Berkeley Lab supports three co-design centers under the Department of Energy’s Exascale Computing Project (ECP). It leads the Block-Structured Adaptive Mesh Refinement Co-Design Center and partners with other national labs and universities in the ExaGraph and ExaLearn Co-Design Centers.
Block-Structured Adaptive Mesh Refinement Co-Design Center
Berkeley Lab leads the Block-Structured Adaptive Mesh Refinement (AMReX) Co-Design Center.
John Bell heads the center with support from Argonne National Laboratory and the National Renewable Energy Laboratory. The Co-Design Center is developing a new framework, AMReX, to support the development of block-structured AMR algorithms for solving systems of partial differential equations (PDE’s) with complex boundary conditions on exascale architectures. Block-structured AMR provides a natural framework in which to focus computing power on the most critical parts of the problem in the most computationally efficient way possible.
Block-structured AMR is already widely used to solve many problems relevant to DOE. Berkeley Lab’s Center for Computational Sciences and Engineering, which Bell previously led, and the lab’s Applied Numerical Algorithms Group have respectively developed BoxLib and Chombo, two of the leading AMR frameworks used by the U.S. research community. Codes in six of the 22 exascale application projects (in the areas of accelerators, additive manufacturing, astrophysics, combustion, cosmology, and multiphase flow) rely on the AMReX software framework.
Berkeley Lab is also a partner in the ECP Co-Design Center that focuses on Graph Analytics — combinatorial (graph) kernels that play a crucial enabling role in many data analytic computing (DAC) application areas as well as several ECP applications.
The ExaGraph Co-Design Center is led by Pacific Northwest National Laboratory (PNNL) in partnership with Berkeley Lab (Aydın Buluç heads the effort), Sandia National Laboratories and Purdue University. It targets a number of key data analytic computational motifs that are currently not being addressed in existing ECP Co-Design Centers, such as graph traversals, graph matching, graph coloring and graph clustering (including clique enumeration, parallel branch-and-bound, and graph partitioning).
Berkeley Lab is also a partner in the ECP Co-Design Center that focuses on Machine Learning. ExaLearn is a co-design center for Exascale Machine Learning (ML) Technologies and is a collaboration initially consisting of experts from eight multipurpose DOE labs.
Rapid growth in the amount of data and computational power is driving a revolution in machine learning (ML) and artificial intelligence (AI). Beyond the highly visible successes in machine-based natural language translation, these new ML technologies have profound implications for computational and experimental science and engineering and the exascale computing systems that DOE is deploying to support those disciplines.
To address these challenges, the ExaLearn co-design center will provide exascale ML software for use by ECP Applications projects, other ECP Co-Design Centers and DOE experimental facilities and leadership class computing facilities. The ExaLearn Co-Design Center will also collaborate with ECP PathForward vendors on the development of exascale ML software.
The timeliness of ExaLearn’s proposed work ties into the critical national need to enhance economic development through science and technology. It is increasingly clear that advances in learning technologies have profound societal implications and that continued U.S. economic leadership requires a focused effort, both to increase the performance of those technologies and to expand their applications. Linking exascale computing and learning technologies represent a timely opportunity to address those goals.