Opening The Future of Science

A new era of U.S. scientific computing is underway. 

Horizon, at the Texas Advanced Computing Center (TACC) at The University of Texas at Austin, will soon begin early operations broadly for the research community. As the flagship supercomputer of the U.S. National Science Foundation Leadership-Class Computing Facility (NSF LCCF), Horizon will provide unprecedented computing power to researchers tackling some of the nation's most ambitious scientific challenges.

Last month, more than 200 members of the national open science community gathered virtually to explore how leadership-class computing is driving the next generation of scientific discovery through the NSF LCCF and its Characteristic Science Applications (CSA) program.

The event featured remarks from NSF and TACC leaders, followed by a moderated panel of researchers using TACC's Frontera and Vista supercomputers to tackle some of science's most complex challenges. Panelists shared how the CSA program is accelerating breakthroughs in astronomy, earthquake science, quantum materials, lattice quantum chromodynamics (QCD), and biophysics, demonstrating the transformative impact of advanced computing on research across disciplines.

“This is an investment of over $400 million and a commitment by the NSF and the nation to support the computational science and AI-driven workloads that will spark the innovations and discoveries of the future,” said Katie Antypas, senior adviser for cyberinfrastructure for the NSF. "This commitment is a long-term investment to match the science missions TACC supports. At the highest level, they have included several Nobel Prize-winning research projects in computational protein design, the detection of gravitational waves, and confirmation of the Higgs boson of particle physics," she added.

For nearly two decades, TACC has defined the leading edge of U.S. academic supercomputing, from the NSF-funded Ranger system to the Stampede family of supercomputers, Frontera, and the AI-focused Vista — each expanding the nation's computational capabilities and paving the way for Horizon in 2026. 

Since 2001, more than 125,000 academic researchers and students have relied on TACC resources to drive discoveries with global scientific and societal impact.

Leadership-Class Computing Arrives

The first LCCF supercomputer, Horizon, has entered the first of two phases in its deployment — its graphical processing unit (GPU) computing racks are operational, totaling 4,000 NVIDIA NVL4 Blackwell GPUs and 2,000 NVIDIA Grace GB200 CPUs interconnected with 800 GB/sec Infiniband networking and 400 petaflops of solid-state storage. 

“The basic premise of the Leadership-Class Computing Facility is to build new and more capable systems to replace systems in the current NSF fleet,” said Dan Stanzione, TACC’s executive director and associate vice president for research at The University of Texas at Austin. “The LCCF has made a long-term investment and commitment to having computing for the scientists who rely on us for their projects and large instrument data.” 

The science applications that run on Horizon will be 15-20 times as fast as on Frontera, with AI-based applications sped up to 100x or more. The CPU part of Horizon is expected to come online in Winter 2026 or early 2027, with 8,000 NVIDIA Vera CPUs. The Ranch data archive system has been in production since 2025, with an exabyte of capacity. 

Characteristic Science Applications Drive Design and Innovation

Helping shape Horizon from the ground up is the Characteristic Science Applications (CSA) program — 11 cutting-edge software projects developed through yearslong collaborations with research teams to prepare the system for some of the most demanding scientific challenges of the coming decade.

These scientists are our co-design partners to make sure we pick the right mix of things to put into the machine, and they're some of our early access users.

“The CSAs are our acceptance tests to make sure Horizon performs," Stanzione said. "These scientists are our co-design partners to make sure we pick the right mix of things to put into the machine, and they're some of our early access users." The CSA teams have been testing running their code on Vista and most recently on 72 Blackwell GPUs TACC added to Vista, to test pre-Horizon nodes.

Five CSA teams on the virtual panel highlighted early results from projects across a diverse range of disciplines, including particle physics, astrophysics, advanced materials, biochemistry related to neurological disorders, and earthquake modeling.

Watch the virtual panel now:

Science Snapshots from the Panelists

Galaxy Evolution at Scale

Tom Quinn (Moderator)
Professor of Astronomy, University of Washington

Quinn's group uses particle-based simulations to investigate galaxy formation across cosmic scales, from the emergence of planets to galaxy clusters and the web-like large-scale structure of the universe.

Unexpected observations from the NASA James Webb Space Telescope (JWST) — including a surprisingly rich population of galaxies formed shortly after the Big Bang — have upended existing models and are fueling the group's efforts to better understand how the early universe evolved.

“On Frontera, we have been able to do simulations of Milky Way-sized galaxies. As we move to larger, more capable facilities such as Horizon, we're moving to clusters of galaxies,” Quinn said.

What’s more, although the JWST can gather images of a galaxy from the early universe, that information can't tell us how the galaxy will evolve — and whether it is a progenitor of a Milky Way-like galaxy or perhaps a seed for the formation of a large cluster of galaxies.

“It's only with the simulations that we can follow a galaxy over the full history of the universe and compare images from JWST to, for example, observations by the Vera Rubin telescope of galaxies we see today."

Quantum Materials

Feliciano Giustino (Panelist)
Professor of Physics and W. A. "Tex" Moncrief, Jr. Chair in Quantum Materials Engineering, Oden Institute for Computational Engineering and Sciences, The University of Texas at Austin

Giustino leads the Center for Quantum Materials Engineering, which develops predictive simulations to design advanced materials for semiconductors, solar energy, and quantum technologies.

