--- 格式版本: 2 标题: "NVIDIA Accelerates AI for Over 80 New Science Systems Worldwide | NVIDIA Blog" 原文链接: "https://blogs.nvidia.com/blog/sc25-new-science-systems-worldwide/" 发布日期: "2026-06-30" 发布时间校准状态: "found" 发布时间来源: "llm:strict_original_body" 发布时间证据: "time class=related-news-date nvidia-article-date datetime=2026-06-30T08:00:57-07:00: Jun 30, 2026" 发布时间校准原因: "该日期来自HTML中的time标签，且class包含nvidia-article-date，位于标题附近，符合文章发布时间的特征。" 发布时间校准置信度: "1" 发布时间候选数量: 32 发布时间严格候选数量: 8 发布时间原页读取状态: "原页面来自已抓取 HTML" 发布时间未找到原因: "候选日期无效或 LLM 未确认" 发布时间校准时间: "2026-07-20T14:13:30+08:00" 发现时间: "2026-07-20T09:26:48+08:00" 入库时间: "2026-07-20T07:06:24.177Z" 来源平台: "NVIDIA Blog 搜索" 搜索渠道: "source_template" 搜索词: "https://blogs.nvidia.com/?s=Hon%20Hai" 匹配关键词: - "GPU" - "NVL72" - "Nvlink" - "Vera Rubin" 相关厂家: - "Hon Hai" - "NVIDIA" - "Oracle" - "Foxconn" 相关专家: [] 内容类型: "网页" 抓取工具: "AgentKey Scrape" 清洗工具: "AgentKey Markdown + LLM 正文裁剪" 原始附件: [] AI优质: "否" AI打分: 68 AI分档: "召回候选" AI质检状态: "不通过" AI打分理由: "NVIDIA官方发布，提及GB200 NVL72/NVL4、Vera Rubin、NVLink Fusion等超节点/机柜级系统，但正文侧重科学计算应用与宏观部署，缺乏机柜架构、供电散热互连等深度技术细节，商业信号偏宏观。" AI质检模型: "qwen3.6-plus" AI质检时间: "2026-07-20T15:06:24+08:00" AI主题相关性: 14 AI来源权威性: 15 AI新颖性: 16 AI技术细节: 10 AI商业部署信号: 10 AI完整性: 3 采集批次: "2026年7月20日9点23分34秒" 采集批次ID: "20260720-092334-062" 去重键: "https://blogs.nvidia.com/blog/sc25-new-science-systems-worldwide" --- Across quantum physics, digital biology and climate research, the world’s researchers are harnessing a universal scientific instrument to chart new frontiers of discovery: accelerated computing. At this week’s [SC25](https://sc25.supercomputing.org/) conference in St. Louis, Missouri, NVIDIA announced that over 80 new scientific systems powered by the NVIDIA accelerated computing platform have been unveiled around the globe in the last year, contributing to a combined total of 4,500 exaflops of AI performance. Newest among them is America’s largest academic supercomputer: the 300-petaflop Horizon system at the Texas Advanced Computing Center (TACC). Slated to be powered by NVIDIA GB200 NVL4 and NVIDIA Vera CPU servers, interconnected with NVIDIA Quantum-X800 InfiniBand networking, Horizon is set to accelerate breakthroughs in science and engineering when it comes online in 2026, offering the nation’s research community unprecedented computing capabilities for discovery and innovation. It’s the latest in a new wave of NVIDIA-accelerated supercomputers fueling global research by nations and private companies in areas such as healthcare, weather and climate modeling, robotics, manufacturing, quantum computing research and materials science. NVIDIA’s full-stack accelerated computing platform — spanning GPUs, CPUs, DPUs, NICs, scale-out switches, as well as CUDA-X libraries and NVIDIA AI Enterprise software — provides the unified architecture, scale and efficiency these systems need to advance science sustainably and at unprecedented speed. ## Scientific Innovation on the Horizon for TACC With 4,000 NVIDIA Blackwell GPUs, the Horizon supercomputer can deliver up to 80 exaflops of AI compute at FP4 precision. It was designed to support a specific set of scientific modeling and simulation applications, including: - **Simulating the mechanics of disease:** Researchers plan to use molecular dynamics software and AI-enhanced simulations to study viruses. - **Modeling stars and galaxies across the universe:** Astrophysicists plan to explore how stars and galaxies form — and simulate distant galaxies uncovered by recent discoveries from the James Webb Space Telescope. - **Investigating novel materials at atomic scale:** Scientists plan to study turbulence, solids with complex crystal structures and the conductivity of quantum materials. - **Mapping seismic waves to prepare for earthquakes:** Researchers plan to improve seismic hazard maps and simulate how faults rupture during earthquakes. “Horizon will enable our scientists to pursue ambitious scientific research at unprecedented scale,” said John Cazes, director of high-performance computing at TACC. “This new system will transform how the research community can pursue AI-driven initiatives to decipher the molecular dynamics of viral infections, explore data from distant galaxies and simulate seismic activity decades into the future.” Horizon: