NVIDIA HGX H100 640GB 8-GPU Baseboard

The World’s Premier AI Supercomputing Platform

Transform your data center into an AI factory with the NVIDIA HGX H100. This 8-GPU baseboard is the engine behind the world’s most powerful supercomputers, designed to solve the most complex problems in AI, data analytics, and high-performance computing (HPC).

Featuring eight NVIDIA H100 Tensor Core GPUs interconnected via high-speed NVSwitch, this board delivers a massive 640GB of HBM3 memory and unified compute performance. It acts as a single, giant accelerator, enabling you to train massive Large Language Models (LLMs) like GPT-4 and run trillion-parameter inference with real-time responsiveness.

Key Features

• 8x H100 SXM5 GPUs: Integrates eight H100 Tensor Core GPUs on a single baseboard, delivering unprecedented density and compute power.
• 640GB HBM3 Memory: With 80GB of ultra-fast HBM3 memory per GPU, the system offers a combined 640GB capacity with unmatched bandwidth, eliminating data bottlenecks for the largest datasets.
• NVSwitch & NVLink Interconnect: The 4th Gen NVLink and NVSwitch technology connect all eight GPUs with 900 GB/s bandwidth per GPU, allowing them to function as a single, unified accelerator for massive model parallelism.
• Transformer Engine: Supercharges AI performance with FP8 precision, speeding up training by up to 9x and inference by up to 30x compared to the previous generation.

Technical Specifications

Ideal For:

• Massive Model Training: The standard for training foundation models (LLMs, Generative AI).
• Real-Time Inference: Deploying trillion-parameter models with low latency.
• Exascale HPC: Climate modeling, genomics, and drug discovery.