{"id":16924,"date":"2025-06-17T16:19:52","date_gmt":"2025-06-17T15:19:52","guid":{"rendered":"https:\/\/ibertronica.es\/blog\/?p=16924"},"modified":"2025-06-17T16:19:52","modified_gmt":"2025-06-17T15:19:52","slug":"nvidia-hgx-open-platform-powering-large-scale-ai-and-hpc","status":"publish","type":"post","link":"https:\/\/ibertronica.es\/blog\/en\/news-en\/nvidia-hgx-open-platform-powering-large-scale-ai-and-hpc\/","title":{"rendered":"NVIDIA HGX: Open Platform Powering Large-Scale AI and HPC"},"content":{"rendered":"

The NVIDIA HGX platform has established itself as the de facto standard for training and deploying large-scale artificial intelligence (AI) models. With a combination of Hopper (H100\/H200) and, starting in 2024, Blackwell (B200\/B300) GPUs, a 1.8 TB\/s NVLink 5 interconnect, 800 Gb\/s InfiniBand and Ethernet networks, BlueField-3 DPUs, and an open, community-defined OCP format, HGX delivers the performance, scalability, and efficiency required by generative models with trillions of parameters.<\/p>\n

The Computational Challenge of Modern AI<\/h2>\n

Language and vision models now have hundreds of billions of parameters, requiring infrastructure capable of moving large amounts of data without bottlenecks. With HGX, NVIDIA brings together everything needed for data centers to scale from a few to thousands of GPUs working together into a single \u00abbuilding block.\u00bb<\/p>\n

How is an NVIDIA HGX node organized?<\/h2>\n

An HGX node is like a super module<\/strong> that sits inside a rack and brings together, in a single chassis, everything needed to train and serve AI models without bottlenecks. Its architecture is divided into five distinct blocks:<\/p>\n

    \n
  1. GPU Group<\/strong>
    \n\u2022 4 to 8 Hopper (H100\/H200) or Blackwell (B200\/B300) GPUs mounted on an SXM socket.
    \n\u2022 Each GPU incorporates HBM memory (up to 141GB per chip on the H200) to support massive datasets.<\/li>\n
  2. NVLink5 Internal Network + NVSwitch<\/strong>
    \n\u2022 Up to 1.8TB\/s of aggregate bandwidth between GPUs, double the previous version.
    \n\u2022 Allows all eight chips to share memory and act as a single logical GPU.<\/li>\n
  3. 800Gb\/s External Network<\/strong>
    \n\u2022 ConnectX-8 cards and Quantum-X800 (InfiniBand) or Spectrum-X (Ethernet) switches.
    \n\u2022 Connect multiple HGX nodes to form clusters of thousands of GPUs with latencies below 2\u00b5s.<\/li>\n
  4. DPUBlueField-3<\/strong>
    \n\u2022 Accelerates zero-trust networking, storage, and security operations without stealing cycles from the GPUs. \u2022 According to NVIDIA, it frees up to 30% of compute capacity that would otherwise be occupied by I\/O tasks.<\/li>\n
  5. Thermal and Power Management<\/strong>
    \n\u2022 Maximum consumption of 6.8kW per node, direct-to-chip liquid cooling, and 54V power backed by the Open Compute Project (OCP).<\/li>\n<\/ol>\n

    Because the HGX form factor is published in OCP, any manufacturer can add x86 or Arm CPUs, additional storage, or custom interconnects, while maintaining full compatibility with CUDA and the NVIDIA AI Enterprise suite.<\/p>\n

    \"Nvidia<\/a><\/p>\n

    HGX solutions available today<\/h2>\n