A look back at SC24: El Capitan crowned as LLNL’s legacy of supercomputing leadership reaches new heights

SC24, held recently in Atlanta, was a landmark event, setting new records and demonstrating Lawrence Livermore National Laboratory’s (LLNL) unparalleled contributions to high-performance computing (HPC) innovation and impact.

SC, the International Conference for High Performance Computing, Networking, Storage and Analysis, is the biggest supercomputing event of the year. SC24 reached an all-time high for the conference series with more than 18,000 attendees and a conference show floor record with 500 exhibitors, serving as a global stage for the latest in HPC technology. LLNL once again stood at the forefront as a world leader in HPC, showcasing groundbreaking hardware, leading discussions on artificial intelligence (AI) and supercomputing-fueled scientific advancements and winning several awards, reflecting the Lab’s instrumental role in shaping the future of computing.

El Capitan: the crown jewel of supercomputing

The centerpiece of LLNL’s SC24 presence was El Capitan, verified by the Top500 at the conference as the world’s fastest supercomputer. At a sustained 1.742 exaFLOPs (1.742 quintillion calculations per second) and over 2.7 exaFLOPs of peak performance, El Capitan topped the November 2024 list of the world’s most powerful supercomputers.

Developed in collaboration with Hewlett Packard Enterprise (HPE) and AMD, El Capitan is a leap forward in speed, precision and efficiency for the National Nuclear Security Administration (NNSA) Tri-Labs (LLNL, Los Alamos and Sandia National Laboratories) and the NNSA’s stockpile stewardship mission.

“El Capitan is more than just a machine — as NNSA’s first exascale computer, it represents a pivotal next step in our commitment to ensuring the safety, security and reliability of our nation's nuclear stockpile without the need to resume underground nuclear testing,” NNSA Acting Deputy Administrator Corey Hinderstein said at a pre-conference press briefing. “This is a momentous day, not only for NNSA, but also for the future of national security, with El Capitan’s computational power and advanced architecture, we are making it clear to the world and to our adversaries that the United States remains at the leading edge of scientific and technical capabilities needed for our defense.”

LLNL’s Weapon Simulation and Computing Associate Director Rob Neely said LLNL and NNSA Tri-Lab scientists “expect El Capitan to make those hero runs of yesterday more commonplace, allowing us to analyze components of the stockpile in great detail and with more precision than ever before.”

“This machine's power will enable us to incorporate various real-world factors, such as materials, manufacturing imperfections, environmental conditions and abnormal and hostile environments,” Neely said. “This means more accurate predictive capabilities and, by extension, better informed decision making for the NNSA stockpile stewardship program.”

El Capitan represents a more than 20-fold increase in peak performance over Sierra, LLNL’s previous most powerful supercomputer. LLNL’s Chief Technical Officer for Livermore Computing Bronis de Supinski credited the new machine’s speed to numerous innovations, including its 40,000-plus AMD Instinct MI300A Accelerated Processing Units (APUs) — a cutting-edge processor combining CPU cores with graphics processing units (GPUs) in a single shared package — allowing for a scale “unprecedented for a GPU-enabled system at an NNSA laboratory.”

LLNL’s Weapon Simulation and Computing Associate Program Director Teresa Bailey called El Capitan “a historic achievement,” adding that the machine will combine simulations with experimental results and AI to achieve breakthroughs in scientific areas such as inertial confinement fusion (ICF), with broad impacts across NNSA’s mission space.

“El Capitan is the right machine for the right time,” Bailey said. “It helps us advance our simulation capabilities, as well as drive AI inference and training based on some of those simulation data sets. We think this is very important to help NNSA and Lawrence Livermore National Laboratory stay on the forefront of high-performance computing.”

Representatives from HPE, AMD, the NNSA and LLNL’s Deputy Director for HPC Terri Quinn and LLNL computational physicist Luc Peterson also participated in the press conference.

