HaiAI123

Curated Global AI Tools Directory

Industry update
Flat

NVIDIA unveils Vera Rubin platform at SC26 with MLPerf results

11/17/2026 — 11/17, 02:18·2 sources (all)·5 reports (all)

Key takeaways

  • At SC26, NVIDIA unveiled its Vera Rubin platform, including the Rubin HPC GPU with over 50,000 CUDA cores and 7th-gen NVLink, and debuted on…

Background & analysis

On November 16, 2026, NVIDIA introduced Rubin HPC, a new GPU within the Vera Rubin platform, at SC26, the supercomputing conference held in Chicago. The chip targets scientific computing, engineering simulation, and data analytics workloads. Each GPU carries more than 50,000 CUDA cores, offers sparse-data acceleration, and uses seventh-generation NVLink interconnect. Alongside the launch, NVIDIA outlined progress on the CUDA-X software libraries, photonic computing, and quantum computing, and disclosed that the Vera CPU rack will reach market early next year.

On November 17, 2026, NVIDIA released its first MLPerf results for the platform. In MLPerf Training v6.1, the Vera Rubin NVL72 posted its debut submission and recorded the fastest training time on six of the seven benchmarks, reaching up to 2x the performance of a highly optimized GB300 NVL72 baseline. For the remaining benchmark, DeepSeek-V3 671B, the round's fastest result came from a GB300 cluster of 10240 GPUs. NVIDIA stated that Vera Rubin retains more than 97% overall efficiency when scaling from 256 to 1024 GPUs.

In its first MLPerf Endpoints v1.0 submission, the platform delivered more than 10x the throughput per watt of GB300 NVL72 under an end-to-end interactivity setting of 50 tokens/s per user. That benchmark uses coding and enterprise workflows through API endpoints to balance serving capacity with responsiveness, and separate SWE-bench Verified evaluations assess end-to-end task-solving ability. NVIDIA's Developer Blog credits the platform with the fastest time on six training benchmarks, while NVIDIA Blog says it swept every MLPerf Training v6.1 item, leaving a discrepancy between the two accounts. NVIDIA added that performance will continue to improve as the CUDA software stack is further optimized, with results and disclosures so far limited to this first round.

AI-generated from 5 reports · updated 6 hours ago

Latest turnNvidia unveiled Rubin HPC, a new GPU designed for scientific computing, engineering simulation and data analysis, at the SC26 supercomputing conference, alongside updates to CUDA-X libraries, photonic computing and quantum computing. The company aims to use hardware-software co-design to let AI agents take part in scientific discovery, from forming hypotheses to validating results.

updream
24-hour heatHeat index 143 · peak 194 · 10h ago
24 hours agonow

Reports on this story headlines open the original

Today
  1. Nvidia unveiled Rubin HPC, a new GPU designed for scientific computing, engineering simulation and data analysis, at the SC26 supercomputing conference, alongside updates to CUDA-X libraries, photonic computing and quantum computing. The company aims to use hardware-software co-design to let AI agents take part in scientific discovery, from forming hypotheses to validating results.

    华尔街见闻AI score 57
  2. NVIDIA's Vera Rubin platform made its MLPerf debut, sweeping every MLPerf Training v6.1 benchmark and, in its first MLPerf Endpoints submission, delivering more than 10x generational gains in throughput per watt for agentic AI. NVIDIA says performance will keep climbing through continued optimization of its CUDA software stack.

    NVIDIA BlogFirst-partyAI score 50
  3. At SC26, NVIDIA announced Rubin HPC, a new GPU in the Vera Rubin platform aimed at simulation and data processing for science, engineering and analytics. It packs more than 50,000 CUDA cores on a single GPU, adds sparse data acceleration and uses seventh-generation NVLink. NVIDIA says it shortens the path from exploration to scientific discovery and industrial design.

    NVIDIA BlogFirst-partyAI score 51
  4. In an NVIDIA developer post, provisional Vera Rubin NVL72 results on MLPerf Endpoints v1.0 show over 10x the throughput per kilowatt of provisioned power of GB300 NVL72, at a shared minimum end-to-end interactivity of 50 tokens per second per user. The benchmark measures serving capacity against responsiveness using coding and enterprise workflows through API endpoints.

    NVIDIA Developer BlogFirst-partyAI score 48
  5. NVIDIA's Vera Rubin NVL72 made its MLPerf Training v6.1 debut, posting the fastest time to train on six of seven benchmarks at up to 2x the performance of a highly optimized GB300 NVL72 baseline. The seventh, DeepSeek-V3 671B, went to a 10,240-GPU GB300 cluster. NVIDIA says Vera Rubin held above 97% overall efficiency when scaling from 256 to 1,024 GPUs.

    NVIDIA Developer BlogFirst-partyAI score 44

Other stories people are talking about

How is heat calculated?About the method

Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.

This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.

Surge
Discussion rising fast
New
First report within 6 hours
Rising
Still gathering discussion

Back to AI industry updates →