93Events
6Days
Nov 15, 2026 – Nov 20, 2026Program window
Sunday, November 15
-
Tutorial
Deep Learning at Scale: Engineering High-Resolution Scientific AI Models
Full description
Deep learning is increasingly used in scientific applications involving high-dimensional, high-resolution data. As models, datasets, and data resolutions scale, so does the need for efficient, distributed training on high performance computing (HPC) systems. This tutorial addresses practical challenges in training scientific AI models on high-resolution data, including data throughput, memory pressure, and communication overhead in distributed HPC workflows. Participants will learn to profile and build efficient data pipelines, optimize single-GPU performance, and scale training across GPUs and nodes using data, tensor, and spatial (domain) parallelism. Through hands-on examples drawn from scientific use cases, we will show how different model families, such as transformers, convolutional neural networks, and neural operators, benefit from these parallelization strategies. We will emphasize the performance trade-offs among computation, memory, and communication that determine scalability for high-resolution scientific workloads. Attendees will gain experience running optimized workloads on GPU-based supercomputers and learn strategies they can apply to train scientific AI models that exceed single-device memory limits while scaling efficiently.- Presenters
- Shashank SubramanianLawrence Berkeley National Laboratory (LBNL)
- Steven FarrellLawrence Berkeley National Laboratory (LBNL)
- Josh RomeroNVIDIA Corporation
- Thorsten KurthNVIDIA Corporation
- Corneel CasertLawrence Berkeley National Laboratory (LBNL)
- Adam RupeLawrence Berkeley National Laboratory (LBNL)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TUT — Tutorial Reg Pass
-
Tutorial
Using Containers To Accelerate HPC
Full description
The use of containers has revolutionized the way in which industries and enterprises have developed and deployed computational software and distributed systems. This containerization model has gained traction within the HPC community as well with the promise of improved reliability, reproducibility, portability, and levels of customization that were not previously possible on supercomputers. This adoption has been enabled by a number of HPC container runtimes that have emerged including Apptainer, Shifter, Sarus, Podman, and others. This hands-on tutorial looks to train users on the use of containers for HPC use cases. We will provide a detailed background on Linux containers, along with an introductory hands-on experience building a container image, sharing the container and running it on an HPC cluster. Furthermore, the tutorial will provide more advanced information on how to run MPI-based and GPU-enabled HPC applications, how to optimize I/O intensive workflows, how to set up GUI-enabled interactive sessions, and how to use containers for agentic AI and MCP servers. Users will leave the tutorial with a solid foundational understanding of how to utilize containers on HPC resources using Podman, Docker, and Apptainer, and in-depth knowledge to deploy custom containers for HPC and AI on their own resources.- Presenters
- Adam LavelyLawrence Berkeley National Laboratory (LBNL); National Energy Research Scientific Computing Center (NERSC)
- Sameer ShendeUniversity of Oregon
- Carlos Eduardo ArangoNVIDIA Corporation
- Sylvain BernardSandia National Laboratories
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TUT — Tutorial Reg Pass
-
Workshop
13th Annual International Workshop on Innovating the Network for Data Intensive Science — INDIS
Full description
Data networks are the circulatory system for science, discovery, society and industry. But scientific data networking is not a solved problem, and significant challenges remain. These challenges reflect scientists’ persistent needs from data networks: the needs for ever increasing capacity, flexibility, consistency, and collaboration. The theme for the 13th INDIS workshop is Connected Science: Bridging Research and Industry — reflecting not only the role of networking in connecting scientific infrastructure, but also our community’s collaborations across different types of organizations to advance data-intensive science. This year INDIS plans to continue its tradition of having a discussion- and interaction-oriented schedule; based on past feedback, this is always well received. INDIS 2026 delivers an essential complement to the HPC-focused SC26 Technical Program, providing a dedicated platform for HPC researchers to explore data-intensive computing, interconnects, and networking.- Session Chairs
- Nik SultanaIllinois Institute of Technology; SCinet
- Ezra KisselEnergy Sciences Network (ESnet)
- Fatema Bannat WalaEnergy Sciences Network (ESnet)
- Akbar KaraCiena Corporation; SCinet
- David MartinLawrence Berkeley National Laboratory (LBNL); Energy Sciences Network (ESnet)
- Cees de LaatUniversity of Amsterdam; Lawrence Berkeley National Laboratory (LBNL)
- Rafael Silva-GuimaraesFederal Institute of Espírito Santo, Department of Informatics
- Esen Gokpinar SheltonIndiana University
- Alfredo GoldmanUniversity of São Paulo
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
-
Workshop
The 2nd International Workshop for Software Frameworks and Workload Management on Quantum-HPC Ecosystems
Full description
The integration of quantum computing (QC) with high performance computing (HPC) is emerging as a transformative paradigm for addressing complex scientific and engineering challenges. Given current quantum hardware limitations, such as noise, limited qubit counts, and constrained circuit depth, hybrid quantum–HPC (Q-HPC) ecosystems provide a practical path forward by combining complementary strengths. This workshop focuses on software frameworks, AI-enabled orchestration, and workload management strategies essential for scalable Q-HPC integration. Topics include architecture co-design, heterogeneous resource management, networking, fault tolerance, error mitigation, programming models, middleware, benchmarking, and large-scale distributed Q-HPC systems. Emphasis will be placed on open, interoperable ecosystems such as openQSE and on ML/AI techniques for adaptive workflow orchestration, topology-aware scheduling, performance prediction, and autonomous runtime control. Through invited talks, technical presentations, panels, and special sessions, the workshop will foster collaboration across academia, industry, national laboratories, and funding agencies to define research priorities and accelerate AI-driven Q-HPC ecosystems.- Session Chairs
- In-Saeng SuhOak Ridge National Laboratory (ORNL)
- Esam El-ArabyUniversity of Kansas
- Edo GiustoUniversity of Naples Federico II; Helix 42
- Katie KlymkoNational Energy Research Scientific Computing Center (NERSC); ESnet; Lawrence Berkeley National Laboratory (LBNL)
- Frank MuellerNorth Carolina State University
- Jorge EchavarriaMunich Quantum Valley (MQV); Leibniz Supercomputing Centre (LRZ)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
-
Workshop
The 9th Annual Parallel Applications Workshop, Alternatives to MPI+X (PAW-ATM)
Full description
Supercomputers get faster and more complex every year. MPI, long the dominant model for distributed computation, has adapted by combining with models for intra-node parallelism (e.g., OpenMP, CUDA). These MPI+X hybrids offer performance but demand significant programmer effort to write, debug, and tune applications. Alternatives to MPI+X are worth exploring as programmer productivity becomes a major component of the time to science. Alternatives include parallel programming languages (e.g., Chapel, Regent, Fortran), general purpose libraries (e.g., Charm++, COMPSs, HPX, Legion, UPC++), and domain-specific libraries (e.g., Arkouda, Dask, Spark). With many options to choose from, it is hard for programmers to know which alternative models are appropriate for their application and for programming model developers to understand the opportunities for improvement. PAW-ATM brings together both application experts and programming model developers to disseminate knowledge about available applications and models, as well as advance their design.- Organizers
- Karla Vanessa Morris WrightSandia National Laboratories
- Daniele LezziBarcelona Supercomputing Center (BSC)
- Katherine RasmussenLawrence Berkeley National Laboratory (LBNL)
- Elliott SlaughterSLAC National Accelerator Laboratory
- Tasuku HiraishiKyoto Tachibana University
- Olivier AumageFrench Institute for Research in Computer Science and Automation (INRIA)
-
Workshop
Fine-Tuning Foundation Models for Micro-CT Image Segmentation Using HPC Platforms
Full description
Foundation models have demonstrated exceptional performance in computer vision, but adapting them to scientific imaging requires domain-specific datasets, scalable training methodologies, and high-performance computing (HPC) resources. This work presents methodologies for fine-tuning and deploying foundation models for synchrotron X-ray micro-computed tomography (micro-CT) image segmentation. We describe the adaptation of SAM3 for instance segmentation and DINOv3-Seg for semantic segmentation. We present strategies for data preprocessing, model configuration, loss design, and hyperparameter optimization for fine-tuning these models. We further describe scalable fine-tuning on local GPU servers through a web-based training interface and on the National Energy Research Scientific Computing Center (NERSC) through Slurm-based job submission. Finally, we present distributed inference strategies on NERSC and the Argonne Leadership Computing Facility (ALCF), enabling near real-time segmentation of large synchrotron datasets across hundreds of GPUs. Together, these methods provide guidance for adapting, training, and deploying foundation models for large-scale scientific image segmentation at synchrotron facilities.- Author/Presenters
- Xiaoya ChongLawrence Berkeley National Laboratory
- David AbramovLawrence Berkeley National Laboratory
- Pradyumna ElavarthiLawrence Berkeley National Laboratory
- Elizabeth ClarkLawrence Berkeley National Laboratory
- Petrus ZwartLawrence Berkeley National Laboratory
- Wiebke KoeppLawrence Berkeley National Laboratory
- Dilworth ParkinsonLawrence Berkeley National Laboratory
- Dylan McReynoldsLawrence Berkeley National Laboratory
- Daniel AllanBrookhaven National Laboratory
- Johanna WekerSLAC National Accelerator Laboratory
- Nicholas SchwarzArgonne National Laboratory
- Francesco De CarloArgonne National Laboratory
- Nghia VoBrookhaven National Laboratory
- Sharif AhmedDiamond Light Source
- Tim SnowDiamond Light Source
- Tamas VargaPacific Northwest National Laboratory
- Tyler FarrLawrence Berkeley National Laboratory
- Harry LisabethLawrence Berkeley National Laboratory
- Andrew McElroneUniversity of California, Davis
- Tanny ChavezLawrence Berkeley National Laboratory
- Alexander HexemerLawrence Berkeley National Laboratory
-
Workshop
Invited talk: "Distributed HPC in Julia without MPI"
Full description
MPI has long been the default substrate for distributed scientific computing, but it is not the only path to scalable performance—and in Julia, it is often not the most productive one. This talk surveys the landscape of MPI-free distributed computing in Julia, centered on the standard-library Distributed package and its surprisingly performant primitives for remote execution, data movement, and cluster management. I will review which widely-used Julia packages already run on top of Distributed, and how their capabilities compare to their MPI-based counterparts. The core of the talk focuses on Dagger.jl, a task-based parallel runtime that provides unified support for both Distributed and MPI: every algorithm expressed in Dagger runs unchanged across either backend, as well as across multi-CPU and multi-GPU configurations. I will highlight Dagger's Datadeps API, which lets users express data dependencies declaratively and have the scheduler automatically handle communication, synchronization, and placement—whether the underlying transport is Distributed or MPI. Through concrete examples, I will show how this abstraction delivers portability without sacrificing performance, and close with an overview of the current state of distributed computing across the broader Julia package ecosystem.- Session Chair
- Damian RousonLawrence Berkeley National Laboratory (LBNL)
- Presenter
- Julian SamarooMIT
-
Workshop
Security Panel
Full description
SC26 has not supplied a description for this event.