Accelerating Scientific Discovery With AI - YouTube NVIDIA 2.23M subscribers ## Argonne, Los Alamos National Laboratories to House New AI Supercomputers The U.S. Department of Energy (DOE) recently announced a partnership with NVIDIA to build seven new AI supercomputers at Argonne National Laboratory (ANL) in Illinois and Los Alamos National Laboratory (LANL) in New Mexico. At ANL, two [AI supercomputing systems featuring NVIDIA Blackwell GPUs](https://nvidianews.nvidia.com/news/nvidia-oracle-us-department-of-energy-ai-supercomputer-scientific-discovery) and NVIDIA networking will connect with the DOE’s network of scientific instruments and data assets, enabling researchers to develop powerful AI models for science and energy applications. The largest system in the lab complex, Solstice, will feature 100,000 NVIDIA Blackwell GPUs. A system of that scale featuring NVIDIA GB200 NVL72 systems can reach a staggering 1,000 exaflops of AI training compute for training. That’s over 50% higher than the sum of AI training compute across the entire TOP500 list from June 2025, at around 650 exaflops. Another ANL system, called Equinox, will be powered by 10,000 NVIDIA Blackwell GPUs. Three more NVIDIA-accelerated systems at the lab — Minerva, Janus and Tara — will support AI inference and workforce development. At LANL, the [Mission and Vision systems](https://www.energy.gov/nnsa/articles/nnsa-announces-two-new-supercomputers-los-alamos-national-laboratory-propelling) — to be built and delivered by HPE — will be powered by the NVIDIA Vera Rubin platform and the NVIDIA Quantum-X800 InfiniBand networking platform. Mission will run classified applications for the National Nuclear Security Administration, while Vision will power open science research, including foundation models and agentic AI. Both are expected to be operational in 2027. These seven DOE systems follow this year’s announcement with Lawrence Berkeley National Laboratory about [Doudna](https://blogs.nvidia.com/blog/dell-nvidia-berkeley-doudna/) — a supercomputer for scientific discovery set to launch in 2026. Doudna will be powered by the NVIDIA Vera Rubin architecture and NVIDIA Quantum-X800 InfiniBand, and is poised to support the work of over 11,000 researchers across fusion energy, materials science, drug discovery and astronomy. ## Europe’s Jülich Supercomputer Breaks Exaflop Barrier on Linpack Benchmark Across the Atlantic, NVIDIA-accelerated supercomputers in Europe — including systems at the Swiss National Supercomputing Centre and Italy’s CINECA supercomputing center — are fueling scientific research throughout the continent. In Germany, the Jülich Supercomputing Centre’s JUPITER system has achieved [exaflop](https://blogs.nvidia.com/blog/what-is-an-exaflop/) performance — calculating 1 quintillion floating point operations per second — on the HPL benchmark, which measures computing performance on double precision (FP64) math. View into a JUPITER rack with compute blades. Image courtesy of Forschungszentrum Jülich / Sascha Kreklau. [JUPITER, inaugurated in September](https://blogs.nvidia.com/blog/jupiter-exascale-supercomputer-live/), is Europe’s first exascale computer, featuring 24,000 NVIDIA GH200 Grace Hopper Superchips interconnected with NVIDIA Quantum-2 InfiniBand. It’s already in use for applications including high-resolution global climate simulation. “With over 1 exaflop of computing power on JUPITER, our researchers can now run global simulations at kilometer-scale resolution,” said Thomas Lippert, director of the Jülich Supercomputing Centre. “This leap in compute capacity enables European researchers to run AI models and simulations across scientific disciplines at new levels of complexity, size and scale.” Other major European supercomputers unveiled in the past year include: - [Blue Lion](https://blogs.nvidia.com/blog/blue-lion-vera-rubin/) — Slated to go online in early 2027, this system at Germany’s Leibniz Supercomputing Centre, LRZ, will be powered by the NVIDIA Vera Rubin platform to support researchers working on climate, turbulence, physics and machine learning. - [Gefion](https://blogs.nvidia.com/blog/denmark-sovereign-ai-supercomputer/) — Denmark’s first AI supercomputer, operated by DCAI, is an [NVIDIA DGX