At the OCP Global Summit, NVIDIA announced its contribution of key components from its NVIDIA Blackwell accelerated computing platform to the Open Compute Project (OCP) and expanded support for OCP standards with its Spectrum-X™ networking platform. NVIDIA will share vital designs of the GB200 NVL72 system, including rack architecture, cooling specifications, and NVLink cable details, aimed at boosting compute density and bandwidth. NVIDIA's previous contributions to OCP include the HGX H100 baseboard design, which enhances the ecosystem's options for AI technology. The enhanced Spectrum-X alignment with OCP standards allows companies to optimize AI workloads while protecting their investments. NVIDIA CEO Jensen Huang emphasized the company's decade-long collaboration with OCP to develop widely adopted data center specifications, aiming to maximize accelerated computing's potential and foster AI factory innovations. The GB200 NVL72 system leverages NVIDIA's MGX™ modular architecture to enable flexible data center infrastructure. It features liquid cooling and integrates 36 Grace™ CPUs and 72 Blackwell GPUs, achieving 30 times faster real-time inference than the H100 GPU.

The next-generation ConnectX-8 SuperNIC™ supports OCP’s standards, offering up to 800Gb/s speeds and enhancing performance for large-scale AI operations. As the demand for AI computing grows, NVIDIA partners with over 40 global electronics manufacturers to simplify AI factory development, with contributions from companies like Meta to further enhance hardware design options. NVIDIA will showcase its advancements and engage with the OCP community at the 2024 OCP Global Summit, set for October 15-17 in San Jose. NVIDIA continues to lead in accelerated computing while managing risks associated with production and market demand. Its forward-looking statements highlight the dynamic nature of technology development and market conditions. For additional information on NVIDIA's contributions and technological advancements, visit NVIDIA's website.