- Panelists
- Adam SlagellEnergy Sciences Network (ESnet)
- Anita NikolichUniversity of Illinois at Urbana-Champaign
- Ryan WalkerUniversity of Illinois at Urbana-Champaign
- Michael SkwarekArgonne National Laboratory
- Matt KwiatkowskiFermi National Laboratory
-
Workshop
Invited talk: "From MPI+X to Domain-Specific Abstractions: Building Sustainable Weather and Climate Software"
Full description
Weather and climate models are large and complex collections of mathematical modelling software. They can take more than a decade to develop and can be in use for a quarter of century or more. They also typically undergo constant revisions and improvements from teams of developers. The proliferation of programming models in the MPI+X paradigm is therefore particularly problematic as cost of developing and maintaining multiple X implementations for different architectures in such a code base becomes prohibitive. The rapid evolution of the programming models has exacerbated the problem. Fortran has been the de facto standard programming language for weather and climate applications and can considered a low-level domain specific language for numerical computation. The ability to write code that in some sense looks likes the maths is part of its appeal and enables developers to be productive. The invasive level of parallel code additions required for performance degrades the sustainability of the code base. One approach to solving these problems has been to raise the level of abstraction from a general-purpose language to a domain specific language (DSL). Here domain specific information is encoded in and exploited by the programming model for optimisation purposes. Key to this approach is a strong separation of concerns between the science/maths code and the parallel and optimisation code. This approach enables developers to write code which looks like maths whilst allowing the portability and performance by having different routes to optimised codes for different architectures. This comes with the cost of designing and maintaining the DSL. The Met Office is part of the Momentum partnership and is developing the Momentum modelling software system. The atmosphere model, LFRic, is developed and co-designed in a its own DSL, embedded in Fortran. Part of the DSL is the python domain specific compiler, PSyclone. The advantages and limitations of the this approach are discussed, including the consequences of design choices such as embedding the DSL in Fortran. Some performance results from LFRic are presented as evidence to support the arguments. Finally, how emerging agentic programming techniques may complement DSL-based approaches and what this implies for future weather and climate software development is considered.- Session Chair
- Damian RousonLawrence Berkeley National Laboratory
- Presenter
- Christopher MaynardMet Office UK
-
Workshop
Navigating the Spectrum of Science Workloads with ESnet JLab FPGA Accelerated Transport (EJFAT)
Full description
Cross-facility workflows are no longer optional but a requirement for the success of current and future science campaigns. From remote labs to multi-institutional science campaigns, advances in instrumentation and technological innovation are enabling experimental and observational scientific user facilities to generate data faster than processing and storage can handle. Our work explores EJFAT as a potential solution to these challenges. While EJFAT has been shown to be performant and scalable, this paper explores considerations for integrating EJFAT into real-world scientific experiments. We explore two applications: one highly integrated, the other highly generalizable. The first is a highly tuned and optimized workflow for pushing data as fast as possible to meet the goals of scientific experiments, while the second is a general ecosystem lowering the barrier for scientists to leverage HPC resources with minimal effort representing two common classes of applications typical found in scientific user facilities.- Author/Presenters
- Harinarayan KrishnanLawrence Berkeley National Laboratory
- David ShapiroLawrence Berkeley National Laboratory
- Ronald PandolfiLawrence Berkeley National Laboratory
- Sophie MorleyLawrence Berkeley National Laboratory
- Yatish KumarLawrence Berkeley National Laboratory
-
Workshop
A Performance-Portable GPU-Accelerated Solution for Particle-Particle Particle-Mesh Method using Cabana and FFTX
Full description
One approach to the development of high-performance computing applications is based on libraries that implement algorithmic motifs in a portable, performant and concise fashion. We develop a GPU-accelerated algorithm for Method of Local Corrections, a particle-particle particle-mesh method for simulating incompressible flow in 3D, using such motif libraries. The algorithm comprises of structured grid calculations, local particle interactions, and free-space convolution. We use two motif libraries: Cabana, a motif library for particles, built on top of Kokkos; and FFTX for fast Fourier Transform, that uses the SPIRAL code generation system. By using motif libraries, we obtain a concise and portable implementation (2K lines of code). We compare the measured performance to that predicted by the roofline model and obtain 52\%-86\% on NVIDIA A100, 42\%-61\% on NVIDIA H100, 22\%-42\% on AMD MI100, and 17\%-38\% on AMD MI300A.- Author/Presenters
- Het MankadOak Ridge National Laboratory (ORNL)
- Zoe BarbeauStanford University
- Phillip ColellaLawrence Berkeley National Laboratory
- Sanil RaoCarnegie Mellon University
- Franz FranchettiCarnegie Mellon University
- Jeffrey VetterOak Ridge National Laboratory
- Keita TeranishiOak Ridge National Laboratory
-
Workshop
Cross-Fabric Policy Reasoning for Hybrid InfiniBand/Ethernet HPC Systems
Full description
SC26 has not supplied a description for this event.
- Presenters
- Vedant Sawal
- Anna GiannakouLawrence Berkeley National Laboratory
-
Workshop
Rising Innovators Research Showcases
Full description
Five invited rising innovators deliver 5-min lightning talks showcasing their research, technical ideas, projects, discoveries, or emerging directions in HPC and AI: → Algorithms as if Architecture Mattered: Moving Less, Computing Faster — Rabab Alomairy (KAUST, Saudi Arabia) → Sparse by Nature, Parallel by Design: HPC for the Life Sciences — Giulia Guidi (Cornell University, USA) → SwarmLoRA: Serverless Shared-Computation Disaggregation for Isolated, High-Throughput Multi-LoRA Serving — Tong Shu (University of North Texas, USA) → Abstract, Portable, Performant: Realizing Sustainable HPC — Floris-Jan Willemsen (Leiden University, Netherlands) → The Meaning of Compute: formal approaches to exascale AI development — Rubén Coll Sánchez (Huawei, Sorbonne University, France)- Session Chair
- Sarra RefaiForschungzentrum Juelich
- Presenters
- Rabab AlomairyKing Abdullah University of Science & Technology
- Giulia GuidiCornell University
- Tong ShuUniversity of North Texas, USA
- Floris-Jan WillemsenLeiden University, Netherlands
- Rubén Coll SánchezHuawei Technologies, France
-
Workshop
Short Talk: Portable AI Workflows and Federated Agents for Experiment-in-the-Loop Synchrotron Science
Full description
SC26 has not supplied a description for this event.
- Presenters
- Tim DunnSLAC National Accelerator Laboratory
- David AbramovLawrence Berkeley National Laboratory
- Justas BalcasEnergy Sciences Network (ESnet)
- Yadu BabujiUniversity of Chicago; Argonne National Laboratory
- Alexander HexemerLawrence Berkeley National Laboratory
- Dylan McReynoldsLawrence Berkeley National Laboratory
- Johanna Nelson WekerSLAC National Accelerator Laboratory
- Hannah ParragaArgonne National Laboratory
- Nicholas SchwarzArgonne National Laboratory
- Bo SuSLAC National Accelerator Laboratory
-
Workshop
Recall Is Not Honesty: Benchmarking Multi-Agent Architectures for Trustworthy HPC Operational Analytics
Full description
Agentic systems answer operational queries over HPC telemetry for domain scientists who cannot verify the responses. Trustworthiness requires recovering the facts telemetry supports and refusing the questions it does not, since no facility instruments everything it runs. Existing architectures are built for capability on the assumption that reliability follows, and inference remains untested. We compare five configurations, one tool-using agent and four multi-agent designs varying only in inter-agent communication and shared-state structure, on EPIQ, a new benchmark of 55 multi-intent queries over Fugaku job telemetry with 119 verifiable facts and 25 unanswerable sub-questions. The single agent recovers the most facts (84.9%) but refuses least. Pairing peer messaging with typed dialogue-act fields refuses most (92.0%) while matching single-agent recall within noise, at 2.5 times the tokens. Refusal follows whether agents exchange results at all; typing those exchanges recovers recall. Agent count explains neither. Recall and refusal must be designed for separately.- Author/Presenters
- Kartik ShindePredii Inc.
- Tirthankar GhosalOak Ridge National Laboratory
- Sreekantam Sai VenkatPredii Inc.
- Siddhant SenapatiPredii Inc.
- Sanjay DasOak Ridge National Laboratory
- Tilak KasturiPredii Inc.
- Khaled IbrahimLawrence Berkeley National Laboratory
- Hugo HernandezspinTwo LLC
- Kanad BasuRensselaer Polytechnic Institute
- Sukumar RanganAdvanced Micro Devices (AMD)
-
Workshop
EESP 2026: 3rd International Workshop on Energy Efficiency with Sustainable Performance: Techniques, Tools, and Best Practices
Full description
The Energy Efficiency and Sustainable Performance (EESP) workshop fosters the exchange of innovative strategies, tools, and best practices to enhance energy efficiency in modern computing environments. As energy costs rise and CO₂ emissions increase, sustainable operation has become critical for HPC, AI, and data center infrastructures. EESP addresses the challenge of balancing performance, power, and sustainability, focusing on actionable, cost-effective solutions for Tier-0/1 and Tier-2 systems. The workshop promotes cross-layer innovation by integrating hardware, software, runtime systems, and data-driven analytics to achieve end-to-end visibility and adaptive energy management. It advances transparency in energy use and carbon emissions, supports hardware innovations (e.g., new architectures and liquid cooling), and enables dynamic adaptation, predictive analytics, and intelligent workload scheduling for real-time optimization. By fostering collaboration between HPC and AI communities, EESP advances reproducible practices and operational insights that empower computing centers, researchers, and industry to implement scalable solutions for long-term sustainable performance. EESP Workshop Link- Session Chairs
- Ayesha AfzalFAU Erlangen-Nürnberg; Erlangen National High Performance Computing Center
- Natalie BatesEnergy Efficient High Performance Computing Working Group (EE HPC WG); Lawrence Berkeley National Laboratory (LBNL)
- Bronis R. de SupinskiLawrence Livermore National Laboratory (LLNL); Queen's University Belfast
- Hatem LtaiefKing Abdullah University of Science and Technology (KAUST)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
-
Workshop
High Performance Python for Science at Scale
Full description
This symposium-style workshop aims to connect researchers, developers, and Python practitioners to share their experiences scaling Python-based applications and workflows on supercomputers. The goal is to provide a platform for topical discussion of best practices, hands-on demonstrations, and community engagement via open-source contributions to new libraries, runtimes, and frameworks. Based on talks and demos that survey and summarize the best practices and recent success stories and developments, the workshop provides attendees a forum for expanding their knowledge of tools and techniques as well as opportunities to engage with representatives from the broader Python community.- Session Chairs
- Pete MendygralHPE
- Sunita ChandrasekaranUniversity of Delaware
- Sam ForemanArgonne National Laboratory
- Daniel MargalaLawrence Berkeley National Laboratory (LBNL)
- Davin PottsAppliomics
- Andy TerrelNVIDIA Corporation
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
-
Workshop
Enhancing Data Center Efficiency and Throughput by Co-Scheduling
Full description
Many HPC data centers have supercomputers with separate CPU-only and GPU compute nodes to support different workloads. One approach to increase system throughput is to run CPU-only jobs on unused CPU resources on the GPU nodes. Analysis of the workload running on NERSC's supercomputer showed that approximately 60% of Perlmutter's GPU node-hours employed fewer than 12 CPU cores and less than 128GB host memory. The remaining 52 CPU cores and 128GB host memory could be used by co-scheduled CPU-only jobs. However, it is critical that the co-scheduled CPU-only jobs and GPU jobs have controlled/bounded interference. This preliminary study assesses the feasibility of co-scheduling in production. Testing of the NERSC-10 benchmark applications on NVIDIA Grace-Hopper nodes using Memory System Resource Partitioning and Monitoring (MPAM) and network/InfiniBand quality of service for performance isolation measured a 70% throughput increase from co-scheduling compared with no co-scheduling, while maintaining an acceptable level of performance isolation.- Author/Presenters
- Paul LinNVIDIA
- Nestor DemeureNVIDIA
- Jonathan SkoneLawrence Berkeley National Laboratory
- Dhruva KulkarniLawrence Berkeley National Laboratory
- Mathias WagnerNVIDIA
- Francesca TartaglioneNVIDIA
- Jack DeslippeLawrence Berkeley National Laboratory
- Ian KarlinQueen's University
- Garima KochharNVIDIA
- Brian AustinLawrence Berkeley National Laboratory
-
Workshop
ORBIT: An AI-Powered Assistant for NOC Operations in a Data-Intensive Science Network
Full description
