Scientific Computing in an AI World: The Conversation Continues
When ICL’s March Newsletter reported on a new perspective by Jack Dongarra, Dennis Gannon (Indiana University), and Daniel Reed (University of Utah), the piece read like the opening of a longer argument. That essay, “Ride the Wave, Build the Future: Scientific Computing in an AI World,” held that the center of gravity in high-performance computing has shifted toward hyperscale AI platforms, and that the scientific computing community should both ride the wave of AI and build the next generation of systems for discovery. In the months since, its authors have carried that argument into new venues.
In May, Dongarra took up the thread on Scientific Software Horizons, a podcast hosted by Michael Heroux and Lois Curfman McInnes. Over a two-part conversation, he traced why AI marks an inflection point for scientific computing and how simulation, data, and machine learning are converging (Part 1), then turned to the workforce, workflow, and policy questions that follow (Part 2), reflecting along the way on lessons from LINPACK and the TOP500.
In June the argument reached its widest audience yet, as a Policy Forum in Science, “Scientific computing in an AI world,” distilled it for a policy and research readership. Dongarra, Reed, and Gannon call for a next-generation strategy that integrates AI with simulation and treats energy efficiency as a primary design constraint, propose an initial “minimum viable” suite of workflow-shaped benchmarks, and describe data and models as “intellectual gold.” They close with a call for a “moonshot” to build a shared foundation that serves both scientific computing and AI.
Across essay, interview, and policy forum, the through-line holds steady: AI is not replacing scientific computing but transforming it, and the community has a chance to shape that transition rather than be swept along by it.
ISC 2026 Connects the Dots in Hamburg
The 41st ISC High Performance conference (ISC 2026) brought more than 4,000 attendees and 188 exhibitors to the Congress Center Hamburg from June 22–26, 2026. Held under the theme “Connecting the Dots,” the meeting drew the international high-performance computing community together with a growing contingent from the AI and quantum computing fields. Hartwig Anzt had a leadership role at the conference, serving as Tutorial Chair for a program of 22 in-depth tutorials that spanned performance optimization, reproducibility, AI methods, and hybrid quantum-classical workflows. This year’s conference sessions are available for free on-demand viewing through the ISC 2026 portal.
Martin Schulz (Technical University of Munich and the Leibniz Supercomputing Centre) opened the program on June 23 with a keynote titled “HPC: A Heterogeneous Future.” Schulz argued that the steady gains of the Moore’s Law era have given way to a “complexity wall,” and that progress now depends on embracing integrated heterogeneity, weaving together quantum, neuromorphic, and photonic technologies rather than treating them as obstacles.
Jack Dongarra and Michela Taufer (University of Tennessee, Knoxville) present the 2026 Jack Dongarra Early Career Award to Devesh Tiwari (Northeastern University)
The 2026 Jack Dongarra Early Career Award went to Devesh Tiwari (Northeastern University), presented on June 24 by Jack Dongarra and by Michela Taufer. Tiwari was recognized for advancing sustainable high-performance computing and for his work on post-Moore systems, including hybrid quantum-classical computing.
Piotr Luszczek organized the HPC on Heterogeneous Hardware (H3) workshop, held June 26. The workshop featured novel work in algorithmic research, software library design, programming models, networking technologies, and workflow development for increasingly heterogeneous hardware, including analog and quantum computing, a variety of accelerators, and many form factors and power envelopes. Its steering committee was made up entirely of ICL alumni: Luszczek and Hartwig Anzt, joined by Bilel Hadri and Hatem Ltaief (both of King Abdullah University of Science and Technology).
Dongarra returned to close the conference on Thursday evening with the keynote “HPC in Transition,” reflecting on how the surge of AI-driven investment is reshaping the field that scientific simulation has long defined.
Looking ahead to next year, ISC named Rio Yokota (Institute of Science Tokyo and the RIKEN Center for Computational Science) as program chair for ISC 2027.
Benchmark Updates
LineShine’s No. 1 TOP500 and HPCG certificates are presented during the TOP500 session at ISC 2026.
The 67th TOP500 list was released at ISC 2026 alongside its companion HPCG and HPL-MxP rankings, and the headline result was a newcomer at the very top. LineShine, a previously unlisted system at the National Supercomputing Center in Shenzhen, debuted at No. 1 with 2.198 Exaflop/s on the High Performance Linpack (HPL) benchmark, displacing El Capitan. Built on a custom Chinese platform with 13.79 million cores, LineShine is the first system to sustain more than two exaflops of double-precision performance using CPUs only, and its debut marks the first time a Chinese system has led the list since 2017.
| TOP500 June 2026 | Country | Exaflop/s | |
|---|---|---|---|
| 1 | LineShineNational Supercomputing Center, Shenzhen |  |
2.198 |
| 2 | El CapitanLawrence Livermore National Laboratory |  |
1.809 |
| 3 | FrontierOak Ridge National Laboratory | |
1.353 |
| 4 | AuroraArgonne National Laboratory | |
1.012 |
| 5 | JUPITER BoosterJülich Supercomputing Centre |  |
1.000 |
LineShine’s arrival pushes the number of systems above the exascale threshold from four to five and, for the first time, places exascale machines in Asia, North America, and Europe at once.
