Showing 1–11 of 11 results for author: Cumming, B

MLPerf HPC: A Holistic Benchmark Suite for Scientific Machine Learning on HPC Systems

Authors: Steven Farrell, Murali Emani, Jacob Balma, Lukas Drescher, Aleksandr Drozd, Andreas Fink, Geoffrey Fox, David Kanter, Thorsten Kurth, Peter Mattson, Dawei Mu, Amit Ruhela, Kento Sato, Koichi Shirahata, Tsuguchika Tabaru, Aristeidis Tsaris, Jan Balewski, Ben Cumming, Takumi Danjo, Jens Domke, Takaaki Fukai, Naoto Fukumoto, Tatsuya Fukushi, Balazs Gerofi, Takumi Honda , et al. (18 additional authors not shown)

Abstract: Scientific communities are increasingly adopting machine learning and deep learning models in their applications to accelerate scientific insights. High performance computing systems are pushing the frontiers of performance with a rich diversity of hardware resources and massive scale-out capabilities. There is a critical need to understand fair and effective benchmarking of machine learning appli… ▽ More Scientific communities are increasingly adopting machine learning and deep learning models in their applications to accelerate scientific insights. High performance computing systems are pushing the frontiers of performance with a rich diversity of hardware resources and massive scale-out capabilities. There is a critical need to understand fair and effective benchmarking of machine learning applications that are representative of real-world scientific use cases. MLPerf is a community-driven standard to benchmark machine learning workloads, focusing on end-to-end performance metrics. In this paper, we introduce MLPerf HPC, a benchmark suite of large-scale scientific machine learning training applications driven by the MLCommons Association. We present the results from the first submission round, including a diverse set of some of the world's largest HPC systems. We develop a systematic framework for their joint analysis and compare them in terms of data staging, algorithmic convergence, and compute performance. As a result, we gain a quantitative understanding of optimizations on different subsystems such as staging and on-node loading of data, compute-unit utilization, and communication scheduling, enabling overall $>10 \times$ (end-to-end) performance improvements through system scaling. Notably, our analysis shows a scale-dependent interplay between the dataset size, a system's memory hierarchy, and training convergence that underlines the importance of near-compute storage. To overcome the data-parallel scalability challenge at large batch sizes, we discuss specific learning techniques and hybrid data-and-model parallelism that are effective on large systems. We conclude by characterizing each benchmark with respect to low-level memory, I/O, and network behavior to parameterize extended roofline performance models in future rounds. △ Less

Submitted 26 October, 2021; v1 submitted 21 October, 2021; originally announced October 2021.

Light Yield Quenching and Quenching Remediation in Liquid Scintillator Detectors

Authors: S. Hans, J. B. Cumming, R. Rosero, R. Diaz Perez, C. Camilo Reyes, S. S. Gokhale, M. Yeh

Abstract: Quenching of light emission from an LAB based scintillator by the addition of organic amines and carboxylic acids is examined. Chemical functional groups of the quenching agents play an important role in this reduction. It is shown that "salt" formation at a 1:1 mole ratio in a mixed amine-acid system, reduces quenching by a factor of 2. Supporting NMR spectra are presented. This "quenching neutra… ▽ More Quenching of light emission from an LAB based scintillator by the addition of organic amines and carboxylic acids is examined. Chemical functional groups of the quenching agents play an important role in this reduction. It is shown that "salt" formation at a 1:1 mole ratio in a mixed amine-acid system, reduces quenching by a factor of 2. Supporting NMR spectra are presented. This "quenching neutralization" has the potential to reduce the light loss incurred when metals complexed with quenching agents are loaded into organic scintillators. △ Less

Submitted 18 November, 2020; v1 submitted 11 August, 2020; originally announced August 2020.

Journal ref: JINST 2020

Arbor -- a morphologically-detailed neural network simulation library for contemporary high-performance computing architectures

Authors: Nora Abi Akar, Ben Cumming, Vasileios Karakasis, Anne Küsters, Wouter Klijn, Alexander Peyser, Stuart Yates

Abstract: We introduce Arbor, a performance portable library for simulation of large networks of multi-compartment neurons on HPC systems. Arbor is open source software, developed under the auspices of the HBP. The performance portability is by virtue of back-end specific optimizations for x86 multicore, Intel KNL, and NVIDIA GPUs. When coupled with low memory overheads, these optimizations make Arbor an or… ▽ More We introduce Arbor, a performance portable library for simulation of large networks of multi-compartment neurons on HPC systems. Arbor is open source software, developed under the auspices of the HBP. The performance portability is by virtue of back-end specific optimizations for x86 multicore, Intel KNL, and NVIDIA GPUs. When coupled with low memory overheads, these optimizations make Arbor an order of magnitude faster than the most widely-used comparable simulation software. The single-node performance can be scaled out to run very large models at extreme scale with efficient weak scaling. HPC, GPU, neuroscience, neuron, software △ Less

Submitted 17 January, 2019; originally announced January 2019.