High-performance networks are critical infrastructure for data-intensive science, and their effective management relies on efficient incident response in Network Operations Centers (NOCs). This paper introduces the Operations Response \& Business Intelligence Toolkit (ORBIT), an AI-powered assistant that enhances incident response by consolidating information from disparate, siloed systems. ORBIT delivers synthesized results directly into the operator's primary work interface, applying agentic AI principles through a centralized reasoning service and a common model-access gateway. The system ensures transparency by logging all AI-generated outputs for operator review and rigorously controls access to sensitive data. In this paper, we describe the ORBIT architecture, present a selection of initial results from operational testing within ESnet’s NOC, and share lessons learned during its development and deployment; the complete findings are available in a companion technical report. Our findings highlight the importance of prompt engineering and a human-in-the-loop design for building trust and facilitating adoption.- Author/Presenters
- Bin DongEnergy Sciences Network (ESnet)
- Sukhada GholbaEnergy Sciences Network (ESnet)
- Brooklin GoreEnergy Sciences Network (ESnet)
- Shawn KwangEnergy Sciences Network (ESnet)
- David MitchellEnergy Sciences Network (ESnet)
- Samuel OehlertEnergy Sciences Network (ESnet)
- Garrett StewartEnergy Sciences Network (ESnet)
- Brendan WhiteEnergy Sciences Network (ESnet)
- Luke BakerEnergy Sciences Network (ESnet)
- Edward BalasEnergy Sciences Network (ESnet)
- Britt GathrightEnergy Sciences Network (ESnet)
- Chin GuokEnergy Sciences Network (ESnet)
- John Paul HerronEnergy Sciences Network (ESnet)
- John MaCauleyEnergy Sciences Network (ESnet)
- Scott RichmondEnergy Sciences Network (ESnet)
- Chris RobbEnergy Sciences Network (ESnet)
- Christopher TracyEnergy Sciences Network (ESnet)
- Kesheng (John) WuLawrence Berkeley National Laboratory (LBNL); ESnet
-
Workshop
Custodian: Enabling Automatic Performance Engineering Agents on the NERSC HPC System
Full description
The advancement of Large Language Models and the rapid evolution of AI assistants and agents create an opportunity for the development of highly specialized HPC agents to enable automatic performance analysis. In this paper, we present Custodian, an HPC performance analysis agent that analyzes previously collected job telemetry and application-output data to characterize application behavior, identify potential performance bottlenecks, and transform performance metrics into natural language descriptions and figures that suggest optimizations or highlight computing inefficiencies. We validate Custodian’s telemetry-derived application states and roofline analyses against NVIDIA Nsight Systems and Nsight Compute across four representative applications executed on NERSC Perlmutter. Finally, we discuss the tradeoffs, cost, productivity and systems applicability of HPC agents on the NERSC HPC system.- Author/Presenters
- AbdulHannan MohammedLawrence Berkeley National Laboratory
- Oscar AnteparaLawrence Berkeley National Laboratory
- Dhruva KulkarniLawrence Berkeley National Laboratory
- Muaaz Gul AwanLawrence Berkeley National Laboratory
- Samuel WilliamsLawrence Berkeley National Laboratory
- Jack DeslippeLawrence Berkeley National Laboratory
-
Workshop
SpSYRK: Half the Work in Distributed Sparse Matrix Multiplication
Full description
The symmetric rank-k update (SYRK), C = AA^⊤, computes the dot product between each pair of rows of A. Its sparse variant is central to similarity-based approaches in machine learning, graph analytics, and genomics. Existing distributed sparse matrix multiplication algorithms, such as Sparse SUMMA, treat sparse SYRK as a generic multiplication, leaving performance untapped. In this paper, we present a distributed sparse SYRK algorithm that leverages symmetry. The approach reduces per-process computation compared with state-of-the-art distributed SpGEMM. A second variant reorders communication to avoid materializing A^⊤ and reduce communication volume. The algorithm achieves a 2× speedup over an optimized Sparse SUMMA on matrices. Our variant achieves superior scaling at high process counts. The approach is a drop- in replacement for any application computing C = AA^⊤ via a distributed SpGEMM routine, and its triangular output can be consumed directly by subsequent operations with reduced compute and memory footprint.- Author/Presenters
- Thomas McFarlandCornell University
- Julian BellavitaCornell University
- Giulia GuidiCornell University; Lawrence Berkeley National Laboratory
-
Workshop
Cheaper by the Batch: Shared Traversal for Genotype Graph Editing
Full description
Adding nodes in a graph while reusing existing structure is a recurring computational problem. In population genetics, this problem arises in genotype representation graph (GRG), a DAG that losslessly encodes genetic variation across thousands of samples by sharing subgraph structure. In GRG, mutation’s carrier is encoded as leaf nodes reachable from the node it is assigned to. Current approaches update mutations individually. This paper introduces a batched mutation-remapping algorithm that replaces independent traversals with a shared reverse-topological pass, identifying reuse candidates for a batch at once. The pass propagates compact bit-parallel per-mutation state and uses an adaptive sparse/dense carrier set representation spanning rare-to-common variant densities. Batching adds memory-scalablility to split-based parallelism, which replicates graph and traversal state per worker. Our remapping approach is evaluated on standalone scenarios and end-to-end allele polarization workload. It is up to 10.5$\times$ faster than independent remapping while preserving exact carrier-set semantics.- Author/Presenters
- Aaron LiCornell University
- Yifan LiCornell University
- Drew DeHaasCornell University
- Giulia GuidiCornell University
-
Workshop
Panel: "Is Explicit Communication Obsolete? Compilers, Runtimes and On What Matters Next"
Full description
As domain-specific accelerators gain importance and workloads increasingly exhibit structured, predictable communication patterns—particularly in machine learning, scientific computing, and tensor operations—the role of explicit communication primitives deserves renewed scrutiny. This panel examines whether the compiler and runtime stack should bear greater responsibility for inferring and optimizing communication, potentially relegating explicit APIs to legacy status. The rise of Tile IRs and polyhedral compilation frameworks suggests a future where communication becomes an emergent property of program structure rather than a programmer directive. When tiling strategies, data dependencies, and memory hierarchies can be analyzed statically or just-in-time, do we still need developers to manually orchestrate point-to-point sends, broadcasts, and reductions? Or does the irregularity and dynamism of real applications—adaptive meshes, sparse data, load imbalance—ensure that explicit control remains essential? This discussion will explore the tension between programmer productivity and performance portability, examining where today's compilation technology succeeds in hiding communication complexity and where it fails. Panelists will debate whether the next decade belongs to smarter runtimes that auto-parallelize and auto-communicate, or whether the most demanding applications will always require the fine-grained control that only explicit communication provides.- Moderator
- Jan CieskoSandia National Laboratories
- Panelists
- Dirk Van EssendelftNational Energy Technology Laboratory
- John ShalfLawrence Berkeley National Laboratory
- Szilárd PállKTH Center for Scientific Computing (KCSC)
- Mathias JacquelinCerebras
- Alex AikenStanford University
-
Workshop
Beyond FLOPS and Kernels: Benchmarking Agentic Science and the Future Role of Python
Full description
Python has become the dominant language for scientific computing, AI, and high-performance computing workflows. As supercomputing enters an era increasingly shaped by foundation models, autonomous agents, and AI-guided scientific discovery, the community faces a fundamental question: what should we benchmark next? This panel brings together leaders from HPC, AI, and scientific software to discuss how benchmarks can continue to influence system architecture in the age of agentic workflows. Should future benchmarks measure isolated kernels and application performance, or end-to-end scientific outcomes? How do we evaluate systems that combine simulation, machine learning, retrieval, reasoning, and autonomous decision-making? And what role should Python play as the orchestration layer for increasingly heterogeneous and distributed computing environments? Panelists will explore whether Python can serve as the foundation for next-generation benchmarks and how benchmark design can guide the architecture of future AI-enabled supercomputers.- Session Chair
- Pete MendygralHPE
- Moderator
- Sunita ChandrasekaranUniversity of Delaware
- Panelists
- Michela TauferUniversity of Tennessee, Knoxville
- Glenn LockwoodVAST Data
- Rollin ThomasLawrence Berkeley National Laboratory
- Minjia ZhangUniversity of Illinois at Urbana-Champaign
-
Workshop
Panel Discussion and Q&A: Challenges in AI-enabled Q-HPC Integration
Full description
SC26 has not supplied a description for this event.
- Session Chair
- Katherine KlymkoNational Energy Research Scientific Computing Center (NERSC)
- Moderator
- Katherine KlymkoNational Energy Research Scientific Computing Center (NERSC)
- Panelists
- Daan CampsMicrosoft Corporation
- Yoonho ParkIBM Thomas J. Watson Research Center
- Kristen PudenzAtom Computing
- Pooja RaoNVIDIA
- Aleksander WennersteenPasqal
Monday, November 16
-
Tutorial
Performance Tuning of HPC and ML/AI Applications with the Roofline Model on GPUs, APUs, and CPUs
Full description
The Roofline performance model offers an insightful and intuitive method for extracting the key execution characteristics of HPC and ML/AI applications and comparing them against the performance bounds of modern CPUs and GPUs. Its abilities to abstract the complexity of memory hierarchies and identify the most profitable optimization techniques have made Roofline-based analysis increasingly popular in the HPC and ML/AI communities. Although different flavors of the Roofline model have been developed, there remains a need for a systematic methodology when applying them to analyze applications running on multicore and accelerated systems. The tutorial aims to bridge this gap on both CPUs and GPUs by both exposing the fundamental aspects behind different Roofline modeling principles as well as providing several practical use case scenarios that highlight their efficacy for performance and energy optimization as well as understanding performance portability. This tutorial presents a unique combination of instruction in Roofline by its creator, hands-on instruction in using Roofline within NVIDIA’s, AMD’s, and Intel’s production performance tools, and discussions of real-world Roofline use cases at NERSC, ALCF, and OLCF. The tutorial presenters have a long history of collaborating on the Roofline model and have presented several Roofline-based tutorials.- Presenters
- Samuel WilliamsLawrence Berkeley National Laboratory (LBNL)
- Neil MehtaLawrence Berkeley National Laboratory (LBNL)
- JaeHyuk KwackArgonne National Laboratory
- Nicholson KoukpaizanOak Ridge National Laboratory (ORNL)
- Magnus StrengertNVIDIA Corporation
- Gina SitaramanAdvanced Micro Devices (AMD)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TUT — Tutorial Reg Pass
-
Workshop
6th International Workshop on RESource DISaggregation in High Performance Computing (RESDIS)
Full description
Disaggregation is an emerging compute paradigm that splits existing monolithic servers into a number of consolidated single-resource pools that communicate over a fast interconnect. This model decouples individual hardware resources and enables the creation of logical compute platforms with flexible and dynamic hardware configurations. The concept of disaggregation is driven by various recent trends in computation. From an application perspective, the increasing importance of data analytics and machine learning workloads brings unprecedented need for memory capacity, which is in stark contrast with the growing imbalance in the peak compute-to-memory capacity ratio of traditional system board-based servers. At the hardware front, the proliferation of heterogeneous, special-purpose computing elements promotes the need for composable platforms, while the increasing maturity of optical interconnects elevates the prospects of distance independence in networking infrastructure. The workshop intends to explore various aspects of resource disaggregation, composability, and their implications for future HPC platforms.- Session Chairs
- Balazs GerofiIntel Corporation; RIKEN Center for Computational Science (R-CCS)
- Ian KarlinNVIDIA Corporation
- John ShalfLawrence Berkeley National Laboratory (LBNL)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
-
Workshop
8th International Workshop on Containers and New Orchestration Paradigms for Isolated Environments in HPC (CANOPIE-HPC)
Full description
The ongoing revolution enabled via containerization, virtualization, and new orchestration models has dramatically changed how applications and services are delivered and managed across the computing industry. This revolution has established a new ecosystem of tools and techniques with new, flexible and agile approaches, and continues to gain traction in the HPC community. In addition to HPC-optimized container runtimes, technologies like Kubernetes create a new set of opportunities and challenges. While adoption is growing, questions regarding best practices, foundational concepts, tools, and standards remain. Our goal is to promote the adoption of these tools and introspect the impact of this new ecosystem on HPC/AI use cases. This workshop serves as a key venue for presenting late-breaking research, sharing experiences and best practices, and fostering collaboration in this field. Our eighth workshop iteration will continue to emphasize real-world experiences and challenges in adopting and optimizing these new approaches for HPC and AI.- Session Chairs
- Alberto MadonnaETH Zürich; Swiss National Supercomputing Centre (CSCS)
- Claudia MisaleCoreWeave, Inc.