On the HPCG benchmark, which rewards performance on data-intensive application patterns rather than peak floating-point throughput, LineShine again leads, with Fugaku and Eni’s new HPC7 system reshuffling the field behind the U.S. Department of Energy machines.
| HPCG June 2026 | Country | Petaflop/s | |
|---|---|---|---|
| 1 | LineShineNational Supercomputing Center, Shenzhen | |
22.00 |
| 2 | El CapitanLawrence Livermore National Laboratory | |
17.41 |
| 3 | FugakuRIKEN Center for Computational Science |  |
16.00 |
| 4 | FrontierOak Ridge National Laboratory | |
14.05 |
| 5 | HPC7Eni S.p.A. |  |
5.95 |
On HPL-MxP, the mixed-precision benchmark maintained at ICL, El Capitan held the lead at 16.7 Exaflop/s, a 9.2x speedup over its standard HPL score. LineShine entered in fourth with a comparatively modest 3.6x speedup, consistent with a CPU-only design that lacks the dedicated low-precision accelerators that boost the GPU-based U.S. systems.
| HPL-MxP June 2026 | Country | Exaflop/s | |
|---|---|---|---|
| 1 | El CapitanLawrence Livermore National Laboratory | |
16.7 |
| 2 | AuroraArgonne National Laboratory | |
11.6 |
| 3 | FrontierOak Ridge National Laboratory | |
11.4 |
| 4 | LineShineNational Supercomputing Center, Shenzhen | |
7.92 |
The lists arrived with news about their own future. ACM, through its SIGHPC special interest group, announced that it will take over stewardship of the TOP500, the result of roughly two years of planning. The transition is intended to preserve the legacy of the project, which Dongarra helped found in 1993, while opening it to new ideas as it continues into its fourth decade.
ICL Hosts PaRSEC Runtime Ecosystem Workshop
ICL hosted the Workshop on Fostering an Open-Source Runtime Eco-System for PaRSEC on May 6–7, 2026, at the University of Tennessee, Knoxville. The two-day meeting brought together users, developers, and collaborators of PaRSEC (Parallel Runtime Scheduling and Execution Controller) to discuss the technical direction of the task-based runtime, strengthen community coordination, and plan the next phase of the project. The workshop was organized by the FOREST team (Natalie Beams, Joseph Schuchart, and Qinglei Cao) with support from the NSF POSE program.
Jack Dongarra opened the workshop with welcome remarks, and the program featured talks from across the PaRSEC community, including George Bosilca and a keynote from Hatem Ltaief (KAUST), with speakers from national laboratories, universities, and industry. Thanks to the ICL operations team for their support in coordinating the event.
Jack Dongarra Named to Forbes Self-Made 250
Jack Dongarra has been named #236 on the Forbes Self-Made 250, the magazine’s ranking of the greatest living self-made Americans, published in honor of the nation’s semiquincentennial. To build the list, Forbes mined its 109-year archive, polled its reporters, and “canvassed AI, running hundreds of queries through both ChatGPT and Gemini” before putting the names to a panel of judges. The resulting list includes rags-to-riches billionaires and also “pioneering scientists, Supreme Court justices and others whose wealth is measured in influence and impact, not just dollar signs.” Oprah Winfrey tops the list at #1, and East Tennessee’s own Dolly Parton ranks #7.
In its entry, Forbes notes that Dongarra was born to an immigrant father with little education and overcame dyslexia as a child. On the list he keeps company with a range of fellow scientists and computing pioneers, among them Nobel-winning neuroscientist John O’Keefe (#235), mRNA pioneer Katalin Karikó (#90), and AI researcher Fei-Fei Li (#191), alongside technology figures such as Nvidia’s Jensen Huang (#81), Oracle’s Larry Ellison (#11), and Google’s Sundar Pichai (#142).
Forbes also published a companion ranking of the 250 greatest historic self-made Americans.
ICL Wins NSF Grant to Study Precision Emulation on AI Chips
ICL has received a four-year, $800,000 award from the National Science Foundation to study how the shift toward AI hardware is reshaping scientific computing. The project, “Precision Emulation on AI Chips (PEACH): Challenges, Opportunities, and Impact on Scientific Computing,” is supported by NSF’s Computer and Information Science and Engineering (CISE) directorate through its Future Computing Research (Future CoRe) program. Ahmad Abdelfattah is the principal investigator with co-PI Natalie Beams. The work will run from July 2026 through June 2030.
As chip makers tune their hardware for AI, they have begun scaling back native support for the 64-bit double precision (FP64) arithmetic that scientific simulations have long relied on; NVIDIA’s Blackwell architecture is the first to deliver lower FP64 throughput than the generation before it. PEACH studies emulation techniques, such as the Ozaki scheme, that reconstruct high-precision results from fast low-precision operations, measuring their accuracy and performance across numerical linear algebra and pinpointing where they break down. The project also looks to develop a new class of mixed-precision algorithms, with the results released to the community through ICL’s open-source MAGMA library.