Comments: PDP 2019 27th Euromicro International Conference on Parallel, Distributed and Network-based Processing

Journal ref: 2019 27th Euromicro International Conference on Parallel, Distributed and Network-Based Processing (PDP), Pavia, Italy, 2019, pp. 274-282

Improving Light Yield Measurements for Low-Yield Scintillators

Authors: J. B. Cumming, S. Hans, M. Yeh

Abstract: Light power spectra are introduced as a new tool for relative light yield (LY) determinations. Light event spectra have commonly been used for this purpose. Theoretical background supporting this change is provided. It is shown that the derivative of a light power spectrum can provide a reliable LY measurement at levels as low as 2% of those for high-yield liquid scintillators. Applications to lig… ▽ More Light power spectra are introduced as a new tool for relative light yield (LY) determinations. Light event spectra have commonly been used for this purpose. Theoretical background supporting this change is provided. It is shown that the derivative of a light power spectrum can provide a reliable LY measurement at levels as low as 2% of those for high-yield liquid scintillators. Applications to light evolution in the PPO+LAB system and to water-based liquid scintillators are described. △ Less

Submitted 5 October, 2018; originally announced October 2018.

Photocurrent Enhancement of Graphene Photodetectors by Photon Tunneling of Light into Surface Plasmons

Authors: Alireza Maleki, Benjamin P. Cumming, Min Gu, James E. Downes, David W. Coutts, Judith M. Dawes

Abstract: We demonstrate that surface plasmon resonances excited by photon tunneling through an adjacent dielectric medium enhance photocurrent detected by a graphene photodetector. The device is created by overlaying a graphene sheet over an etched gap in a gold film deposited on glass. The detected photocurrents are compared for five different excitation wavelengths, ranging from nm to nm. The photocurren… ▽ More We demonstrate that surface plasmon resonances excited by photon tunneling through an adjacent dielectric medium enhance photocurrent detected by a graphene photodetector. The device is created by overlaying a graphene sheet over an etched gap in a gold film deposited on glass. The detected photocurrents are compared for five different excitation wavelengths, ranging from nm to nm. The photocurrent excited with incident p-polarized light (the case for resonant surface plasmon excitation) is significantly amplified in comparison with that for s-polarized light (without surface plasmon resonances). We observe that the photocurrent is greater for shorter wavelengths (for both s and p-polarizations) due to the increased photothermal current resulting from higher dam** of surface plasmons at shorter wavelength excitation. Position-dependent Raman spectroscopic analysis of the optically-excited graphene photodetector indicates the presence of charge carriers near the metallic edge. In addition, we show that the polarity of photocurrent switches across the gap as the incident light spot moves across the gap. Graphene-based photodetectors offer a simple architecture which can be fabricated on dielectric waveguides to exploit the plasmonic photocurrent enhancement of the evanescent field for detection. Applications for these devices include photo-detection, optical sensing and direct plasmonic detection. △ Less

Submitted 12 October, 2016; originally announced October 2016.

Comments: 14 pages, 6 figures

First Experiences With Validating and Using the Cray Power Management Database Tool

Authors: Gilles Fourestey, Ben Cumming, Ladina Gilly, Thomas C. Schulthess

Abstract: In October 2013 CSCS installed the first hybrid Cray XC-30 system, dubbed Piz Daint. This system features the power management database (PMDB), that was recently introduced by Cray to collect detailed power consumption information in a non-intrusive manner. Power measurements are taken on each node, with additional measurements for the Aries network and blowers, and recorded in a database. This en… ▽ More In October 2013 CSCS installed the first hybrid Cray XC-30 system, dubbed Piz Daint. This system features the power management database (PMDB), that was recently introduced by Cray to collect detailed power consumption information in a non-intrusive manner. Power measurements are taken on each node, with additional measurements for the Aries network and blowers, and recorded in a database. This enables fine-grained reporting of power consumption that is not possible with external power meters, and is useful to both application developers and facility operators. This paper will show how benchmarks of representative applications at CSCS were used to validate the PMDB on Piz Daint. Furthermore we will elaborate, with the well-known HPL benchmark serving as prototypical application, on how the PMDB streamlines the tuning for optimal power efficiency in production, which lead to Piz Daint being recognised as the most energy efficient petascale supercomputer presently in operation. △ Less

Submitted 12 August, 2014; originally announced August 2014.

Comments: This paper was presented at the 2014 Cray User Group (CUG) user meeting in Lugano, Switzerland,First Experiences With Validating and Using the Cray Power Management Database Tool, Gilles Fourestey and Ben Cumming and Ladina Gilly, Proceedings of the CUG meeting, 2014

Temperature Dependence of Light Absorption by Water

Authors: J. B. Cumming

Abstract: A model is described that relates the temperature coefficient of the optical absorption spectrum of pure water to the frequency derivative of that spectrum and two parameters that quantify the dependence of a peak's amplitude and its position on temperature. When applied to experimental temperature coefficients, it provides a better understanding of the process than the analysis currently in use. A model is described that relates the temperature coefficient of the optical absorption spectrum of pure water to the frequency derivative of that spectrum and two parameters that quantify the dependence of a peak's amplitude and its position on temperature. When applied to experimental temperature coefficients, it provides a better understanding of the process than the analysis currently in use. △ Less

Submitted 9 January, 2013; originally announced January 2013.