- Richard Shane CanonLawrence Berkeley National Laboratory (LBNL)
- Andrew YoungeSandia National Laboratories
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
-
Workshop
Opening Invited Talk: Tightly Coupled and Customizable Heterogeneity
Full description
The slowing of Moore’s Law and the rapid growth of AI are pushing systems toward greater specialization across compute, memory, networking, and storage. At the same time, AI for Science is bringing simulation, analytics, training, inference, and agentic execution into increasingly connected workflows that need to share and move data efficiently. No single architecture is well suited to every part of these workflows. This talk looks at tightly coupled and customizable heterogeneity as a way to build systems around these changing needs. A key focus will be macroheterogeneity, where different computing architectures work together as part of a larger system and each part of a workflow can run where it fits best. We will also look at customizable scale-up hypernodes, bringing compute, memory, accelerators, and high-bandwidth fabrics together in configurations shaped by workload needs. The goal is not simply to add more forms of heterogeneity, but to make them easier to couple, customize, and use as part of an end-to-end AI-HPC system.- Session Chairs
- Ian KarlinQueen’s University
- John ShalfLawrence Berkeley National Laboratory
- Balazs GerofiIntel Corporation; RIKEN Center for Computational Science (R-CCS)
- Presenter
- Samantika SuryHPE
-
Workshop
Enabling Performance Isolation for Co-scheduled Workflows
Full description
Heterogeneous systems such as NVIDIA GH200 and GB200 pair powerful GPUs with substantial CPU capacity, but GPU jobs targeting these systems frequently leave much of the CPU capacity idle, lowering CPU utilization and overall system throughput. Co-scheduling CPU workloads alongside GPU jobs improves resource utilization, but can introduce interference and undermine performance predictability. We present a proof of concept for co-scheduling CPU and GPU jobs on GH200 nodes with hardware-enforced resource isolation using Arm Memory System Resource Partitioning and Monitoring (MPAM) for CPU, L3-cache, and memory bandwidth and InfiniBand QoS for host network traffic. We augment existing resource isolation mechanisms in Slurm to allocate jobs to dedicated CPU and GPU "shards" with allocations for compute cores, L3-cache, memory bandwidth, and network bandwidth. This proof-of-concept implementation provides a practical framework for evaluating co-scheduled workloads with hardware-enforced resource isolation and establishes an operational model for multi-tenant co-scheduling on heterogeneous HPC/AI nodes.- Author/Presenters
- Daniel MargalaLBNL NERSC
- Jonathan SkoneLBNL NERSC
- Christopher SamuelLBNL NERSC
-
Workshop
RSE-HPC Panel: Meeting RSE Challenges
Full description
Hannah Cohoon, Lawrence Berkeley National Laboratory Thomas Flynn, Durham University Helen Kershaw, National Center for Atmospheric Research (NCAR)- Participants listed in source description
- Hannah CohoonLawrence Berkeley National Laboratory
- Thomas FlynnDurham University
- Helen KershawNational Center for Atmospheric Research (NCAR)
-
Workshop
Stress-Testing Exascale Scientific Data Output - ADIOS2 Configuration Strategies for WarpX under Extreme Load Imbalance
Full description
Spatially imbalanced, temporally irregular simulation output challenges I/O configurations tuned for balanced workloads. This work evaluates ADIOS2 BP5 configurations through openPMD-api using WarpX on Frontier at ORNL and Perlmutter at NERSC, contrasting balanced 3D Cartesian output with boosted-frame back-transformed diagnostics (BTD). BTD combines rank-level output skew with snapshot contributions spanning multiple simulation steps. Balanced output reaches 3.64~TB/s without compression on 2,048 Frontier nodes and over 1~TB/s on 512 Perlmutter nodes; lossless Blosc2 compression achieves approximately 7~TB/s application-level effective throughput on Frontier. In the tested BTD configurations, SPAN combined with asynchronous writing provides the highest throughput on Frontier, FlattenSteps supports assembly of complete lab-frame snapshots for checkpointing without reader modification and with negligible write-time overhead and no penalty on read performance, and DataSizeBased aggregation improves throughput when subfile counts are constrained. These results identify complementary mechanisms for handling irregular output and provide configuration guidance for these workloads and systems.- Author/Presenters
- Greg EisenhauerGeorgia Institute of Technology
- Junmin GuLawrence Berkeley National Laboratory
- Axel HueblLawrence Berkeley National Laboratory
- Scott KlaskyOak Ridge National Laboratory
- Norbert PodhorszkiOak Ridge National Laboratory
- Alexander SimLawrence Berkeley National Laboratory
- Jean-Luv VayLawrence Berkeley National Laboratory
- Kesheng (John) WuLawrence Berkeley National Laboratory (LBNL); ESnet; Lawrence Berkeley National Laboratory (LBNL)
-
Workshop
Fair, Random IO Operations are Hard
Full description
The IO500 Committee continuously works to improve and expand what the IO500 benchmarks cover to keep them both current and relevant. The latest change we have been working on is a random 4 KiB read phase. Our first attempt yielded some unexpected results prompting a deep dive and re-implementation from scratch. Where we are now requires community feedback to ensure the best path forward.- Author/Presenters
- Jay LofsteadSandia National Laboratories; University of New Mexico
- Dean HildebrandGoogle LLC
- Andreas DilgerThe Lustre Collective
- Jean Luca BezESnet; Lawrence Berkeley National Laboratory (LBNL)
- Julian KunkelGWDG
-
Early Career
Mentorship in Practice: Building Relationships for Early-Career Growth
Full description
Early-career professionals are often told to “find a mentor,” with little guidance on where to start or how to navigate the relationship once they have one. This panel brings together professionals with experience as both mentors and mentees to discuss how these relationships begin, what makes them useful, what to talk about, and how they can evolve over time. Panelists will also discuss the distinction between mentors and sponsors and share their thoughts on building a broader network of people who can offer guidance, perspective, and advocacy at different career stages. The moderated discussion will be followed by an open Q&A, where attendees are invited to bring their own questions and experiences into the conversation.- Moderator
- Melissa RomanusLawrence Berkeley National Laboratory (LBNL)
-
Workshop
Evaluation of LLVM Flang for Production HPC Applications and Modern Fortran Features
Full description
In 2025, LLVM released its first Flang Fortran compiler version considered ready for widespread evaluation. We know of no published assessment of Flang compiling a workload-derived portfolio of high-performance computing (HPC) applications. We address this gap using workload data from the National Energy Research Scientific Computing Center (NERSC), which supports more than 10,000 scientists on approximately 1,000 projects. We selected ten Fortran packages heavily used at NERSC with available source code and compiled them using recent versions of Flang on NERSC’s Perlmutter system. Seven compiled without code modifications, though some required build-system changes. Three compiled after minor source edits, mostly to address Fortran standard violations. We evaluated seven additional packages selected for their use or enablement of standard Fortran parallel features: multi-image execution and do concurrent. Six such codes compiled with most or all unit tests passing.- Author/Presenters
- Katherine RasmussenLawrence Berkeley National Laboratory (LBNL)
- Damian RousonLawrence Berkeley National Laboratory
- Dan BonacheaLawrence Berkeley National Laboratory
- Rahulkumar GayatriLawrence Berkeley National Laboratory
- Phillip ThomasLawrence Berkeley National Laboratory
- Brandon CookLawrence Berkeley National Laboratory
-
Workshop
WHPC Career Pathways
Full description
Three women working in industry, national laboratories, and academia will share their career journeys and perspectives on navigating professional transitions, mentorship, challenges, and opportunities. This session offers practical insights for students and early-career professionals exploring diverse pathways in careers within HPC.- Session Chair
- Jessica Imlau DagostiniUCSC
- Presenters
- Wendy WilhelmHPE
- Rebecca Hartman-BakerNERSC
- Fernanda FoertterUniversity of Alabama
-
Workshop
Evaluating OpenMP API Version 6.1 Target Kernel Language
Full description
The OpenMP API has steadily evolved into a comprehensive, highly portable alternative to vendor-specific APIs for GPU acceleration. The upcoming OpenMP 6.1 specification introduces two key features for code acceleration: access to dynamic shared memory and multidimensional, kernel-like programming. In this paper, we provide an overview of these newly standardized features for C++ and Fortran codebases. We implement a functional prototype based on the official LLVM infrastructure, extending both the Clang and Flang compilers to support these additions. We evaluate our implementation across several C++ and Fortran benchmarks, confirming that these new OpenMP capabilities effectively help close the performance and expressiveness gap with vendor-specific APIs. During the evaluation, we also observe a few remaining opportunities where OpenMP could become even more competitive for specific computational patterns.- Author/Presenters
- Kevin SalaLawrence Livermore National Laboratory
- Rahulkumar GayatriLawrence Berkeley National Laboratory
- Krishna Chaitanya SankisaAdvanced Micro Devices (AMD)
- Krzysztof ParzyszekAdvanced Micro Devices (AMD)
- Michael KlemmAdvanced Micro Devices (AMD)
- Johannes DoerfertLawrence Livermore National Laboratory
-
Workshop
Community Discussion and Q&A
Full description
SC26 has not supplied a description for this event.
- Session Chairs
- Alberto MadonnaETH Zurich / Swiss National Supercomputing Centre
- Claudia MisaleCoreWeave, Inc.
- Richard Shane CanonLawrence Berkeley National Laboratory
- Andrew YoungeSandia National Laboratories
-
Demo
IRI 2.0: A Common Language for DOE Research Infrastructure
Full description
The DOE's Integrated Research Infrastructure (IRI) provides consistent, machine-readable descriptions of facility resources, capabilities, access paths, and services. We'll show how these descriptions support both human-driven workflows and agentic systems that discover and interact with DOE research infrastructure.- Speakers
- John MacAuley (ESnet/LBL), Justas Balcas (ESnet/LBL)
-
Featured booth talk
ESnet's 40th Anniversary Celebration
Full description
The DOE's sole high-performance network for science, the Energy Sciences Network (ESnet), was formally created in 1986. On Nov. 16 at 7:30pm in the DOE Booth (#4829), ESnet Executive Director Inder Monga will give a brief presentation followed by lightning talks by DOE leadership and ESnet partners and staff, including Dylan Jacob, SCinet 2026 Chair and ESnet core routing and optical networking group lead. Come raise a toast to 40 years of accelerating discoveries in fusion, physics, cosmology, and more across the DOE complex. More info: www.es.net/40- Speakers
- Inder Monga, Ben Brown, others
-
Demo
AIDRIN: The Data Readiness Infrastructure
Full description
AIDRIN is open-source infrastructure for quantitatively assessing and improving the readiness of scientific datasets for intelligent learning systems. AIDRIN helps users to prepare, transform, and remediate to ready the datasets. AIDRIN offers multiple interfaces, including a command-line interface (CLI), web-based service, and MCP server with agentic interface.- Speakers
- Jean Luca Bez (LBNL), Suren Byna (LBNL)
-
Demo
ESnet on AWS: A Cloud Facility Behind the IRI API
Full description
We'll show ESnet AWS cloud capabilities exposed through the Integrated Research Infrastructure (IRI), demonstrating how to make cloud infrastructure usable together with other DOE facility resources. We present AWS-backed compute (batch-scheduled, container-execution, GPU-capable) and storage endpoints (plus a Globus DTN for data transfer) through the same DOE-IRI interface enabled at the ASCR facilities. This demo facility implements AWS support through IRI, and the same adapter pattern could use Google Cloud or Azure resources without changing the interface. The demo runs the identical client seamlessly across all ASCR facilities - NERSC, ALCF, OLCF, and ESnet on the AWS Cloud, switching only the facility endpoint while the researcher workflow remains the same. As the science workload, the Stanford Synchrotron Radiation Lightsource (SSRL) at SLAC National Accelerator Laboratory will run its containerized X-ray tomography and segmentation workflow on the AWS facility, submitted from the SSRL cross-facility workflow dashboard through the same IRI API calls it uses for NERSC and ALCF. SSRL is a Basic Energy Sciences user facility, so this is a beamline workload from outside ASCR running on the ASCR centers and the cloud facility from a single interface.- Speakers
- Justas Balcas (ESnet/LBL), John MacAuley (ESnet/LBL), Tim Dunn (SLAC)
Tuesday, November 17
-
Demo
Agentic Scientific AI at Scale: From Data to Insight with ClearML
Full description
From a browser session, a scientist directs an AI agent through a scientific AI campaign on NERSC systems. Using ClearML as a unified platform for data management, HPC job submission, monitoring, experiment tracking, and visualization, the agent prepares training runs, orchestrates hyperparameter tuning, follows their progress, compares results, and explains the findings. The demo shows how ClearML’s MCP integration can make the full workflow accessible to AI agents while keeping the scientist in charge of the research decisions.- Speakers
- Steven Farrell (LBNL), Shashank Subramanian (LBNL)
-
Posters
Faster Biobank-Scale Genomics Through Sparse Linear Algebra on Genotype Representation Graph
Full description
Biobank initiatives include half a million phased genomes, outpacing the tools available to analyze them. Conventional genotype formats treat variants independently and fail to exploit the shared ancestry structure underlying population genetic variation. Prior work introduced the genotype representation graph, a compact graph encoding that captures shared ancestry directly but relies on traversal algorithms that map poorly onto modern accelerators. Here, we show that this computation can be expressed exactly as sparse linear algebra: under a topological ordering, the graph’s adjacency matrix becomes strictly block triangular, so the core genotype matrix–vector product reduces to a block sparse triangular solve. Further, we decompose this solve into a pipelined sequence of blocked sparse matrix-vector multiplications, exposing fine-grained parallelism and enabling mature, hardware-optimized primitives in place of hand-tuned kernels. The result is substantial, portable speedups across accelerator hardware, from cloud deployments to supercomputers.- Authors
- Yifan LiCornell University
- Qingyao SunCornell University
- Drew DeHaasCornell University
- Xinzhu WeiCornell University
- Giulia GuidiCornell University
-
Posters
How the NERSC User Community Connects
Full description
High performance computing centers, such as NERSC, serve thousands of projects and users across many science domains and funding offices, yet staff have few systematic ways to see how these communities form, persist, and connect for planning and outreach. As an exploratory step toward surfacing actionable insights from allocation data, 2020–2026 Perlmutter allocation records are turned into yearly project and user networks, and community detection, weighted betweenness bridge detection, and liaison ranking. Null-model and matched-popularity tests confirm the structure is real and actionable: communities stay above the 95% null modularity band every year; about 10% of projects carry 66%–74% of cross-office connectivity; and the top 200 liaison candidates reach 69% of active users. To our knowledge, this is the first network analysis of an HPC facility’s community.- Authors
- Sumaiya AzadUniversity of Utah
- Camelia Daniela BrumarTufts University
- Coby SandholtzBrigham Young University
- Kevin GottLawrence Berkeley National Laboratory (LBNL)
- Rebecca Hartman-BakerLawrence Berkeley National Laboratory (LBNL)
-
Posters
Performance Analysis of Phaseless Auxiliary Field Quantum Monte Carlo
Full description
Predicting molecular behavior requires knowing a molecule's ground-state energy, a calculation whose exact cost grows exponentially with system size. Phaseless auxiliary-field quantum Monte Carlo (ph-AFQMC) approximates this energy via guided random walks, where accuracy and cost depend heavily on the quality of the trial wavefunction used to steer the walk. We compare two quantum methods for preparing this trial wavefunction, the Variational Quantum Eigensolver (VQE) and the Generative Quantum Eigensolver (GQE), benchmarked on NERSC Perlmutter using linear hydrogen chains. At H8 (16 qubits), GQE reduces end-to-end pipeline wall-time by roughly 5x relative to ADAPT-UCCSD, while matching VQE-family accuracy near equilibrium bond length. These results motivate extending the study to H10 chains and multi-node scaling to determine whether GQE's cost advantage persists at larger, more strongly correlated systems.- Authors
