NVIDIA has introduced the H200 NVL PCIe GPU at the Supercomputing 2024 (SC24) conference, expanding its Hopper architecture portfolio. Designed for air-cooled enterprise racks under 20kW, the H200 NVL offers flexible configurations and enhanced performance for artificial intelligence (AI) and high-performance computing (HPC) workloads. With a 1.5x increase in memory and a 1.2x boost in bandwidth compared to the H100 NVL, the H200 NVL accelerates large language model (LLM) fine-tuning and inference while reducing energy consumption.
The H200 NVL incorporates NVIDIA’s advanced NVLink technology, enabling GPU-to-GPU communication up to seven times faster than PCIe Gen 5. This capability supports demanding applications like LLM inference and HPC tasks. Additionally, the GPU is bundled with a five-year subscription to NVIDIA AI Enterprise, a cloud-native software platform optimized for AI development and deployment. NVIDIA’s Spectrum-X Ethernet platform ensures the system’s networking is tailored for peak performance.
Global hardware partners, including Dell Technologies, HPE, Lenovo, and Supermicro, are set to deliver H200 NVL-enabled platforms starting in December. Modular systems leveraging NVIDIA MGX architecture will further streamline deployment for various data center configurations. NVIDIA is also releasing an Enterprise Reference Architecture to help customers and partners build high-performance AI infrastructure using H200 NVL technology.
Product Highlights:
• 1.5x memory and 1.2x bandwidth increases over H100 NVL.
• Up to 1.7x faster AI inference and 2.5x improved HPC performance over the Ampere generation.
• NVLink technology delivers 7x faster GPU-to-GPU communication compared to PCIe Gen 5.
• Includes NVIDIA AI Enterprise software with five-year subscription.
• Optimized networking with NVIDIA Spectrum-X Ethernet platform.
• Availability:
• Systems from Dell, HPE, Lenovo, and Supermicro launch in December.
• Additional platforms from Aivres, ASUS, GIGABYTE, MSI, and others.
