Information Highlights:
- NVIDIA know-how runs 81% of the TOP500 and 90% of the techniques new to the record.
- 26 techniques on the TOP500 adopted the NVIDIA Grace CPU, up eight from the earlier record.
- The highest eight techniques on the Green500 run on NVIDIA GPUs and 9 of the prime 10 use NVIDIA applied sciences.
- No. 1 on the Green500, KAIROS, makes use of a single NVIDIA Grace Hopper Superchip.
- 376 of the TOP500 techniques are interconnected utilizing NVIDIA networking.
NVIDIA applied sciences energy greater than 400 of the world’s 500 quickest supercomputers — 81% of the TOP500 — in line with the newest rankings launched this week at the ISC Excessive Efficiency convention in Hamburg, Germany.
That’s a achieve of 17 techniques from the earlier record, with the momentum in new deployments: almost 9 of each 10 techniques new to the rating are constructed on NVIDIA applied sciences.
That proportion displays a deliberate desire for machines constructed for AI, simulation and science collectively. And it’s compounding: NVIDIA techniques throughout the TOP500 now ship greater than 2x the AI coaching and almost 3x the AI inference throughput of each different platform mixed.
GPU and networking adoption every hit new highs, with NVIDIA GPUs accelerating a report 238 techniques and NVIDIA networking connecting a report 376 — the overwhelming majority on NVIDIA Quantum InfiniBand, the spine of large-scale AI and high-performance computing, and the relaxation on Ethernet.
The pattern behind the numbers is greater than anybody record: Accelerated computing is changing into the basis for the techniques taking over the world’s most demanding work, throughout AI and science.
Up to date twice a 12 months, the TOP500 ranks the world’s quickest supercomputers, whereas the Green500 record measures how a lot computing every delivers per watt.
A Full-Stack Footprint
NVIDIA’s attain now spans the full system — GPU, networking and, more and more, the CPU — with NVIDIA Grace CPU adoption reaching 26 techniques, up eight from the earlier record, with almost 2.5 million Grace CPUs shipped.
NVIDIA Grace-based machines sit atop each rankings: JUPITER at No. 5 and Alps at No. 10 on the TOP500, and KAIROS at No. 1 on the Green500.
Every pairs an NVIDIA GPU with the Grace CPU in a single NVIDIA Grace Hopper Superchip, letting the two share reminiscence with minimal overhead — a design constructed for the memory-intensive calls for of trendy AI.
The NVIDIA Vera CPU, introduced earlier this 12 months, builds on the success of Grace, taking CPU efficiency and power effectivity to new ranges for the most demanding AI workloads in trendy knowledge facilities — the place brokers transfer from answering fundamental inquiries to taking actions, operating code, utilizing instruments and evaluating outcomes.
Topping the Effectivity Listing
NVIDIA swept the Green500 rating of the most energy-efficient supercomputers: The highest eight all run on NVIDIA GPUs and 9 of the prime 10 use NVIDIA applied sciences.
Main the record is KAIROS, an NVIDIA Grace Hopper system at France’s College of Toulouse, at 73.3 gigaflops per watt — with Grace Hopper techniques taking the prime 4 spots, throughout France, Germany and the U.Ok.
From Exascale Science to the Subsequent Wave
A report 35 NVIDIA AI HPC supercomputers are in improvement throughout Europe — equipping greater than 3 million researchers with next-generation infrastructure for continental AI, accelerated science and industrial innovation.
Amongst these techniques is JUPITER, Europe’s quickest supercomputer and its first to achieve exascale, at the Jülich Supercomputing Centre in Germany.
JUPITER is mapping the human mind at mobile scale, simulating Earth’s local weather and advancing the AI behind next-generation 6G networks.
The latest arrivals to the record run on the NVIDIA Blackwell structure, with B200 and GB200 techniques coming into the rankings throughout Asia, Europe and the U.S. — and the first GB200 techniques debuting in Japan.
The buildout is world, from a brand new AI manufacturing unit in South Africa to nationwide AI techniques in Saudi Arabia, Singapore and Vietnam.
It’s the identical story up and down the record: the world’s AI buildout is operating on NVIDIA.
Source link
#NVIDIA #Powers #Worlds #Fastest #Supercomputers