For the past three years, his group has been porting the EPW code from CPUs to GPUs, a major software engineering effort that enables faster, larger-scale simulations of electron-phonon interactions critical to understanding superconductivity and other quantum phenomena.

"We demonstrated a 30-fold speed up of this code on the GPU accelerators of Vista,” Giustino said. “Also exciting is the performance portability achieved on the Aurora supercomputer of the Argonne National Lab, where we managed to scale the code up to 6,000 GPUs. This gives us confidence that we'll be able to utilize Horizon to its fullest.”

At an NVIDIA Hackathon hosted by TACC in May 2026, the group's EPW team achieved an 80-fold acceleration in optical absorption spectra calculations on Horizon test nodes, demonstrating the performance gains possible on next-generation GPU-based systems.

Netting New Particles and Forces 

Steven A. Gottlieb (Panelist)
Distinguished Professor Emeritus; Provost Professor Emeritus of Physics, Indiana University Bloomington

Gottlieb’s group works with the MILC (MIMD Lattice Computation) collaboration and Fermilab to search for evidence of physics beyond the Standard Model, the theory describing the fundamental particles and the electromagnetic, strong, and weak forces that govern their interactions. 

Using lattice quantum chromodynamics (QCD) to model the strong force with unprecedented precision, the team produces calculations that help interpret results from major particle physics experiments, including the Large Hadron Collider, Fermilab’s Muon g-2 experiment, KEK, and BESIII.

His group models QCD strong interactions between quarks and gluons using a 4D space-time grid, which is the main source of systematic errors. 

“We need to take a range of lattice spacing to the continuum limit, a computational heavy job, which is why we want to use Horizon,” Gottlieb said. “We project that Horizon is so powerful that this much tougher problem of a finer grid resolution will only need six months to generate our targeted 1,000 configurations."

Dynamic Evolution of Biomolecular Simulations

David Hardy
Senior Research Programmer, Theoretical and Computational Biophysics Group, Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign 

Hardy’s group and the Computational Biophysics Group develop NAMD, the widely used molecular dynamics software behind some of the world’s largest biomolecular simulations. Written in C++ and built on the CHARM++ parallel programming model, the Gordon Bell Award-winning code enables researchers to simulate complex biological systems efficiently on the world’s most powerful supercomputers.

In their latest work, the team simulated a billion-atom protocell of the Satellite Tobacco Mosaic Virus, a model system for understanding how viruses assemble and remain stable. They also optimized the GPU version of NAMD on TACC’s AI-focused Vista supercomputer, improving performance and scalability on NVIDIA Grace CPU and Blackwell GPU architectures, the same technology that will power Horizon.

“We were seeing really good results scaling up to 512 nodes of Vista, and we're gratified to see that,” Hardy said.

He also highlighted recent simulations of glutamate transportation regulated by the synaptic gap that occurs when neurons are talking to each other, where dysregulation is associated with various neurological disorders. This work is in collaboration with the lab of fellow University of Illinois colleague Emad Tajkhorshid, a developer and principal investigator of the NAMD project.

“They've run this now on Vista, where they can run our very fast GPU resident code for NAMD, and they're getting simulation rates of about 90 nanoseconds per day. This is scaling quite nicely, and they’re looking forward to being able to use Horizon so they can scale this out to one system per GPU or one simulation per GPU,” Hardy added.

Shaking Up Seismic Hazard Analysis Simulations

Yifeng Cui (Panelist)
Director, High Performance GeoComputing Laboratory, San Diego Supercomputer Center; Principal Investigator, Statewide California Earthquake Center

Cui traced the evolution of earthquake simulations, from the groundbreaking TeraShake project in 2004 to the ShakeOut scenarios run at TACC that serve as the scientific foundation for earthquake preparedness drills involving tens of millions of people worldwide.

Today, his group is advancing the state of the art with CyberShake, a physics-based probabilistic seismic hazard analysis (PSHA) framework that models earthquake risk at unprecedented fidelity. The project runs vast ensembles of simulations across more than 1,000 sites, executing approximately one million jobs per site on leadership-class supercomputers to generate detailed seismic hazard forecasts.

“With Horizon, we are going to push the PSHA map from one to two hertz and beyond,” Cui said. "This would cover the earthquake resilience of structures at the house-to-house level. Ultimately, it would provide a more reliable loss estimate for both insurance models and building codes.”

A ShakeOut 2.0 version is also planned for development on Horizon with five hertz linear wave propagation, as well as another CSA developing code for Seisol, which captures dynamic rupture simulations. Also, aftershock simulations are in the works for studying potential magnitude 7.0 quakes in the Los Angeles basin.

“Breaking this frequency barrier with Horizon will be a game changer,” Cui said.

LCCF — A Facility for Other NSF Facilities

Soon, the earliest users will gain access to Horizon and the broader distributed system spanning the LCCF and partner sites, including the Atlanta University Center Consortium, the National Center for Supercomputing Applications, the Pittsburgh Supercomputing Center, and the San Diego Supercomputer Center.

In this age of AI, we expect this system will accelerate science even far beyond what we could imagine.

"Something else I wanted to note was just how LCCF is enabling other NSF investments, particularly our largest facilities," Antypas said. “You can see that in the partnerships that the LCCF has developed, whether that's with the Vera Rubin Observatory, the National Radio Astronomy Observatory, LIGO, Earthscope, or many others that will rely on the LCCF for data analysis."

“In this age of AI, we expect this system will accelerate science even far beyond what we could imagine,” Antypas added.