Gordon Bell Prize Finalists Push Open Science Boundaries With NVIDIA-Powered Supercomputers

5 finalists for the Gordon Bell Prize for excellent achievements in high-performance computing (HPC) are utilizing NVIDIA-powered supercomputers for his or her vital work in local weather modeling, supplies science, fluid simulation, geophysics and digital design.

Introduced as we speak at SC25, the finalists’ initiatives are driving AI and HPC for science utilizing physics simulation, high-precision math and different superior supercomputing methods, accelerating breakthroughs throughout climate forecasting, semiconductor design, house exploration and different fields. Their outcomes are open and accessible on ArXiv.

The supercomputers powering their work embody:

• Alps — hosted on the Swiss Nationwide Supercomputing Centre (CSCS) and powered by greater than 10,000 NVIDIA GH200 Grace Hopper Superchips. 
• Perlmutter — hosted on the Nationwide Power Analysis Scientific Computing Heart (NERSC) and powered by NVIDIA accelerated computing.
• JUPITER — Europe’s first exascale supercomputer, hosted on the Jülich Supercomputing Centre (JSC) and powered by the NVIDIA Grace Hopper platform and Quantum-X800 InfiniBand networking.

“At CSCS, we don’t simply help open science — we speed up it,” stated Thomas Schulthess, director of CSCS. “The extraordinary breakthroughs by this 12 months’s 5 Gordon Bell finalists in local weather modeling, supplies science, fluid dynamics and digital twins stand as irrefutable proof: with out the Alps supercomputer, these scientific discoveries merely wouldn’t exist. Pushing computational boundaries turns daring targets into actuality, delivering scientific revolutions that can redefine our world.”

Be taught extra concerning the 5 finalists’ initiatives beneath.

A novel configuration for the ICON Earth system mannequin — developed by researchers on the Max Planck Institute for Meteorology, German Local weather Computing Centre (DKRZ), CSCS, JSC, ETH Zurich and NVIDIA — is poised to allow extra correct climate forecasts and a deeper understanding of how the planet works. 

By modeling the complete Earth’s techniques at kilometer-scale decision, ICON can seize the circulation of vitality, water and carbon by way of the ambiance, oceans and land with distinctive element and unprecedented temporal compression — permitting about 146 days to be simulated each 24 hours — which allows extra environment friendly local weather simulations projecting as much as many years ahead.

“Integrating all important elements of the Earth system within the ICON mannequin at an unprecedented decision of 1 kilometer permits researchers to see full world Earth system info on native scales and be taught extra concerning the implications of future warming for each individuals and ecosystems,” stated Daniel Klocke, computational infrastructure and mannequin improvement group chief at Max Planck Institute for Meteorology.

Developed as a part of a collaboration between Oak Ridge Nationwide Laboratory, NVIDIA and others — and operating on the Alps supercomputer — ORBIT-2 is an AI basis mannequin for climate and local weather downscaling that demonstrates unparalleled scalability and precision.

Tapping into exascale computing and algorithmic innovation, ORBIT-2 overcomes challenges confronted by conventional local weather fashions with spatial hyper-resolution downscaling, a method that creates high-resolution information from lower-resolution sources. This allows groups to seize and predict much more localized phenomena like city warmth islands, excessive precipitation occasions and delicate shifts in monsoon patterns.

“NVIDIA’s superior supercomputing applied sciences enabled ORBIT-2 to attain distinctive scalability, reliability and influence on the intersection of AI and high-performance computing on NVIDIA platforms,” stated Prasanna Balaprakash, director of AI packages and part head for information and AI techniques at Oak Ridge Nationwide Laboratory.

A group from ETH Zurich has superior nanoscale digital gadget modeling with QuaTrEx, a bundle of algorithms that may increase the design of next-generation transistors.

Operating on the Alps supercomputer with NVIDIA GH200 Superchips, QuaTrEx can simulate units with greater than 45,000 atoms with FP64 efficiency and excessive parallel-computing effectivity. This allows sooner, extra correct design of transistors, known as NREFTs, that might be essential for the semiconductor trade.

“Entry to Alps was instrumental within the improvement of QuaTrEx,” stated Mathieu Luisier, full professor of computational nanoelectronics at ETH Zurich. “It allowed us to simulate units that we couldn’t think about dealing with just some months in the past.”

Designing spacecrafts, particularly these with many small engines, requires detailed simulation, as engines packed carefully collectively may cause their exhaust to work together and warmth up a rocket’s base. 

Operating on the Alps supercomputer, MFC, an open-source solver developed by the Georgia Institute of Expertise in collaboration with NVIDIA and others, allows fluid circulation simulation 4x sooner and with over 5x higher vitality effectivity whereas sustaining the identical accuracy because the earlier world report. Primarily based on full-scale runs on Alps, MFC is anticipated to run at 10x the size of the earlier world report on JUPITER. This paves the best way for sooner, extra correct design of vital elements for house exploration.

“Our new info geometric regularization technique, mixed with the NVIDIA GH200 Superchip’s unified digital reminiscence and mixed-precision capabilities, has drastically improved the effectivity of simulating complicated computational fluid flows, enabling us to simulate rocket engine plumes at unprecedented scales,” stated Spencer Bryngelson, assistant professor in computational science and engineering on the Georgia Institute of Expertise.

The College of Texas at Austin, Lawrence Livermore Nationwide Laboratory and the College of California San Diego have created the world’s first digital twin that may problem real-time probabilistic tsunami forecasts based mostly on a full-physics mannequin. 

Utilized to the Cascadia subduction zone within the Pacific Northwest, the digital twin completed complicated computations that may usually take 50 years on 512 GPUs in simply 0.2 seconds on the Alps and Perlmutter supercomputers, representing a ten billion-fold speedup.

“For the primary time, real-time sensor information might be quickly mixed with full-physics modeling and uncertainty quantification to offer individuals an opportunity to behave earlier than catastrophe strikes,” stated Omar Ghattas, professor of mechanical engineering at UT Austin. “This framework gives a foundation for predictive, physics-based emergency-response techniques throughout varied hazards.”

For the tsunami digital twin, ICON and MFC initiatives, NVIDIA CUDA-X libraries performed a key position in maximizing the efficiency and effectivity of the complicated simulations. ICON additionally faucets into NVIDIA CUDA Graphs, which permit work to be outlined as graphs somewhat than single operations.

Be taught extra concerning the newest supercomputing developments by becoming a member of NVIDIA at SC25, operating by way of Thursday, Nov. 20.