Exascale Co-design Center

Exascale Co-design Center
Name	Exascale Co-design Center
Formation	2010s
Type	Research consortium

Exascale Co-design Center. The Exascale Co-design Center is a collaborative research consortium formed to guide the transition of high-performance computing to exascale-class systems. It brings together national laboratories, university research groups, industrial partners, and standards bodies to coordinate hardware and software innovations needed for sustained exascale performance. The Center acts as a focal point connecting program offices, procurement agencies, and vendor ecosystems to align scientific applications, compiler stacks, and system architectures for next-generation supercomputers.

Overview

The Center functions at the intersection of long-range computing roadmaps and near-term procurement programs, engaging stakeholders such as Oak Ridge National Laboratory, Argonne National Laboratory, Lawrence Berkeley National Laboratory, Los Alamos National Laboratory, and Sandia National Laboratories alongside industry participants like Cray Inc., Intel Corporation, NVIDIA, AMD, and IBM. It liaises with initiatives including the Department of Energy exascale initiatives, the U.S. Advanced Scientific Computing Research (ASCR), international efforts at Jülich Research Centre, Riken, and collaborations with university consortia such as Berkeley Lab–affiliated groups and Massachusetts Institute of Technology research teams. Governance models draw on precedents from consortia like The Open Group and U.S. Council on Competitiveness-style public–private partnerships.

Objectives and Scope

Primary objectives include defining target exascale use cases, shaping processor and accelerator requirements, and driving co-design feedback loops among application science teams, compiler developers, and systems engineers. The scope spans numerical simulation communities rooted in projects such as ASCI, DOE Computational Science Graduate Fellowship, and domain sciences represented by programs like Climate Modeling Intercomparison Project, Large Hadron Collider simulation workflows, and Human Genome Project–style bioinformatics pipelines. The Center prioritizes energy efficiency, resilience, programmability, and I/O subsystem evolution in line with procurement frameworks from agencies like National Nuclear Security Administration and standards advanced by IEEE and ISO committees.

Organizational Structure and Partnerships

Organizationally, the Center establishes thematic working groups (architecture, system software, applications, verification, and performance) staffed by scientists from University of Illinois Urbana-Champaign, University of California, Berkeley, Stanford University, University of Michigan, and research staff from laboratories including Pacific Northwest National Laboratory. Industrial partnerships incorporate server OEMs, semiconductor foundries, and software vendors such as Microsoft Research and Google research labs, while standards engagement spans Khronos Group and OpenMP committees. Funding and oversight involve stakeholders like National Science Foundation, Defense Advanced Research Projects Agency, and international partners from institutions such as CERN and European Centre for Medium-Range Weather Forecasts.

Research and Development Activities

R&D activities include co-design cycles coupling application kernels from communities like computational fluid dynamics teams supporting NASA missions, multi-physics codes used by ITER plasma modeling, and cosmology simulations akin to Sloan Digital Sky Survey scale codes with architecture prototyping at facilities like NERSC and Oak Ridge Leadership Computing Facility. Projects address memory hierarchy innovations, network topologies inspired by designs used at Blue Gene and modern torus and dragonfly fabrics, and resilience methods drawing on checkpoint/restart approaches from Los Alamos workflows. The Center promotes cross-cutting efforts in compiler optimizations using toolchains from LLVM and runtime systems influenced by Charm++ and MPI implementations.

Hardware and Software Co-design Projects

Representative projects include co-design of accelerator-integrated CPU nodes leveraging processor roadmaps from ARM Holdings and heterogeneous designs from NVIDIA and AMD, development of power-aware scheduling influenced by research at Princeton University and University of Texas, and storage stack innovations using parallel file systems such as Lustre and object storage models informed by Ceph. Software stacks emphasize portability layers like Kokkos and RAJA, domain-specific languages similar to work at Los Alamos and Sandia, and verification tools adapted from formal methods communities linked to Carnegie Mellon University and Massachusetts Institute of Technology projects.

Performance Evaluation and Benchmarking

The Center defines benchmarks and performance metrics derived from community-supported suites and standards established by groups like SPEC and collaborations with benchmarking consortia. Evaluation combines scaled runs on predecessor petascale systems such as Titan (supercomputer), Sequoia (supercomputer), and prototype platforms at Argonne Leadership Computing Facility with emulation on FPGA clusters from partners including Xilinx and simulation frameworks developed at Lawrence Livermore National Laboratory. Metrics include sustained floating-point throughput, energy-to-solution, parallel efficiency across networks inspired by InfiniBand, and end-to-end science throughput for codes used in climate science, astrophysics, and materials science.

Impact and Legacy

The Center’s coordinated outputs influence procurement of flagship exascale systems, inform academic curricula at institutions like Georgia Institute of Technology and University of Illinois, and seed startup activity in the HPC ecosystem connecting to incubators backed by Department of Energy and National Science Foundation programs. Its legacy includes contributions to open-source toolchains, influence on international supercomputing roadmaps such as those produced by Top500 analysts, and durable collaborations that shaped subsequent petascale-to-exascale transitions for projects tied to national research priorities like Project X and large-scale science facilities. Category:High-performance computing