- Namita ShahUniversity of Michigan
- Neil MehtaLawrence Berkeley National Laboratory (LBNL)
- Katherine KlymkoLawrence Berkeley National Laboratory (LBNL)
- Ermal RrapajLawrence Berkeley National Laboratory (LBNL)
- Pooja RaoNVIDIA Corporation
- Nickolas JoynerLawrence Berkeley National Laboratory (LBNL)
- Lisa ClausLawrence Berkeley National Laboratory (LBNL)
- Charles LivelyLawrence Berkeley National Laboratory (LBNL)
-
Posters
Sparse Matrix-Vector Multiplication on Cerebras WSE-3
Full description
Sparse matrix-vector multiplication (SpMV) is bounded by data movement rather than arithmetic, since every nonzero fetches from an unpredictable address. We present a general tile-wise SpMV kernel for the Cerebras WSE-3, whose ~893,000 processing elements (PE) hold the working set in on-chip SRAM. The kernel partitions the matrix with a deterministic host-side balancing algorithm and sums per-PE partials eastward along each grid row. Sweeping to the full wafer, the local compute scales linearly and a two-parameter model predicts it exactly across 17 configurations; the eastward reduction, parallel along one axis only, is the sole limit on total scaling, and a standalone chain-latency model underpredicts it by up to 26% once it overlaps compute. On a biobank-scale genotype matrix of 5.28 billion nonzeros the kernel sustains 64,700 GFLOP/s; on a matched smaller instance it completes in 59.6 μs against 5.46 ms for four A100 GPUs.- Authors
- Zenchang SunCornell University
- Arman MargarianCornell University
- Jefferson ZhouCornell University
- Giulia GuidiCornell University
-
Posters
Streaming Nuclear Physics at Scale
Full description
Nuclear physics experiments have long challenged the limits of high-rate and low-latency logic. Presently, streaming readout can move decision making from fixed detector electronics to software, allowing raw detector data to be processed continuously as they are produced. We highlight the state of Jefferson Lab and ESnet collaborative work with example workflows. First, existing data from the CLAS12 detector were replayed across ESnet through the EJFAT network load balancer to the Perlmutter supercomputer at NERSC, reaching 100 Gb/s as jobs dynamically scaled the workflow. Second, live data direct from the CLAS12 electromagnetic calorimeter were streamed, showing end-to-end capabilities and real-time cosmic ray identification. These demonstrations go beyond earlier transport benchmarks to physics-based streaming workflows, thus combining high-rate data transport with online scientific processing. We also discuss applying this approach to Jefferson Lab’s development of streaming-readout modules for positron emission tomography (PET) processing, enabling future whole-area, whole-body scanning otherwise considered impractical.- Authors
- Jonathan ZarlingJefferson Lab
- Vardan GyurjyanJefferson Lab
- David LawrenceJefferson Lab
- Derek HowardEnergy Sciences Network (ESnet)
- Yatish KumarEnergy Sciences Network (ESnet)
- Stacey SheldonEnergy Sciences NetworkLab
- David AbbottJefferson Lab
- Benjamin RaydoJefferson Lab
-
Posters
Trustworthy Agentic AI for HPC Operational and Predictive Data Analytics and Forecasting
Full description
Large language models (LLMs) and agentic systems offer promising pathways for automating high performance computing (HPC) operations, yet their real-world deployment is severely constrained by silent operational failures, uncalibrated confidence, and brittle central control. In this work, we chart a new vision for reliable, agentic AI across supercomputing facilities. We explore the architectural paradigm shift from centralized orchestration to decentralized, peer-to-peer agent topologies, illustrating how typed dialogue and bounded knowledge limits ensure defensible, transparent operational guidance. Furthermore, we examine the critical challenge of stance fidelity, demonstrating how verbalized confidence must be audited as dynamic system telemetry rather than taken at face value. Based on these insights, we advocate for a fundamental re-architecting of facility AI: prioritizing honest refusal over raw recall, establishing auditable ground-truth benchmarks, and transitioning AI tools from passive historical log lookups to active, pre-submission job steering.- Authors
- Kartik ShindePredii
- Tirthankar GhosalOak Ridge National Laboratory (ORNL)
- Sreekantam Sai VenkatPredii
- Siddhant SenapatiPredii
- Sanjay DasOak Ridge National Laboratory (ORNL)
- Tilak KasturiPredii
- Khaled IbrahimLawrence Berkeley National Laboratory (LBNL)
- Hugo HernandezspinTwo
- Kanad BasuRensselaer Polytechnic Institute
- Sukumar RanganAdvanced Micro Devices (AMD)
-
Posters
Understanding the NERSC User Community Through Network Analysis
Full description
NERSC users collaborate, forming a network. Which users occupy the most structurally influential positions in the NERSC user community? Using 2020–2026 NERSC allocation records, a proxy collaboration graph was analyzed to reveal the user community’s dynamics. Two definitions of user connectivity were compared using available data on project allocation size: unweighted degree and repository-size-adjusted (RSA) strength. Degree distributions, program-office affiliation, and user roles were analyzed. Results show that RSA strength is a more accurate measurement of user connectivity than unweighted degree. The strongest connectors in the network are career scientists and users associated with high-energy physics. These findings suggest that in HPC network data, unweighted connectivity is not a reliable measurement of network influence or cross-disciplinary reach. Findings also suggest that allocation networks reflect certain field-specific collaboration patterns, implying that allocation networks may provide a useful proxy for co-authorship networks in the same field.- Authors
- Coby SandholtzBrigham Young University
- Camelia Daniela BrumarTufts University
- Sumaiya AzadUniversity of Utah
- Kevin GottLawrence Berkeley National Laboratory (LBNL)
- Rebecca Hartman-BakerLawrence Berkeley National Laboratory (LBNL)
-
Featured Booth Talk
Preparing NERSC for Quantum Computing
Full description
The National Energy Research Scientific Computing Center (NERSC) actively provides quantum hardware resources to the scientific community through the Quantum User Expansion for Science and Technology (QUEST) program. This initiative is explicitly designed to connect researchers with state-of-the-art commercial quantum hardware and user support while developing effective support structures for next-generation systems. This talk will describe NERSC's evolving user model, which actively bridges the gap between traditional scientific computing communities and emerging quantum resources. Through strategic private partnerships with IBM and QuEra Computing, NERSC explores distinct modes of user engagement, training, and resource allocation across diverse quantum architectures. Additionally, NERSC supports access to on-prem hardware through collaborations with the Advanced Quantum Testbed and IQM. By detailing these access pathways, our presentation will offer concrete perspectives on facility vision, user models, and ecosystem development to help shape the roadmap for future DOE Office of Science investments in quantum information science.- Speakers
- Katie Klymko
-
Roundtable
IRI Developer & User Community Work Session
Full description
Join an open working session for Integrated Research Infrastructure developers, facility teams, and users to discuss current implementations, integration challenges, priorities, and future capabilities. Bring questions, use cases, and feedback to help shape the next steps for IRI.- Speakers
- John MacAuley (ESnet/LBL), Paul Rich (ANL)
-
Invited Talk
Sparse Linear Algebra Accelerates Genotype Representation Graph Computation at Biobank Scale
Full description
Biobank-scale genomic analyses are increasingly constrained by computational costs, as hundreds of thousands to millions of samples and variants must be analyzed together. The genotype representation graph (GRG) compactly encodes population genetic variation to accelerate computation, but the current approach does not exploit modern accelerator architectures. This work introduces Mikado, a new methodology for expressing GRG-based computation using sparse linear algebra primitives. Under a reverse topological ordering of the graph nodes, the GRG adjacency matrix is strictly block-lower-triangular, and the genotype matrix-vector product becomes a sparse triangular solve that can be further decomposed into a pipelined sequence of blocked sparse matrix-vector multiplies. By decoupling computation from graph representation, our approach exposes fine-grained parallelism and enables hardware-optimized sparse primitives on GPUs. Mikado was evaluated across the Perlmutter supercomputer at NERSC and Google Cloud. Mikado achieves an order-of-magnitude speedup for PCA and BOLT-LMM compared with the original GRG traversal approach, including on All of Us Research Program cohorts, demonstrating that advanced computing infrastructure is now central to progress in biological research. Portability across supercomputer nodes and cloud instances makes these gains broadly accessible, providing a scalable, hardware-portable, researcher-friendly tool for population genetics at biobank scale.- Presenter
- Giulia GuidiDepartment of Computer Science, Cornell University
-
Birds of a Feather
Data Readiness for AI in HPC: Building a Community Around Dimensions, Metrics, and Tools
Full description
AI now drives discovery across HPC science, yet models are only as trustworthy as the data they use. Many teams still lack a shared way to judge whether their data is ready for AI. This BoF brings together the HPC and scientific communities to define data readiness for AI using seven practical dimensions: quality, understandability, fair and unbiased representation, structure and organization, feature importance, governance covering ethics, security and privacy, and AI-specific needs such as labels and distributions. Through short talks and open discussion, we will compare practices, identify gaps, and begin building shared metrics, tools, and benchmarks.- Session Leader
- Suren BynaOhio State University; Lawrence Berkeley National Laboratory (LBNL)
- Additional Session Leaders
- Jean Luca BezLawrence Berkeley National Laboratory (LBNL)
- Wes BrewerOak Ridge National Laboratory (ORNL)
- Shreyas CholiaLawrence Berkeley National Laboratory (LBNL)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
-
Birds of a Feather
Open-Source Hardware Tools: Advancing Architecture Research and Chip Prototyping
Full description
As demand for specialized architectures grows in the post-Moore era, this BoF will foster a sustained community around chip prototyping using open-source hardware tools. Building on themes explored in previous editions of this BoF at SC24 and SC25, the session will focus on practical tool experiences, current gaps, domain-specific accelerators, AI-assisted design workflows, and chip prototyping as an emerging HPC workload. Invited speakers from architecture, open-source EDA, accelerator design, and HPC will seed discussion. Expected outcomes include actionable insights, stronger collaboration, and follow-up community initiatives to advance open-source chip prototyping.- Session Leader
- Kazutomo YoshiiArgonne National Laboratory
- Additional Session Leaders
- Kentaro SanoRIKEN Center for Computational Science (R-CCS)
- John ShalfLawrence Berkeley National Laboratory (LBNL)
- Antonino TumeoPacific Northwest National Laboratory (PNNL)
- Junichiro KadomotoThe University of Tokyo
- Franz FranchettiCarnegie Mellon University
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
-
Birds of a Feather
Workflows Community: Next-Generation Workflows for HPC — AI, Agentic Systems, and Scientific Discovery
Full description
This BoF will convene the workflows community to discuss the challenges of next-generation scientific workflows. A central focus will be on how the role of the human changes as workflows become increasingly autonomous: where people remain in the loop, how they steer and validate agentic systems, and how our community can support workflow developers. The session will bring together workflow users, developers, theorists, and industrial collaborators, highlighting emerging research directions and practical needs from scientific campaigns. Through interactive discussions, participants will explore collaborative strategies to enable resilient, scalable, and adaptive workflows that meet the evolving demands of scientific discovery.- Session Leader
- Stephen HudsonArgonne National Laboratory
- Additional Session Leaders
- Kyle ChardUniversity of Chicago; Argonne National Laboratory
- Rafael Ferreira da SilvaOak Ridge National Laboratory (ORNL)
- Daniela CassolLawrence Berkeley National Laboratory
- Woong ShinOak Ridge National Laboratory
- Mathew CherukaraArgonne National Laboratory
- Mathew SinclairArgonne National Laboratory
- Logan WardNVIDIA
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
-
Demo
AI-assisted Operations of Data Acquisition Systems for Neutrino Experiments
Full description
This demonstration presents an AI Shifter for the NOvA Data Acquisition system deployed on NERSC resources. It combines real-time anomaly detection with an LLM-based solution engine grounded in experiment documentation and expert logbooks, helping operators diagnose problems and identify mitigation actions through natural-language interaction. LLM inference runs on Perlmutter compute nodes, while frontend and orchestration services run on Spin. The approach aims to improve data quality, reduce detector downtime, and provide a foundation for AI-assisted operations in DUNE and other high-energy physics experiments.- Speakers
- Pengfei Ding (LBNL), Jianming Bian (LBNL)
-
Roundtable
EJFAT Streaming Technology and User Feedback
Full description
The ESnet-JLab FPGA Accelerated Transport, or EJFAT (pronounced “Edge-Fat”) prototype was designed to seamlessly integrate edge and cluster computing in order to allow data from multiple types of scientific instruments to be streamed and processed in near real time by multiple HPC facilities — and, if needed, to redirect those data streams dynamically. EJFAT has progressed significantly since SC25 in terms of users and use cases. This is an opportunity to get updates, provide the team input and engage with new ideas and use cases.- Speakers
- Yatish Kumar (ESnet/LBL), Eli Dart (ESnet/LBNL)
-
Paper
Tackling Parallelization Challenges of Randomized Preconditioners with Dependency Tracking
Full description
Incomplete Cholesky (IC) is a common memory-efficient preconditioner for large graph Laplacian systems, but classical IC can break down on irregular graphs. Approximate Cholesky (AC) addresses this with randomized dropping that is provably breakdown-free. Its stochastic fill-in pattern, however, resists static parallelization, creating a "setup bottleneck." We present ParAC, which parallelizes the AC sampling rule of Gao, Kyng, and Spielman using dynamic dependency tracking and architecture-aware memory management to execute it efficiently on CPUs and GPUs while preserving its numerical updates. Total time-to-solution speedups over deterministic IC reach 57.3× on CPU and 52.2× on GPU. Algebraic multigrid (AMG) generally remains preferable on structured PDE discretizations. On the irregular graph Laplacians tested, ParAC is up to 7.2× faster than AMG on CPUs and 6.0× faster on GPUs. On 200 heterogeneous SDDM systems, ParAC converges on every instance, whereas the AMG solvers fail on 15 and 19, on CPU and GPU, respectively.- Authors
- Tianyu LiangUniversity of California, Berkeley
- Chao ChenNorth Carolina State University
- Yotam YanivLawrence Berkeley National Laboratory (LBNL)
- Hengrui LuoRice University
- David TenchLawrence Berkeley National Laboratory (LBNL)
- Xiaoye LiLawrence Berkeley National Laboratory (LBNL)
- Aydın BuluçLawrence Berkeley National Laboratory (LBNL)
- James DemmelUniversity of California, Berkeley
- Jie ChenAmazon Web Services (AWS)
-
SCinet Network Research Exhibition
ORBIT: An AI-Powered Assistant for NOC Operations in a Data-Intensive Science Network
Full description
ESnet shares its pragmatic journey developing AI tools that deliver measurable improvements to Network Operations Center efficiency. Rather than chasing trends, ESnet focused on AI-assisted ticket resolution. This presentation details two concrete applications — Intelligent Resolution Guidance that proactively surfaces relevant documentation, and Adaptive Incident Summarization that generates audience-tailored ticket summaries. Through systematic stakeholder engagement and incremental development, ESnet demonstrates how responsible AI implementation can genuinely augment operational capabilities. Attendees will gain practical insights on balancing AI capabilities with limitations, ensuring tools address real workflow pain points while integrating seamlessly into existing processes, and learn from both successes and challenges encountered.- Presenter
- John WuESnet
-
Birds of a Feather
Confidential Supercomputing: Essential Evolution or Elegant Illusion?