LLNL showcases exascale dominance and AI-based innovation

Top500 representatives made El Capitan’s No. 1 ranking official at the SC24 press briefing on Nov. 18, where they unveiled the November 2024 lists of the worlds’ most powerful (Top500) and energy-efficient (Green500) supercomputers. The list underscored the dominance of the Department of Energy in computing, as El Capitan, ORNL’s Frontier and ANL’s Aurora claimed the top three slots. LLNL currently boasts a total of 14 machines on the Top500, by far the most of any supercomputing site in the world.

Later that evening, during the SC24 opening night gala, a fireside chat with NNSA Deputy Assistant Deputy Administrator for Advanced Simulation and Computing Thuc Hoang, LLNL’s Neely, HPE’s Chief Product Officer and Senior Vice President for HPC, AI & Labs Trish Damkroger (formerly of LLNL), and AMD Corporate Fellow Steve Scott delved deeper into the impacts and capabilities of El Capitan. Held at the DOE booth, speakers discussed El Capitan’s ability to perform high-fidelity 3D simulations in support of national security and highlighted the system's innovative technologies and AI capabilities. The event concluded with a discussion on the future of HPC and AI integration.

The combination of AI and exascale computing again took center stage at the DOE booth on Nov. 19, as LLNL’s Peterson gave a talk focused on leveraging AI for ICF on El Capitan. With the multi-disciplinary ICECap (Inertial Confinement Energy on El Capitan) project, Peterson said his team aims to pave the way for exascale digital design and engineering and enhance fusion ignition reliability by using AI to optimize ICF target designs.

“With ICECap we’re putting this all together,” Peterson said. “These advanced simulations are really unprecedented — these are the highest-resolution ICF simulations that have ever been done, and the team has done a lot into making them robust.”

To read more about ICECap, click here.

LLNL also was recognized by the press for groundbreaking contributions to HPC, as teams involving LLNL computer scientists secured two HPCwire Editor’s Choice Awards. One award honored Spack, an influential HPC package manager with more than 6,000 monthly users, as the Best HPC Programming Tool or Technology, presenting the award to creator Todd Gamblin and team. The publication also awarded a collaboration of LLNL, Los Alamos and Sandia national laboratories and Cerebras Systems with Best Use of HPC in Physical Sciences for achieving a 457-fold speed improvement in molecular dynamics simulations using Cerebras’s second-generation Wafer-Scale Engine AI accelerator. The paper describing the work with Cerebras was a finalist for the most prestigious honor in supercomputing, the ACM Gordon Bell Prize, which was presented at SC24 this year to a paper on an electron-modeling algorithm run on ORNL’s Frontier.

During the show, Kathryn Mohror, a Distinguished Member of LLNL’s Technical Staff, was officially presented with the ACM SIGHPC Emerging Woman Leader in Technical Computing award. The recognition celebrated Mohror’s leadership and contributions to the field of HPC, particularly her work on scalable fault-tolerant systems, workflows and performance tools, as well as her mentorship and efforts to support diversity in computing.

Shaping the future of HPC with insights, innovations and industry collaboration

SC24 was not just a celebration of achievements, but also served as a platform for envisioning the future of HPC, as discussions on exascale readiness, sustainable computing and hybrid infrastructure in the age of AI signaled the direction of the field. In all, the conference featured 44 workshops and 89 Birds-of-a-Feather (BoF) sessions, many of which featured LLNL employees playing leading roles.

The technical program began on Nov. 17, with LLNL computer scientists Gamblin, Greg Becker and Alec Scott leading a tutorial on Spack. Other tutorials held early in the conference included LLNL’s Johannes Doerfert (on Portable GPU Acceleration of HPC Applications with Standard C++23), Giorgis Georgakoudis (Parallel Programming in Python with OpenMP) and Peter Lindstrom (lossy compression of scientific data).

Other LLNL-led workshops, BoFs and discussions emphasized diversity and inclusivity in HPC, exploring ways to empower underrepresented groups in HPC and developing programs to enhance career opportunities for the next generation of computational scientists. Computing Workforce Administrator Jamie Lewis, Computing Workforce Manager Marisol Gamboa and Organizational Development Consultant Andrekka “AJ” Lanier helped lead the Students@SC program — with sessions on creating resumes, overcoming imposter syndrome and career coaching. LLNL Computing Communications Specialist Holly Auten presented on creating engaging social media at a BoF on HPC-related outreach (co-led by LLNL HPC Technical Consultant Jane Harriman), and Elsa Gonsiorowski co-chaired the 20th international Women in HPC workshop.

