Stories

  1. Image: NVIDIA Newsroom

    Nvidia unveils Rubin HPC GPU with more than 50,000 CUDA cores at SC26

    Nvidia announced Rubin HPC at the SC26 supercomputing conference, a new GPU in its Vera Rubin platform aimed at simulation and data processing for agent-driven science, engineering and analytics work. The company says it has more than 50,000 CUDA cores, its highest count on any GPU, compared with 28,672 on the current Rubin GPU. Rubin HPC GPUs will be linked using NVLink 7. Nvidia also introduced Nemotron for Science, pairing its open model with a new open chemistry dataset.

  2. Image: NVIDIA Newsroom

    NVIDIA unveils Rubin HPC GPU with over 50,000 CUDA cores and Vera CPU rack platform at SC26

    NVIDIA announced Rubin HPC at SC26, a new GPU in its Vera Rubin platform aimed at simulation and data processing for science, engineering and analytics. The chip puts more than 50,000 CUDA cores on a single GPU, uses seventh-generation NVLink, and is expected in 2028; NVIDIA says it triples simulation and analytics performance. The company also announced a Vera CPU rack-scale HPC platform, with a 192-CPU rack due in early 2027, alongside Nemotron for Science models, new CUDA-X libraries and a quantum codesign platform with IBM and Fermilab.

  3. Image: NVIDIA Newsroom

    NVIDIA Vera Rubin debuts in MLPerf Training v6.1 and Endpoints with leading results

    NVIDIA's Vera Rubin platform made its MLPerf debut, with the company saying it swept every MLPerf Training v6.1 benchmark. The Vera Rubin NVL72 system delivered up to 2x the performance of Blackwell Ultra in training and the fastest time-to-train in its first appearance. In its first MLPerf Endpoints submission, NVIDIA reported more than 10x generational gains in throughput per watt for agentic AI. NVIDIA says the results show the CUDA software stack for the new architecture is already mature.

  4. Image: NVIDIA Developer Blog

    NVIDIA releases CUDA 13.5 with production support for Vera Rubin

    NVIDIA released CUDA 13.5, bringing general availability support for its Vera Rubin platform. The release moves Vera Rubin from early developer access to production availability, according to NVIDIA’s announcement. The release also coincided with NVIDIA’s 620.70.06 Linux driver release.

  5. Image: NVIDIA Newsroom

    NVIDIA and Meta to Expand Collaboration on Muse and Muse Family of Models Amid Robust Growth

    Nvidia and Meta are expanding their partnership on Muse and the Muse family of models, with a focus on improving inference and training performance. Meta will deploy more Vera Rubin systems and Vera CPUs, while Nvidia plans to invest $3 billion in Meta’s planned equity fundraise. The investment comes as Meta seeks to raise as much as $95 billion through stock offerings.

  6. Image: NVIDIA Newsroom

    NVIDIA unveils DSX POD reference design targeting 14-day AI factory deployment

    NVIDIA introduced DSX POD, a factory-built containerized infrastructure design for AI factories, at GTC Berlin. The design combines Vera Rubin compute with networking, cabling, rack-level power and liquid cooling, with systems integrated and tested before shipment. NVIDIA says deployment at a prepared site could take as little as 14 days, compared with six months; that estimate excludes construction of the supporting facility. Partner-built systems are targeted for delivery worldwide in early 2027.

  7. Image: NVIDIA Newsroom

    Nvidia unveils containerized DSX POD systems to speed Vera Rubin data center deployments

    Nvidia introduced DSX POD, a reference design for modular containers pre-built with Vera Rubin AI hardware, networking and cooling. Each pod is described as holding seven Vera Rubin NVL72 racks with 504 GPUs, and is integrated and validated before delivery. Nvidia says a prepared site could begin serving tokens in as few as 14 days, compared with months for conventional rack installation; power, cooling and network connections must already be in place. The company says partners will offer the systems starting in 2027.

  8. Image: The Register

    Meta details Prometheus supercluster used to train its Muse AI models

    Meta published details about Prometheus, its New Albany, Ohio, supercluster used to train the Muse family of AI models. The system combines Nvidia Grace Blackwell-based clusters, with a Vera Rubin-based cluster expected to join; Meta says Prometheus will eventually scale to a gigawatt of capacity and about a million accelerators. Meta also described measures to keep training running through hardware failures, including a pool of redundant GPUs: more than 11 percent of the GPUs are dead or idle as replacements, one account says.

  9. Image: NVIDIA Newsroom

    NVIDIA opens Vera Rubin architecture and announces AI infrastructure partnerships

    NVIDIA announced new partnerships and open-design contributions at the Open Compute Project Global Summit, aiming to help scale AI infrastructure. The company said its efforts cover power, cooling and networking, and include opening its Vera Rubin architecture. The announcements come as AI factories expand from megawatts to gigawatts, increasing the need for infrastructure that can scale efficiently and reliably.

Topics

Publishers