Full description
Sensitive workloads across health sciences, financial modeling, and AI/ML on proprietary data increasingly demand confidentiality guarantees traditionally out of scope for supercomputing infrastructure. Confidential computing—hardware-enforced trusted execution environments established in cloud and edge contexts—is proposed as a path to securing computation at HPC scale. But is this achievable, or does it conflict with the principles that make supercomputers fast? This BoF gathers the HPC community to debate whether confidential supercomputing is a necessary, achievable evolution or an aspirational posture satisfying compliance optics without meaningful security. Join us to map where it genuinely delivers—and where gaps remain.- Session Leader
- Sadaf AlamUniversity of Bristol; Bristol Centre for Supercomputing
- Additional Session Leaders
- Hai Ah NamLawrence Berkeley National Laboratory (LBNL)
- Albert ReutherMIT Lincoln Laboratory
- Nathaniel McCallumAdvanced Micro Devices (AMD)
- Nigel EdwardsHPE
- CJ NewburnNVIDIA Corporation
- Katerina AntypasNational Science Foundation (NSF)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
-
Birds of a Feather
Ecosystems for Scientific Computing in the Age of AI
Full description
AI is rapidly reshaping scientific software through code generation, simulation acceleration, intelligent workflows, and emerging agentic systems. Yet the software ecosystems, collaboration models, trust frameworks, and workforce practices that underpin high performance computing (HPC) were built for a different era. Drawing on findings from a recent workshop series (https://arxiv.org/abs/2510.03413), this BoF invites the international community to challenge assumptions, identify emerging priorities, and help define the next generation of scientific computing ecosystems. Discussion will focus on human-AI collaboration, trust, sustainability, user experience, and workforce evolution — turning shared insights into actionable directions for the community.- Session Leader
- Lois Curfman McInnesArgonne National Laboratory
- Additional Session Leaders
- Anshu DubeyRIKEN Center for Computational Science (R-CCS)
- Dorian ArnoldEmory University
- Prasanna BalaprakashPrimaLabs
- Mike BernhardtTeam Libra LLC
- Thomas HeraultFrench Institute for Research in Computer Science and Automation (INRIA)
- Keita TeranishiOak Ridge National Laboratory (ORNL)
- Mary Ann LeungSustainable Horizons Institute
- Olivia NewtonUniversity of Montana
- Stefan WildLawrence Berkeley National Laboratory (LBNL)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
Wednesday, November 18
-
SCinet Network Research Exhibition
Customized End-to-End Services for Science Workflows via SENSE
Full description
nri114- Primary contact
- Tom LehmanEnergy Sciences Network (ESnet)
- Contributors
- Xi YangEnergy Sciences Network (ESnet)
- Justas BalcasEnergy Sciences Network (ESnet)
-
Birds of a Feather
Applications, Libraries, and Tools in Modern Fortran (alt.fortran)
Full description
Fortran plays a crucial role in numerous applications. This BoF provides a forum for Fortran developers to engage with the language's modern programming features. With features introduced in recent language revisions, Fortran 2023 supports modern programming practices and high performance computing (HPC). This BoF gathers developers from diverse domains to share experiences and explore Fortran's evolving capabilities. After some brief presentations, the session will focus on an interactive discussion where audience members will be encouraged to share their own experiences and ask questions of our experts.- Session Leader
- Katherine RasmussenLawrence Berkeley National Laboratory (LBNL)
- Additional Session Leaders
- Christopher MaynardMet Office
- Thomas CluneNASA Goddard Space Flight Center
- Liam PattinsonUniversity of York
- Wael ElwasifOak Ridge National Laboratory (ORNL)
- Manuel ArenazCodee; University of A Coruña (UDC)
- Heidi PoxonNVIDIA Corporation
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
-
Birds of a Feather
IO500: The High-Performance Storage Community
Full description
The IO500 is the de facto benchmark standard for HPC storage. Official IO500 lists have been released at ISC and SC events since SC17 and now have over 300 entries. The purpose is to foster the high-performance storage community and ensure forward progress towards the common goals of creating, sharing, and benefiting from a large corpus of shared storage data. IO500 also serves as the largest repository of detailed HPC storage information for researchers and system designers to analyze and evaluate over time. A key BoF highlight is the presentation of the latest IO500 Production and Research lists.- Session Leader
- Andreas DilgerThe Lustre Collective
- Additional Session Leaders
- Jean Luca BezLawrence Berkeley National Laboratory (LBNL)
- Dean HildebrandGoogle LLC
- Julian KunkelGeorg-August-Universität Göttingen; GWDG
- Jay LofsteadSandia National Laboratories
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
-
Birds of a Feather
Navigating Complexity: Achieving Performance Portability in the Evolving Landscape of Accelerated HPC Systems
Full description
With the increasing demand for AI in HPC, there has been a rapid rise in accelerated architectures, portable programming models, and frameworks. The already-daunting task of programming for interconnected accelerated systems has become even more complex. This BoF, organized by IXPUG, will focus on portable programming across a wide range of heterogeneous architectures—including Intel, NVIDIA, AMD, and ARM supporting diverse simulation, data analytics, and AI workloads. The session will explore key challenges, state-of-the-art solutions, and emerging best practices for programming across these systems, identifying common principles and methodologies that support development and long-term maintenance across sites, architectures, and scientific applications.- Session Leader
- Amit RuhelaTexas Advanced Computing Center (TACC); University of Texas
- Additional Session Leaders
- David MartinLawrence Berkeley National Laboratory (LBNL)
- Steffen ChristgauZuse Institute Berlin
- Hatem LtaiefKing Abdullah University of Science and Technology (KAUST)
- Clayton HughesSandia National Laboratories
- Toshihiro HanawaThe University of Tokyo
- Nalini NaliniIntel Corporation
- Neil MehtaLawrence Berkeley National Laboratory (LBNL)
- Peter DindaNorthwestern University
- Cristina BeldicaIntel Corporation
- Mathias PuetzNextSilicon Inc
- Ilan TayariNextSilicon Inc
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
-
Birds of a Feather
OpenQSE: Building an Open Quantum-HPC Software Ecosystem
Full description
Quantum computing resources are increasingly integrated into HPC and cloud environments, but software stacks remain isolated, vendor-specific, and often proprietary. OpenQSE is a community effort developing an open, modular ecosystem of vendor-neutral specifications to close the interoperability gaps that fragment hybrid quantum-classical workflows. This BoF will briefly share the initiative's roadmap and working-group progress, then open the floor for structured discussion with HPC facility staff, application teams, vendors, and researchers to identify priorities and shape the direction together. The intended outcome is a prioritized set of integration artifacts the community identifies as most urgent for near-term development.- Session Leader
- Thomas NaughtonOak Ridge National Laboratory (ORNL)
- Additional Session Leaders
- Amir ShehataOak Ridge National Laboratory (ORNL)
- Sebastian SternAmazon Web Services (AWS)
- Yoonho ParkIBM
- Michael FergusonHPE
- Lukas BurgholzerMunich Quantum Software Company (MQSC); Technical University of Munich
- Ermal RrapajLawrence Berkeley National Laboratory (LBNL)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
-
Birds of a Feather
Operational Data Analytics: Agentic AI for HPC Facilities
Full description
Operational data analytics (ODA) has become essential for understanding and optimizing HPC systems. However, converting the vast amounts of operational data they generate into actionable insights still requires significant manual effort and domain expertise. Agentic AI has the potential to extend traditional analytics by reading telemetry, correlating signals, diagnosing problems, and drafting fixes for an operator to approve. This BoF will explore the state of the art in agentic AI’s applicability to ODA, promising use cases, limitations, and governance challenges. Participants will share experiences and discuss how agentic AI can enhance future HPC operations and efficiency.- Session Leader
- Dong DaiUniversity of Delaware
- Additional Session Leaders
- Kevin MenearNational Laboratory of the Rockies (NLR)
- Avisek NaugHPE
- Keiji YamamotoRIKEN Center for Computational Science (R-CCS)
- Steven FarrellLawrence Berkeley National Laboratory (LBNL)
- Jeff HansonHPE
- Kathleen ShogaLawrence Livermore National Laboratory (LLNL)
- Melissa RomanusLawrence Berkeley National Laboratory (LBNL)
- Michael OttLeibniz Supercomputing Centre (LRZ); Energy Efficient High Performance Computing Working Group (EE HPC WG)
- Nil MuArizona State University
- Paul BrunkUniversity of Georgia
- Tim OsborneOak Ridge National Laboratory (ORNL)
- Woong ShinOak Ridge National Laboratory (ORNL)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
-
Birds of a Feather
Talking to Your Users: Ask Better Questions Before You Build
Full description
User experience (UX) is often misunderstood as a nice-to-have, rather than a necessary element of scientific computing. In reality, UX encompasses all the ways a user interacts with a product and is highly relevant to HPC systems. This interactive BoF will introduce UX work as a practical approach for improving usability and adoption. Participants will discuss how to ask users better questions to uncover needs and identify opportunities to improve systems, services, and support. Participants will leave with practical methods, starter resources, and connections to a growing cross-institutional UX community. This session benefits developers, documentation specialists, and facility support staff.- Session Leader
- Beth CernyArgonne National Laboratory
- Additional Session Leaders
- Johanna CohoonLawrence Berkeley National Laboratory (LBNL)
- Sarah ReadPortland State University
- Jordan FrithClemson University
- Charles Lively III,Lawrence Berkeley National Laboratory (LBNL)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
-
Demo
IRI 2.0: A Common Language for DOE Research Infrastructure
Full description
The DOE's Integrated Research Infrastructure (IRI) provides consistent, machine-readable descriptions of facility resources, capabilities, access paths, and services. We'll show how these descriptions support both human-driven workflows and agentic systems that discover and interact with DOE research infrastructure.- Speakers
- John MacAuley (ESnet/LBL), Justas Balcas (ESnet/LBL)
-
Roundtable
Streaming Workflows to ASCR Supercomputers
Full description
ASCR compute facilities are often the desired end points for EJFAT (for ESnet-JLab FPGA Accelerated Transport) workflows. Real-time compute brings unique requirements on availability, scheduling and run control integration. Join this discussion to help the EJFAT team think through and shape the way we access resources in these unique high capacity environments.- Speakers
- Yatish Kumar (ESnet/LBL), Eli Dart (ESnet/LBNL)
-
SCinet Network Research Exhibition
StarLight Agentic AIDTN for High-Performance Data/Process Transport with Research Platforms
Full description
NRI103- Primary contacts
- Jim Hao ChenNorthwestern University; StarLight
- Chun-Hao YangNational Yang Ming Chiao Tung University; Northwestern University
- Contributors
- Fei YehNorthwestern University
- Joe MambrettiNorthwestern University
- Se-Young YuEnergy Sciences Network (ESnet)
- Ting-Yu LinNCHC/NIAR
- Meng-Hsun TsaiNational Yang Ming Chiao Tung University; NCHC/NIAR
- Marc LyonnaisCiena
- Gaurauvdeep ShamiCiena
-
Panel
The Future of the HPC Workforce: Skills, Scale, and Sustainability
Full description
High performance computing (HPC) is undergoing rapid transformation with the convergence of simulation, AI, data-intensive science, and emerging quantum computing. This evolution is reshaping the skills required to effectively use and advance these systems, creating urgent challenges in workforce development. This panel brings together perspectives from national laboratories, academia, and industry to discuss how to build a sustainable, inclusive, and future-ready HPC workforce. Key questions include: What will the HPC workforce need to look like in the next 5–10 years? Where are the most critical gaps in skills, training pathways, and access? What skills are currently missing or undervalued in existing curricula? Which collaboration models across institutions have proven effective? How can we broaden participation and lower barriers to entry? The session will include interactive audience engagement, with opportunities to submit and prioritize questions in real time, ensuring a dynamic and relevant discussion.- Moderator
- Patrick DiehlLos Alamos National Laboratory (LANL)
- Panelists
- Rebecca Hartman-BakerNational Energy Research Scientific Computing Center (NERSC)
- Scot HalversonNVIDIA Corporation
- Richard KnepperCornell University
- Martin SchulzTechnical University of Munich
- Benedikt von St. ViethForschungszentrum Jülich
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
-