The technical program continued Nov. 19, as LLNL computer scientist Hariharan Devarajan presented DFTracer, a data flow tracing tool for AI-driven workflows. In the afternoon, computer scientist Ignacio Laguna presented on the challenges and opportunities of foundational large language models (LLMs). Computational scientist Cody Balos and software developer Kristi Belcher co-led a BoF on tools and strategies to enhance software development efficiency in HPC and harness the full potential of exascale systems, while Livermore Computing (LC) I/O Strategist Cameron Harr presented on LC’s use of the open-source Lustre file system.

Meanwhile, LLNL computer scientist Harshitha Menon co-led a BoF session on embedding LLMs in HPC workflows, where she spoke about adapting HPC infrastructure to accommodate AI workloads and the potential benefits for scientific research. LLNL Systems Engineer Chris DePrater also spoke about liquid cooling controls at exascale, highlighting LLNL’s commitment to sustainability — an increasingly critical focus in the HPC community. At the evening’s Top500 BoF, de Supinski accepted the certificate from Top500 representatives for El Capitan becoming the most powerful computing system in the world and spoke about the machine’s technical capabilities, as well as those of other machines procured as part of the El Capitan contract, including Tuolumne, RZAdams and Sandia National Laboratories’ new and similar El Dorado machine.

At the DOE booth, LLNL researcher Peer-Timo Bremer also demonstrated a tabletop model of an autonomous AI laboratory called Sidekick, which replicates a high-repetition laser experiment in miniature. The project is part of a collaboration with NVIDIA fostered through the AI Innovation Incubator. The following day, High-Performance Computing for Energy Innovation (HPC4EI) Director Aaron Fisher discussed how the HPC4EI program, managed by LLNL for DOE, collaborates with DOE labs to help industry leverage supercomputing resources and expertise to accelerate energy efficiency, decarbonization and other aims.

“We're in the next industrial revolution right now, but this one isn't being powered by machinery; this one is being powered by data,” Fisher said. “In the past few decades, we have really multiplied the amount of data that we can process, collect and use to put together better and better manufacturing techniques. Industry is engaging with this and it’s coming to fruition as we speak.”

In the afternoon, LLNL’s Laguna participated in a panel discussion on bridging gaps between AI and traditional HPC, while LLNL computer scientist Brian Van Essen spoke on unstructured hydrodynamics in spatial dataflow architectures, for a BoF on AI accelerators for HPC. To end the day, Ulrike Meier Yang participated in a panel discussion on the importance of collaboration among the SIAM Activity Group on Supercomputing and other organizations as a means to deliver impactful services in HPC.

The technical program continued Nov. 21, as LLNL computer scientist Olga Pearce discussed running tools Caliper and Thicket at exascale — part of a panel on performance evaluation tools. LLNL summer student intern Zane Fink of the University of Illinois Urbana-Champaign then presented on HPAC-ML, a programming model that enables the use of embedded machine learning surrogates in scientific applications.

SC24 concluded on Nov. 22 with a handful of workshops and panels, including the 11^th  International Workshop on HPC User Support Tools (HUST) co-led by Gonsiorowski, and a panel featuring de Supinski, which tackled the topic of sustainability in AI. Other LLNL session chairs and co-chairs included Erik Draeger (15^th Workshop on Latest Advances in Scalable Algorithms for Large-Scale Heterogeneous Systems), Maya Gokhale (MEMO’24: International Workshop on Memory System, Management and Optimization) and Laguna (8^th International Workshop on Software Correctness for HPC Applications).

Chaired by LLNL’s Principal Deputy Director for Computing Lori Diachin, SC25 will be held Nov. 16-21, 2025, in St. Louis. For a full list of LLNL SC24 contributors, including SC committee members, visit the website.