Conference Reports
In China, Dongarra Chairs the ASC26 Finals and Keynotes HACI 2026
Jack Dongarra’s travels took him to China in May 2026. As Chair of the Advisory Committee, he helped oversee the finals of the ASC Student Supercomputer Challenge (ASC26), held May 16–20 at Wuxi University in Jiangsu Province. Now in its 13th year, ASC is one of the world’s three premier student supercomputing competitions, alongside the student cluster competitions at ISC and SC. More than 300 teams registered worldwide, and the 25 that reached the finals built their own clusters under a strict 5,000-watt power budget, taking on challenges that ranged from embodied AI and gravitational-wave simulation to quantum circuit simulation and climate modeling. Peking University took the championship, with Tsinghua University as runner-up.
Days later, in Shenzhen, Dongarra delivered the opening talk at the International Forum for HPC & AI Co-Driven Innovation (HACI 2026), held May 22–25. Inaugurated in 2024, HACI gathers researchers in the Guangdong–Hong Kong–Macao Greater Bay Area to explore the convergence of high-performance computing and artificial intelligence. His talk, “HPC in Transition,” examined how the surge of investment in computing infrastructure is now driven largely by AI workloads, and what that shift means for the scientific simulations that have long defined the field. He later joined a panel, “The Age of HPC+AI: From Vision to Infrastructure,” alongside other international HPC leaders.
Piotr Luszczek Co-Chairs New Mixed-Precision Workshop at IPDPS 2026
The inaugural MxP4S workshop at IPDPS 2026 in New Orleans.
Piotr Luszczek co-chaired the first Workshop on Mixed- and Adaptive-Precision Numerics for Scientific Computing (MxP4S), held May 25, 2026, at the 40th IEEE International Parallel & Distributed Processing Symposium (IPDPS) in New Orleans. Luszczek organized the workshop with Aditya Kashi (Oak Ridge National Laboratory), with a steering committee that included Hartwig Anzt.
The workshop responded to a widening gap in modern hardware: new HPC and AI platforms offer rapidly growing low-precision throughput while double-precision performance plateaus, yet many scientific simulations cannot use low precision directly without sacrificing accuracy or stability. Through a program of invited talks, MxP4S explored how computational scientists can harness low-precision hardware responsibly and move toward adaptive-precision methods that balance performance, accuracy, and energy efficiency across application domains.
Inaugural Midwest Randomized Linear Algebra Workshop
ICL alumni Daniel Bielich and Max Melnichenko with Piotr Luszczek at the Midwest Randomized Linear Algebra Workshop.
ICL alumni Daniel Bielich (Ansys) and Max Melnichenko (University of California, Berkeley) joined with Piotr Luszczek at the Inaugural Midwest Randomized Linear Algebra Workshop, held May 11–12, 2026, at the Fluno Center in Madison, Wisconsin. The workshop gathered researchers to take stock of the state and direction of randomized numerical linear algebra (RLA), a fast-growing set of methods that use randomness to accelerate large-scale matrix computations. The workshop also charted community priorities through tutorials, plenary talks, and breakout sessions.
Interview
Daniel Barry
Congratulations on your PhD achievement. Can you catch us up on your academic journey that led to your current role as a Research Scientist?
Thank you so much! Throughout my graduate studies, I did various experiments using the Performance API (PAPI), such as designing benchmarking strategies for accurate uncore-event monitoring. I also developed a methodology to automatically compose performance metrics using natively available hardware events. This allowed me to garner familiarity with PAPI’s structure and innerworkings.
What are you currently working on at ICL?
Now, as a Research Scientist, I am developing new features for PAPI, such as integrating AMD’s ROCTX library with PAPI’s rocp_sdk component, as well as developing infrastructure to automate creating tables of performance events that PAPI can expose.
What are you most interested in exploring next in your research or what research areas are you most excited about looking forward?
I am very interested in developing support for PC sampling in the PAPI framework. In addition, I would like to do some experiments utilizing the various power-monitoring capabilities of PAPI.
What advice would you give to future students at ICL who hope to follow a similar path as yours?
I would tell students two pieces of advice. (1) When it comes to coursework, pace yourself! Staying up all night working on projects and assignments sometimes feels necessary, but doing so can have a more negative academic AND research impact than simply submitting an incomplete or imperfect assignment earlier on. (2) Have humility in the face of new evidence in research. Sometimes results from experiments are unexpected. It is important to honor newly acquired knowledge because it can lead to richer and more nuanced conclusions and models.
What are some of your interests or hobbies outside of work?
I really enjoy swimming, cooking, watching movies, and playing video games. Throughout grad school, I accrued a backlog of movies to watch, and I have been enjoying catching up!
In your 2018 interview, you said that if you weren’t working at ICL, you’d want to be working here. How does it feel to have made that a reality?
Continuing my career with ICL has been very gratifying. It is an honor to work with the PAPI team, and we have great synergy. Having my role at ICL is rewarding because I am able to contribute to HPC technology through my research and software development.