SuperPOD](https://www.nvidia.com/en-us/data-center/dgx-superpod/) providing sovereign AI capacity for the country’s innovators to advance research in areas including quantum computing, clean energy and biotechnology. - [Isambard-AI](https://blogs.nvidia.com/blog/isambard-ai/) — The U.K.’s most powerful AI supercomputer, housed at the University of Bristol, is being used for projects including Nightingale AI, a multimodal foundation model trained on National Health Service data, and [UK-LLM](https://blogs.nvidia.com/blog/uk-llm-nemotron/), an initiative to enable high-quality AI reasoning for Welsh and other U.K. languages. Gefion is Denmark’s first AI supercomputer, consisting of an NVIDIA DGX SuperPOD interconnected with NVIDIA Quantum-2 InfiniBand networking. Image courtesy of DCAI. ## Supercomputers in Japan, South Korea and Taiwan Fuel Research Across Industries Through [sovereign AI](https://blogs.nvidia.com/blog/what-is-sovereign-ai/) investments and private-sector initiatives, NVIDIA-accelerated AI infrastructure is also supporting scientific research in Japan, South Korea and Taiwan. RIKEN, Japan’s top research institute, announced at SC25 that it is integrating NVIDIA GB200 NVL4 systems in two new supercomputers — a 1600-GPU system for AI for science and a 540-GPU system for quantum computing. RIKEN is integrating NVIDIA Blackwell with two new supercomputers in Japan — one built for AI for science and the other for quantum computing. RIKEN is also working with Fujitsu and NVIDIA to codesign [FugakuNEXT](https://blogs.nvidia.com/blog/fugakunext/) (development code name), a supercomputer that will power earth systems modeling, drug discovery research and advanced manufacturing applications. It’ll feature FUJITSU-MONAKA-X CPUs, which can be paired with NVIDIA technologies using [NVLink Fusion](https://nvidianews.nvidia.com/news/nvidia-nvlink-fusion-semi-custom-ai-infrastructure-partner-ecosystem). Tokyo University of Technology has built an AI supercomputer with [NVIDIA DGX B200](https://www.nvidia.com/en-us/data-center/dgx-b200/) systems capable of achieving 2 exaflops of FP4 theoretical computing performance with under 100 GPUs. The system will be used to develop large language models and build digital twins to serve as core infrastructure for fostering the next generation of AI talent. Japan’s National Institute of Advanced Industrial Science and Technology recently launched [ABCI-Q](https://nvidianews.nvidia.com/news/nvidia-powers-worlds-largest-quantum-research-supercomputer), the world’s largest research supercomputer dedicated to quantum computing, featuring over 2,000 NVIDIA H100 GPUs. NVIDIA powers ABCI-Q, the world’s largest research supercomputer dedicated to quantum computing. Image courtesy of AIST G-QuAT. The South Korean government plans to deploy [over 50,000 NVIDIA GPUs across sovereign clouds and AI factories](https://nvidianews.nvidia.com/news/south-korea-ai-infrastructure). Industry leaders [Samsung](https://nvidianews.nvidia.com/news/samsung-ai-factory), [SK Group](https://nvidianews.nvidia.com/news/sk-group-ai-factory) and [Hyundai Motor Group](https://nvidianews.nvidia.com/news/hyundai-motor-group-ai-factory) are also building AI factories with NVIDIA Blackwell GPUs to accelerate research and manufacturing. And in Taiwan, NVIDIA is working with [Foxconn Hon Hai Technology group](https://nvidianews.nvidia.com/news/foxconn-ai-factory-tsmc-taiwan-nvidia) to build an AI factory supercomputer with 10,000 NVIDIA Blackwell GPUs to fuel innovation across researchers, startups and industries. *Read more about* [*NVIDIA-accelerated supercomputing*](https://www.nvidia.com/en-us/high-performance-computing/) *and watch the* [*SC25 fireside chat by Ian Buck*](https://nvevents.nvidia.com/sc25fireside)*, vice president of hyperscale and high-performance computing at NVIDIA.* ![Why Performance per Watt Is the Ultimate Metric for AI Infrastructure Efficiency](./assets/img-83d2f103.jpg) ![AI Innovators Adopt NVIDIA Vera — Why Max Single-Threaded CPU at Scale Matters](./assets/img-1f6575f1.png) ![NVIDIA Unlocks AI Compute at Scale, Inviting Partners to Power the AI Infrastructure Buildout](./assets/img-3511966c.jpg) ![How NVIDIA’s Inference Software Stack Powers the Lowest Token Cost](./assets/img-e47e99dc.png)