Product Overview

The NVIDIA HGX H100 8-GPU Computing Server is the flagship integrated system designed for large-model training, deep learning and high-performance computing (HPC). It features 8× NVIDIA H100 80GB SXM5 GPUs interconnected via NVLink and NVSwitch to deliver unprecedented AI compute throughput. With dual-socket 4th/5th Gen Intel Xeon Scalable processors, high-speed DDR5 memory and NVMe all-flash storage, the platform is optimized for trillion-parameter model pre-training, scientific computing and AI inference clusters.

Key Features

• 8× NVIDIA H100 80GB SXM5 GPUs, HBM3e, NVLink 4th-gen 900GB/s interconnect
• NVSwitch full-mesh topology enabling all-to-all GPU P2P bandwidth
• Dual Intel Xeon Scalable CPUs, up to 64 cores per node
• Up to 2TB DDR5-4800 ECC RDIMM memory
• 8× NVMe U.2 hot-swappable SSDs, up to 61.44TB local storage
• Redundant platinum-level PSUs, intelligent thermal design

Typical Use Cases

• Large language model (LLM) pre-training & fine-tuning
• AI inference cluster for online / batch workloads
• Scientific computing, molecular simulation, computational fluid dynamics
• Autonomous driving and multi-modal model training platforms

Warranty & Services

3-year limited hardware warranty included; optional 24/7 on-site maintenance. Custom configurations, cluster delivery and on-site commissioning available. Contact our engineering team for detailed quotes.