In this regular feature, HPCwire highlights newly published research in the high-performance computing community and related domains. From parallel programming to exascale to quantum computing, the details are here.
DISTAL: the distributed tensor algebra compiler
Stanford University researchers introduced DISTAL, a “compiler for dense tensor algebra that targets modern distributed and heterogeneous systems.” In this paper, the authors tackle the problem programmers face when implementing “distributed tensor algorithms that are both correct and achieve high performance.” DISTAL, they write, “compiles a tensor algebra domain specific language to a distributed task-based runtime system and supports nodes with multi-core CPUs and multiple GPUs.” The code “generated by DISTAL is competitive with optimized codes for matrix multiply on 256 nodes of the Lassen supercomputer and outperforms existing systems by between 1.8x to 3.7x (with a 45.7x outlier) on higher order tensor operations.”
Authors: Rohan Yadav, Alex Aiken, Fredrik Kjolstad
A multi-institution, international team of researchers introduced the thesan project in this pre-print paper. Thesan is “a suite of large volume (box = 95.5 cMpc) radiation-magneto-hydrodynamic simulations that simultaneously model the large-scale statistical properties of the intergalactic medium during reionization and the resolved characteristics of the galaxies responsible for it.” Using the Gauss Centre for Supercomputing supercomputer SuperMUC-NG at Leibniz Supercomputing Centre, the researchers show that “the flagship simulation has dark matter and baryonic mass resolutions of 3.1 × 106 M and 5.8 × 105 M, respectively. The gravitational forces are softened on scales of 2.2 ckpc with the smallest cell sizes reaching 10 pc at = 5.5, enabling predictions down to the atomic cooling limit.” The researchers “demonstrate that different reionization models give rise to varied bubble size distributions that imprint unique signatures on the 21 cm emission, especially on the slope of the power spectrum at large spatial scales, enabling current and upcoming 21 cm experiments to accurately characterize the sources that dominate the ionizing photon budget.”
Authors: R. Kannan, E. Garaldi, A. Smith, R. Pakmor, V. Springel, M. Vogelsberger and L. Hernquist
This paper by a research team from the Oak Ridge National Laboratory proposes the use of compiler directives as “an attractive programming model for providing portability across different GPU vendors, in which case the porting process may proceed in the reverse direction: from low-level, architecture-specific code to higher level directive-based abstractions.” The researchers “extend MiniMDock to GPU offloading with OpenMP directives, and compare to the performance of kernels using CUDA, and HIP on both Nvidia and AMD GPUs, as well as across different compilers, exploring performance bottlenecks.” The paper documents “this reverse-porting process, from highly optimized device code to a higher-level version using directives, compare code structure, and describe barriers that were overcome in this effort.”
Authors: Mathialakan Thavappiragasam, Wael Elwasif, and Ada Sedova
FastFold: reducing AlphaFold training time from 11 days to 67 hours
An international team of researchers tackles the challenges associated with the training and inference of AlphaFold, which is “an end-to-end model that uses amino acid sequences as model input and directly outputs the three-dimensional structure of the protein.” In this paper from the National University of Singapore, HPC-AI Technology Inc., and Helixon Shanghai Jiao Tong University, the authors “propose FastFold, a highly efficient implementation of protein structure prediction model for training and inference. FastFold includes a series of GPU optimizations based on a thorough analysis of AlphaFold’s performance.” The authors “scaled FastFold to 512 GPUs and achieved aggregate 6.02 PetaFLOPs with 90.1 percent parallel efficiency.” In addition, “experimental results show that FastFold reduces overall training time from 11 days to 67 hours…” The code can be accessed on Github: https://github.com/hpcaitech/FastFold
Authors: Shenggan Cheng, Ruidong Wu, Zhongming Yu, Binrui Li, Xiwen Zhang, Jian Peng, and Yang You
Exascale grid optimization toolkit
The open source Exascale Grid Optimization toolkit was created by a team of researchers from the Pacific Northwest National Laboratory with the goal of solving “large-scale stochastic, security-constrained, multi-period alternating current optimal power flow problems on high-performance computers.” In this paper supported by the Exascale Computing Project, researchers detail “the ExaGO library including its architecture, formulations, modeling details, and its performance for several optimization applications.”
Authors: Shrirang Abhyankar, Slaven Peles, Tamara Becejac, Jesse Holzer, Asher Mancinelli, Cameron Rutherford
Massively parallel modeling and inversion of electrical resistivity tomography data using PFLOTRAN
“An open-source implementation of massively parallel 3D electrical resistivity tomography modeling and inversion algorithms” is presented in this pre-print paper from a team of researchers from the Pacific Northwest National Laboratory. “The forward modeling is based on the FV method and inversion employs the Gauss–Newton method. The computations are parallelized so that all available HPC resources can be exploited, provided they are beneficial. The algorithms are implemented within the framework of PFLOTRAN, which is an open-source, state-of-the-art massively parallel subsurface flow and transport simulation code and has been used extensively for various subsurface applications.” According to the researchers, the “capabilities are accurate, robust, and highly scalable on HPC platforms.”
Authors: Piyoosh Jaysaval, Glenn E. Hammond, and Timothy C. Johnson
Modeling carbon capture on metal-organic frameworks with quantum computing
A team of researchers from Cambridge Quantum Computing, a U.K. quantum software solutions provider, and TotalEnergies, a multi-energy company from France, aim to enable the “use of quantum computing techniques in the quest of sorbents optimization for more efficient carbon capture and conversion applications.” In their effort, documented in this paper, the researchers apply quantum computing to “the problem of CO adsorption in Al-fumarate Metal-Organic Frameworks. Fragmentation strategies based on Density Matrix Embedding Theory are applied, using a variational quantum algorithm as a fragment solver, along with active space selection to minimize qubit number.” The results “suggest that quantum computing methodologies are successful in capturing many-body correlations in the MOF+CO2 system, with physical dissociation curves.”
Authors: Gabriel Greene-Diniz, David Zsolt Manrique, Wassil Sennane, Yann Magnin, Elvira Shishenina, Philippe Cordier, Philip Llewellyn, Michal Krompiec, Marko J. Rančić, David Muñoz Ramo
Do you know about research that should be included in next month’s list? If so, send us an email at [email protected]. We look forward to hearing from you.