AbacusSummit is the world’s largest suite of high-performance cosmological N-body simulations, developed to meet and exceed the analysis requirements of the Dark Energy Spectroscopic Instrument. These simulations allow one to predict the large-scale structure that results from a wide range of cosmological models, enabling detailed investigations of theories of cosmological structure formation and comparison to the coming decade of observational surveys.
AbacusSummit was performed with the Abacus code on the Summit supercomputer at the Oak Ridge Leadership Computing Facility (OLCF) using time from the ALCC program. AbacusSummit provides nearly 60 trillion simulated particles over 97 cosmologies, including 140 examples of 330 billion particles in 2 Gpc/h boxes. It generated about 2 petabytes of data products, comprising halo catalogs, merger trees, and particle subsamples. These data products are being placed in the public domain, with access arranged for large collaborations. More details are available here.
The Abacus code aims to produce extremely high-performance and high-accuracy simulations, leveraging the opportunities of a new mathematical algorithm for the solution of gravitational forces and the high speed of graphics processing units (GPUs). The code has been exquisitely optimized and indeed co-developed with custom-built computers. Abacus has been developed over a decade. Currently it is actively developed by Lehman Garrison, Nina Maksimova, and Daniel Eisenstein, with contributions from Boryana Hadzhiyska and Sownak Bose and consulting from Philip Pinto. Abacus was initially developed by Marc Metchnik and Philip Pinto, with contributions from Daniel Eisenstein and later development led by Douglas Ferrer.
Abacus found a superb match with the OLCF Summit supercomputer, which featured a combination of GPU power, CPU power, memory, and storage that allowed large Abacus simulations to be run. Abacus achieved very high performance on Summit, completing this world-leading suite in about 350,000 node-hours.