Demo
ESnet on AWS: A Cloud Facility Behind the IRI API
Full description
We'll show ESnet AWS cloud capabilities exposed through the Integrated Research Infrastructure (IRI), demonstrating how to make cloud infrastructure usable together with other DOE facility resources. We present AWS-backed compute (batch-scheduled, container-execution, GPU-capable) and storage endpoints (plus a Globus DTN for data transfer) through the same DOE-IRI interface enabled at the ASCR facilities. This demo facility implements AWS support through IRI, and the same adapter pattern could use Google Cloud or Azure resources without changing the interface. The demo runs the identical client seamlessly across all ASCR facilities - NERSC, ALCF, OLCF, and ESnet on the AWS Cloud, switching only the facility endpoint while the researcher workflow remains the same. As the science workload, the Stanford Synchrotron Radiation Lightsource (SSRL) at SLAC National Accelerator Laboratory will run its containerized X-ray tomography and segmentation workflow on the AWS facility, submitted from the SSRL cross-facility workflow dashboard through the same IRI API calls it uses for NERSC and ALCF. SSRL is a Basic Energy Sciences user facility, so this is a beamline workload from outside ASCR running on the ASCR centers and the cloud facility from a single interface.- Speakers
- Justas Balcas (ESnet/LBL), John MacAuley (ESnet/LBL), Tim Dunn (SLAC)
-
Birds of a Feather
AI and ML for HPC Workload Analysis: From Isolation to Reproducibility — Building Common Benchmarks
Full description
HPC facilities each build operational data analytics in isolation, with no shared benchmarks to compare methods or reproduce results across sites. This seventh-edition BoF convenes the community around a common goal: shared standards, open benchmarks, and cross-facility collaboration for HPC workload analysis. Using the HPC ODA Commons, introduced to the community at ISC 2026, as a working exemplar, we examine how standardized schemas and benchmark recipes enable reproducible comparison across systems. We present an emerging multi-facility study, invite sites to contribute data, schemas, or models, and work together to prioritize analyses and plan a community report.- Session Leader
- Kadidia KonateLawrence Berkeley National Laboratory (LBNL)
- Additional Session Leaders
- Sergio IserteBarcelona Supercomputing Center (BSC)
- Kevin MenearNational Laboratory of the Rockies (NLR)
- Estela SuarezJülich Supercomputing Centre; University of Bonn
- Ahmad Maroof KarimiOak Ridge National Laboratory (ORNL)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
-
Birds of a Feather
Quantum Jumps: From NISQ to FTQC to Hybrid HPC
Full description
Quantum computing is experiencing accelerating transitions, from noisy intermediate-scale quantum (NISQ) devices to fault-tolerant quantum computers, and from quantum computers being objects of research to their becoming scientific computation tools. Quantum programming habits and assumptions must be revisited, and for the HPC community, this is more than an incremental hardware upgrade; it affects entire stacks. This session will explore how compilation workflows, execution models, and algorithmic strategies are expected to change to incorporate error correction and integrate with production HPC workflows. Thought leaders from the quantum industry, HPC, and academic research will exchange perspectives and engage the audience in discussion.- Session Leader
- Kevin KissellAlice & Bob
- Additional Session Leaders
- Dave BaconGoogle Quantum
- Katherine KlymkoLawrence Berkeley National Laboratory (LBNL)
- Matthew OttenUniversity of Wisconsin-Madison
- Aaron LottHPE; University of Maryland, College Park
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
Thursday, November 19
-
Paper
Fault Injection and Resilience Across Emerging Platforms
Full description
This session studies resilience across quantum, GPU, and systolic platforms. It combines transient-fault mitigation, cross-architecture injection, operational characterization, and radiation-induced error analysis.- Session Chair
- David MartinLawrence Berkeley National Laboratory (LBNL); Energy Sciences Network (ESnet)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
-
Panel
HPC–Quantum Integration: Software, Systems, and Hybrid Workflows
Full description
High performance computing (HPC) and quantum computing are coming together to enable hybrid workflows that combine classical and quantum resources within a single computing environment. As quantum technologies evolve at different paces, integrating them into HPC systems introduces challenges for software, hardware, and scientific applications. This panel brings together perspectives from HPC centers, quantum hardware development, and academic research to examine how HPC–quantum integration is approached in practice. Panelists will discuss software and systems layers enabling hybrid execution, including programming models, workflow orchestration, and middleware connecting HPC environments to quantum resources. Drawing on early integration efforts, the discussion will highlight how communities are engaging with hybrid workflows — from deploying and evaluating quantum resources within HPC facilities to developing hardware and exploring algorithms and use cases. Panelists will share insights into challenges such as latency, resource management, as well as differing expectations around performance, usability, and readiness across quantum paradigms.- Moderator
- Heather WestInternational Data Corporation (IDC)
- Panelists
- Katherine KlymkoLawrence Berkeley National Laboratory
- Kristel MichielsenJülich Supercomputing Centre (JSC)
- Trevor LantingD-Wave Quantum
- Steven GirvinYale University
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
-
Paper
GLOW: Accelerating Geometric Multigrid Layers on Cerebras Wafer-Scale Engine
Full description
This paper describes the mapping of geometric multigrid (GMG) onto the Cerebras Wafer Scale Engine (WSE-3). The WSE wafer integrates almost a million cores along with a low-latency 2D mesh network, providing unprecedented on-chip bandwidth and fine-grained communication. Although the architecture is predominantly used for deep learning, we explore the suitability of WSE for accelerating GMG. GMG solves sparse linear systems efficiently by recursively coarsening the grid, which leads to more rapid convergence with less computation. As compared to prior implementations of stencils on WSE, this demonstration highlights the challenges of mapping a layered algorithm – and the full set of operators – onto a dataflow architecture. We compare the performance of this implementation to that of the HPGMG benchmark on an NVIDIA GH200 GPU. Across different GMG configurations and problem sizes, we observe a speedup of up to 25× as compared to a single GH200 node.- Authors
- Sameeran JoshiUniversity of Utah
- Oscar AnteparaLawrence Berkeley National Laboratory (LBNL)
- Samuel WilliamsLawrence Berkeley National Laboratory (LBNL)
- Hans JohansenLawrence Berkeley National Laboratory (LBNL)
- Mary HallUniversity of Utah
-
Birds of a Feather
Bringing Quantum Computing to the HPC Datacenter
Full description
As quantum computing matures, challenges with the adoption of this technology shift from theoretical feasibility to practical deployment within established supercomputing infrastructures. This BoF examines the holistic framework required to operationalize quantum systems, from the physical datacenter floor requirements to the high-level application layer. We dive into the complex interplay between infrastructure installation, platform requirements, security protocols, and the necessary evolution of the software stack to support quantum algorithms and hybrid QC-HPC workflows. As such, this BoF outlines a strategic blueprint for building resilient, high-performance quantum-classical hybrid computing environments suitable for next-generation scientific discovery.- Session Leader
- Ermal RrapajLawrence Berkeley National Laboratory (LBNL)
- Additional Session Leaders
- Seetharami SeelamIBM
- Kentaro YamamotoQuantinuum
- Jonathan WurtzQuera
- Amir ShehataOak Ridge National Laboratory (ORNL)
- Katherine KlymkoLawrence Berkeley National Laboratory (LBNL)
- Thomas Naughton IIIOak Ridge National Laboratory (ORNL)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
-
Birds of a Feather
Designing Open, Heterogeneous System Monitoring Around OpenCHAMI
Full description
While modern management platforms like OpenCHAMI streamline cluster orchestration, they do not inherently provide the comprehensive monitoring and data collection required to observe every facet of a next-generation data center. Operating heterogeneous, large-scale systems requires a unified, open-source observability framework capable of aggregating distributed logs, metrics, and traces across thousands of bleeding-edge compute nodes. This BoF session brings together HPC systems engineers, data center architects, and site reliability engineers to discuss the architecture, integration challenges, and best practices for building robust telemetry pipelines around OpenCHAMI, ensuring operational visibility, rapid fault isolation, and open-source compliance in next-generation supercomputing environments.- Primary Session Leader
- Massimo BeniniSwiss National Supercomputing Centre (CSCS)
- Additional Session Leaders
- John StileLawrence Berkeley National Laboratory (LBNL)
- James BrandtSandia National Laboratories
- Matthew PhillipsLos Alamos National Laboratory (LANL)
- Mark PlaggeSandia National Laboratories
- Sarra RefaiJülich Supercomputing Centre (JSC)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
-
Birds of a Feather
HPC User Support in the AI and Exascale Era
Full description
As HPC enters an era of extreme heterogeneity and AI integration, user support is a critical strategic enabler. However, rising system complexity and finite staffing strain traditional models. To thrive in the AI and exascale era, a collaborative community approach is essential. This BoF session invites practitioners, providers, and users to collaboratively reshape HPC support. Featuring insights from facility leaders on managing rapid AI adoption, an interactive live survey, and open group discussions, participants will share experiences regarding scaling, AI automation, and knowledge transfer. Outcomes will form a post-conference report dedicated to fostering a sustainable, community-driven support ecosystem.- Session Leader
- Haritha Siddabathuni SomArgonne National Laboratory
- Additional Session Leaders
- Rebecca Hartman-BakerLawrence Berkeley National Laboratory (LBNL)
- Chris FusonOak Ridge National Laboratory (ORNL)
- Smita DarmoraUniversity of Illinois Chicago
- Thomas BreuerJülich Supercomputing Centre (JSC)
- Helena Vela BeltranDo IT Now
- Ilya ZhukovJülich Supercomputing Centre (JSC)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
-
Birds of a Feather
Real-Time Scientific Data Streaming to HPC Nodes: Challenges and Innovations VI
Full description
Emerging scientific needs are driving a new class of workflows that require near real-time processing, urgent computing, and time-sensitive decision-making. The traditional "compute-store-analyze" paradigm, which relies on dumping massive datasets to file systems for post-processing, has become an unsustainable bottleneck. It introduces unacceptable latency, and often renders high-frequency data analysis practically impossible. This BoF discusses this challenge by exploring the architectural shift from file-based workflows to high-performance streaming. In this new paradigm, storage is treated not as a buffer for post-processing, but as a sink for final scientific results, while analysis, visualization, and reduction occur in-transit or in-situ.- Session Leader
- Bjoern EndersLawrence Berkeley National Laboratory (LBNL); National Energy Research Scientific Computing Center (NERSC)
- Additional Session Leaders
- Adam ThompsonNVIDIA Corporation
- Alex UptonSwiss National Supercomputing Centre (CSCS)
- Rafael Ferreira da SilvaOak Ridge National Laboratory (ORNL)
- Eli DartEnergy Sciences Network (ESnet); Lawrence Berkeley National Laboratory (LBNL)
- Rajive JoshiRTI
- Rachana AnanthakrishnanUniversity of Chicago
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
-
Birds of a Feather
Trusted Research Environments for AI and Integrated Science: Launching the Genesis Mission
Full description
This BoF aims to build on the successful launch at SC25 of a multi-national community of practice. This year we focus on the Genesis Mission and DOE’s American Science Cloud and AI Models Consortium projects, to include establishing best practices for multi-facility or multi-institutional workflows, lessons learned from international contexts, sharing and accessing resources, developing mixed public/private partnerships, and protecting sensitive data privacy and security. Our goal is to advance progress in developing secure and trustworthy infrastructures for AI and integrated science. We seek to understand trustworthiness in supercomputing environments from both technical and socio-technical perspectives.- Session Leader
- Ben BrownDOE Office of Advanced Scientific Computing Research
- Additional Session Leaders
- Jeremy CramptonU.S. Department of Energy
- Lois Curfman McInnesArgonne National Laboratory
- Debbie BardLawrence Berkeley National Laboratory (LBNL)
- Arjun ShankarOak Ridge National Laboratory (ORNL)
- Sadaf AlamUniversity of Bristol
- Carol HawkDOE Office of Advanced Scientific Computing Research
