks, providing scalability, performance and flexibility to meet the demands of modern AI and HPC workloads.