The NVIDIA HGX H200 8-GPU Baseboard is a cutting-edge accelerated computing platform designed for the most intensive AI, HPC, and data center workloads. It integrates 8 NVIDIA H200 SXM GPUs with a massive 1.1 TB of HBM3e high-bandwidth memory, offering unprecedented capacity and throughput for training trillion-parameter large language models (LLMs), powering generative AI, and running complex scientific and industrial simulations. Equipped with the NVLink 4 switch, the system delivers up to 3.6 TB/s of all-to-all GPU bandwidth, ensuring ultra-fast interconnectivity, minimal latency, and maximum scalability across GPUs. Based on the NVIDIA Hopper architecture, the H200 introduces advanced features such as the Transformer Engine with FP8 precision, enhanced sparsity acceleration, and superior energy efficiency, enabling faster training and inference at lower power consumption. With unmatched compute density, scalability, and performance, the HGX H200 8-GPU Baseboard is the ideal foundation for building AI supercomputers, exascale HPC systems, and hyperscale cloud infrastructure, driving innovation in the era of accelerated computing.
The NVIDIA HGX H200 8-GPU Baseboard combines 8 H200 SXM GPUs, 1.1 TB HBM3e memory, and an NVLink 4 switch to deliver extreme performance and scalability for AI, HPC, and data center workloads.