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
-
Birds of a Feather
Watts, FLOPs, and Impact: Rethinking Energy in the Exascale Era
Full description
Energy consumption has become a key constraint in high performance computing (HPC), shaping system design, procurement, software, and scientific workflows in the exascale era. As large-scale simulations and AI-driven workloads increase power demand, the community must reconsider how performance is measured, optimized, and reported beyond FLOPs alone. This BoF session will convene researchers, facility operators, vendors, and application scientists to discuss energy measurement, performance-per-watt metrics, efficient algorithms and runtimes, workload scheduling, and carbon-aware computing. The session will identify practical gaps, share lessons from production systems, and define community priorities for sustainable, energy-efficient HPC at scale.- Session Leader
- Sameh AbdulahKing Abdullah University of Science and Technology (KAUST)
- Additional Session Leaders
- Bilel HadriKing Abdullah University of Science and Technology (KAUST)
- Natalie BatesLawrence Berkeley National Laboratory (LBNL)
- Steven MartinHPE
- Siddhartha JanaIntel Corporation; Energy Efficient High Performance Computing Working Group (EE HPC WG)
- Mathis BodeForschungszentrum Jülich
- Tapasya PatkiLawrence Livermore National Laboratory (LLNL)
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
- TUT — Tutorial Reg Pass
- XO/EX — Exhibits Reg Pass
Friday, November 20
-
Panel
Accelerated Quantum Supercomputing: AI, HPC, and the Quantum Applications Worth Betting On
Full description
Quantum processors are not useful in isolation. Today's small-scale QPUs require only modest classical hardware, but as systems scale toward utility and fault tolerance, dependencies on classical HPC and AI grow rapidly. These dependencies span the full stack: real-time decoders for fault-tolerant error correction, AI-assisted state preparation and circuit synthesis, hybrid quantum-classical algorithms, HPC-scale data processing, and verification. The serious questions for the SC community span all timelines: Which applications can quantum-HPC-AI integrated systems meaningfully accelerate now, which become reachable with fault tolerance, and which remain infeasible despite widespread touting? Integration approaches differ widely — RIKEN's Fugaku-IBM/Quantinuum hybrid platform, DOE lab user programs at NERSC and ORNL, NVIDIA's NVQLink GPU-QPU interconnect, and Google and AWS's standalone fault-tolerant roadmaps with classical co-processing serving error correction. These distinct framings reflect different visions for HPC-AI integration, and our community lacks consensus on which application and architecture pairings are optimal.- Moderator
- Elica KyosevaNVIDIA Corporation
- Panelists
- Katherine KlymkoLawrence Berkeley National Laboratory (LBNL); NERSC
- Salvatore MandràGoogle Quantum AI
- Travis HumbleOak Ridge National Laboratory (ORNL)
- Fred ChongUniversity of Chicago; Infleqtion
- Mitsuhisa SatoRIKEN Center for Computational Science (R-CCS); Juntendo University
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
-
Workshop
Eighth Workshop on Interactive and Urgent High Performance Computing
Full description
Interactivity enables the exploitation of HPC in new and revolutionary ways, delivering many new and exciting opportunities for our community. Interactive HPC involves users being in the loop during job execution monitoring a job, steering the experiment, or visualizing results to make immediate decisions. Likewise, urgent computing combines interactive computational modeling with time-sensitive systems in the real world such as the near-real-time analysis of unfolding disasters to inform real-time decision making. Supporting interactive and urgent workloads on HPC requires expertise in a wide range of areas and the solving of numerous technical and organizational challenges. This workshop brings together stakeholders, researchers, and practitioners from across interactive and urgent computing within the wider HPC community. We will share success stories, case studies, and technologies to continue community building around leveraging urgent and interactive HPC as an important tool for scientific research, responding to disasters, and addressing societal issues.- Session Chairs
- Thierry GoubierFrench Alternative Energies and Atomic Energy Commission (CEA)
- Bill ArndtLawrence Berkeley National Laboratory (LBNL); National Energy Research Scientific Computing Center (NERSC)
- Robert HenschelIndiana University
- Michael RingenburgAllen Institute
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
-
Workshop
Eighth Workshop on Interactive and Urgent High Performance Computing (WIUHPC)
Full description
Interactivity enables the exploitation of HPC in new and revolutionary ways, delivering many new and exciting opportunities for our community. Interactive HPC involves users being in the loop during job execution monitoring a job, steering the experiment, or visualizing results to make immediate decisions. Likewise, urgent computing combines interactive computational modeling with time-sensitive systems in the real world such as the near-real-time analysis of unfolding disasters to inform real-time decision making. Supporting interactive and urgent workloads on HPC requires expertise in a wide range of areas and the solving of numerous technical and organizational challenges. This workshop brings together stakeholders, researchers, and practitioners from across interactive and urgent computing within the wider HPC community. We will share success stories, case studies, and technologies to continue community building around leveraging urgent and interactive HPC as an important tool for scientific research, responding to disasters, and addressing societal issues.- Organizers
- Thierry GoubierCEA
- Will ArndtNERSC
- Robert HenschelIndiana University
- Michael RingenburgAllen Institute
-
Workshop
Workshop on Operational Data Analytics in HPC (HPC-ODA)
Full description
Operational data analytics (ODA) provides unique opportunities to analyze, understand, and optimize HPC operations. Readily available open-source frameworks make the collection of telemetry from different operational domains of HPC increasingly easy. However, analyzing the data, extracting actionable knowledge from it, and leveraging it to optimize HPC operations is challenging. While most HPC sites have sophisticated monitoring capabilities in place, their analysis capabilities often do not go beyond visual inspection by humans via dashboards or static thresholds for alarms. Few early adopter sites have pioneered advanced usage of data, but its wider adoption will depend on success stories and the availability of analysis tools and models. This workshop will bring together HPC users, practitioners in HPC operations, monitoring experts, and data analytics researchers to share experiences, discuss state-of-the-art research, and foster collaboration. It will solicit papers with concrete, reproducible, and empirical data gathered through case studies, research for practice, and surveys.- Session Chairs
- Ayse Kivilcim CoskunBoston University; Emerald AI
- Jeff HansonHPE; EE HPC WG
- Melissa RomanusLawrence Berkeley National Laboratory (LBNL)
- Michael OttLeibniz Supercomputing Centre (LRZ)
- Tim OsborneOak Ridge National Laboratory (ORNL)
- Woong ShinOak Ridge National Laboratory (ORNL); EE HPC WG
- Recordings
- Livestreamed, Recorded
- Registration Categories
- TP — Tech Program Reg Pass
- W — Workshop Reg Pass
-
Workshop
On preprocessing heterogeneous HPC telemetry for use in foundation models and digital twins
Full description
HPC facilities publish telemetry supporting foundation-model training and digital-twin replay but lack common fields, units, and privacy conventions. We extend REDI and AIDRIN to HPC telemetry via a new domain processor and agentic telemetry skill, and apply SetGo, forming an end-to-end pipeline that transforms heterogeneous telemetry into AI-ready and digital-twin-ready datasets. For digital twins, we demonstrate cross-facility replay fidelity: naively replaying Marconi100 telemetry through a Frontier twin violates its physical power envelope on 10.23% of samples; renormalization reduces this to 0.0042%. The pipeline supports representative replay and targeted emulation. For foundation models, we find static scheduler metadata alone gives no signal for predicting job-level details; however, ExaFM predicts job runtime with R2 = 0.960 when conditioned on a pre-anonymization identity embedding; without it, transferring that head to Marconi100, unseen during pretraining, reaches R2 = 0.716. We close with actionable recommendations for publishing AI- and digital-twin-ready telemetry toward future pretraining corpora.- Author/Presenters
- Wesley BrewerOak Ridge National Laboratory (ORNL)
- Ketan MaheshwariOak Ridge National Laboratory
- Qing LiuMississippi State University
- Sean WilkinsonOak Ridge National Laboratory
- Bertrand CirouCINES
- Gérard GilCINES
- Jean Luca BezLawrence Berkeley National Laboratory
- Kaveen HinidumaOhio State University
- Suren BynaOhio State University
-
Workshop
Measuring the Impact of the Better Scientific Software Fellowship Program
Full description
The Better Scientific Software Fellowship (BSSwF) Program was launched in 2018 in the U.S. to recognize, develop, and support leaders advancing practices, processes, and tools for better scientific software. Over eight years, the program has supported 43 Fellows and 43 Honorable Mentions across institutions and career stages to advance developer productivity and software sustainability and strengthen the scientific software community. A 2025 impact assessment examined program reach, contributions to developer productivity and software sustainability, and effects on careers, mentorship, and community engagement. Based on a landscape analysis, review of Fellow projects, and interviews with Fellows, findings indicate that the program expanded professional networks and visibility and created opportunities for collaboration, mentoring, and career advancement. Program elements include community-facing projects, professional engagement, mentoring, and dissemination of Fellows’ work. BSSwF provides a model and insights for training, professional engagement, networking, and workforce-development programs supporting individual careers and the broader scientific software ecosystem.- Author/Presenters
- Mary Ann LeungSustainable Horizons Institute
- Danielle JacquesIndependent Consultant
- Connie McNeelyGeorge Mason University
- David BernholdtOak Ridge National Laboratory
- Elsa GonsiorowskiLawrence Livermore National Laboratory
- Michael HerouxParaTools, Inc.
- Adam LavelyLawrence Berkeley National Laboratory
- Lois McInnesArgonne National Laboratory
-
Workshop
Perlmutter-Agent: Design and Early Deployment of a Local, Tool-Grounded LLM Assistant for HPC Incident Investigation
Full description
HPC incident investigation requires system administrators to correlate metrics, logs, and ticket history across heterogeneous data sources. We present Perlmutter-Agent, a locally deployed, tool-grounded large language model (LLM) assistant for incident investigation on NERSC’s Perlmutter supercomputer. An orchestrator converts Slack messages into structured requests and dispatches them to specialized metrics, logs, and ticket sub-agents. Guided by site-specific agent skills, the sub-agents query operational data through MCP servers and return evidence that the orchestrator summarizes into likely causes. All agents use an open-weight Qwen3.8-27B model served locally by vLLM, keeping operational data from external model providers. In a restricted beta, 339 of 380 user requests (89.2\%) completed successfully; diagnostic accuracy and operator usefulness remain to be evaluated.- Author/Presenter
- Siqi DengLawrence Berkeley National Laboratory (LBNL)
-
Workshop
Towards MPI Communication Modeling From Passive Network Telemetry
Full description
The distribution of MPI message sizes issued by an application is a first-order determinant of its communication behavior, yet recovering this distribution typically requires application-level instrumentation and/or a priori knowledge of the communication pattern. We present an application-agnostic model of MPI communication that reconstructs the MPI message-size distribution from passive network telemetry: packet-level counters that modern NICs expose and that facility operators already collect continuously, system-wide, at negligible cost. The model requires no knowledge of the network topology, the communication peers, or the MPI operations in use, and can be applied retroactively to archived telemetry. Instantiated on the Cassini counters of NERSC's Perlmutter, the model estimates OSU Micro-Benchmark message counts within 0.5% and the dominant LAMMPS size classes within 10%, which is a first step toward instrumentation-free prediction of communication time.- Author/Presenters
- Rui LiuLawrence Berkeley National Laboratory
- Paul LinNVIDIA
- Evan WeinbergNVIDIA
- Dhruva KulkarniLawrence Berkeley National Laboratory
- Zhengji ZhaoLawrence Berkeley National Laboratory
- Ian KarlinQueen's University
- Garima KochharNVIDIA
- Brian AustinLawrence Berkeley National Laboratory
No events match your search and selected filters. Try another search or clear these filters.
Source: SC26 Conference Program. Program details may change.
Last edited: November 9, 2026
