NVIDIA HGX: The benchmark for AI supercomputing and generative performance

Large language models (LLMs) and generative AI have fundamentally changed the requirements for data centres. Where classic simulations used to dominate, GPU memory bandwidth and interconnect efficiency now determine the success of AI projects. With the latest generation of the NVIDIA HGX™ platform – now including the groundbreaking Blackwell architecture – sysGen offers the technological answer to these peak loads.

• Request information about the HGX platform

Massive data sets and complex simulations cannot be solved by brute force alone. The NVIDIA HGX™ platform is a finely tuned ecosystem:

• NVLink & NVSwitch: Instead of running GPUs side by side, they act as a single, gigantic graphics processor in an HGX system. This eliminates bottlenecks in data exchange.
• Convergence of AI & HPC: Whether it's computational fluid dynamics (CFD), genome research or the development of new drugs, HGX accelerates the entire stack from data analysis to final visualisation.
• NVIDIA NGC & software stacks: All systems are optimised for access to the NVIDIA NGC™ catalogue, which provides containerised software, pre-trained models and SDKs for maximum productivity.

Focus areas

1. Generative AI & LLMs: Training and deployment of models such as GPT-4, Llama 3 or specialised enterprise solutions.
2. Autonomous driving & robotics: Processing enormous amounts of sensor data in real time.
3. Digital twins: Simulation of complex physical worlds with NVIDIA Omniverse.