Production of phi Mesons in Au+Au Collisions at 11.7 A GeV/c

Authors: E917 Collaboration, B. B. Back, R. R. Betts, J. Chang, W. C. Chang, C. Y. Chi, Y. Y. Chu, J. B. Cumming, J. C. Dunlop, W. Eldredge, S. Y. Fung, R. Ganz, E. Garcia-Solis, A. Gillitzer, G. Heinzelman, W. F. Henning, D. J. Hofman, B. Holzman, J. H. Kang, E. J. Kim, S. Y. Kim, Y. Kwon, D. McLeod, A. C. Mignerey, M. Moulson , et al. (15 additional authors not shown)

Abstract: We report on a measurement of phi-meson production in Au+Au collisions at a beam momentum of 11.7 A GeV/c by Experiment E917 at the AGS. The measurement covers the midrapidity region 1.2 < y < 1.6. Transverse-mass spectra and the rapidity distribution are presented as functions of centrality characterized by the number of participant projectile nucleons. The yield of phi's per participant projec… ▽ More We report on a measurement of phi-meson production in Au+Au collisions at a beam momentum of 11.7 A GeV/c by Experiment E917 at the AGS. The measurement covers the midrapidity region 1.2 < y < 1.6. Transverse-mass spectra and the rapidity distribution are presented as functions of centrality characterized by the number of participant projectile nucleons. The yield of phi's per participant projectile nucleon increases strongly in central collisions in a manner similar to that observed for kaons. △ Less

Submitted 5 April, 2004; v1 submitted 22 April, 2003; originally announced April 2003.

Comments: 12 pages, 3 table, 14 eps figures which use \includegraphics, revtex 4, revised version submitted to Phys. Rev. C

Journal ref: Phys.Rev.C69:054901,2004

Antilambda production in Au+Au collisions at 11.7 A GeV/c

Authors: E917 Collaboration, B. B. Back, R. R. Betts, J. Chang, W. C. Chang, C. Y. Chi, Y. Y. Chu, J. B. Cumming, J. C. Dunlop, W. Eldredge, S. Y. Fung, R. Ganz, E. Garcia-Solis, A. Gillitzer, G. Heinzelman, W. F. Henning, D. J. Hofman, B. Holzman, J. H. Kang, E. J. Kim, S. Y. Kim, Y. Kwon, D. McLeod, A. C. Mignerey, M. Moulson , et al. (15 additional authors not shown)

Abstract: We present results from Experiment E917 for antilambda and antiproton production in Au+Au collisions at 11.7 A GeV. We have measured invariant spectra and yields for both species in central and peripheral collisions. We find that the antilambda/antiproton ratio near mid-rapidity increases from 0.26+0.19-0.15 in peripheral collisions to 3.6+4.7-1.8 in central collisions, a value that is substanti… ▽ More We present results from Experiment E917 for antilambda and antiproton production in Au+Au collisions at 11.7 A GeV. We have measured invariant spectra and yields for both species in central and peripheral collisions. We find that the antilambda/antiproton ratio near mid-rapidity increases from 0.26+0.19-0.15 in peripheral collisions to 3.6+4.7-1.8 in central collisions, a value that is substantially larger than current theoretical estimates. △ Less

Submitted 22 January, 2001; originally announced January 2001.

Comments: 5 pages, 1 table, 3 eps figures which use \includegraphics, revtex 4

Journal ref: Phys.Rev.Lett.87:242301,2001

Baryon Rapidity Loss in Relativistic Au+Au Collisions

Authors: E917 Collaboration, B. B. Back, R. R. Betts, J. Chang, W. C. Chang, C. Y. Chi, Y. Y. Chu, J. B. Cumming, J. C. Dunlop, W. Eldredge, S. Y. Fung, R. Ganz, E. Garcia-Solis, A. Gillitzer, G. Heintzelman, W. F. Henning, D. J. Hofman, B. Holzman, J. H. Kang, E. J. Kim, S. Y. Kim, Y. Kwon, D. McLeod, A. C. Mignerey, M. Moulson , et al. (15 additional authors not shown)

Abstract: An excitation function of proton rapidity distributions for different centralities is reported from AGS Experiment E917 for Au+Au collisions at 6, 8, and 10.8 GeV/nucleon. The rapidity distributions from peripheral collisions have a valley at midrapidity which smoothly change to distributions that peak at midrapidity for central collisions. The mean rapidity loss increases with increasing beam e… ▽ More An excitation function of proton rapidity distributions for different centralities is reported from AGS Experiment E917 for Au+Au collisions at 6, 8, and 10.8 GeV/nucleon. The rapidity distributions from peripheral collisions have a valley at midrapidity which smoothly change to distributions that peak at midrapidity for central collisions. The mean rapidity loss increases with increasing beam energy, whereas the fraction of protons consistent with isotropic emission from a thermal source at midrapidity decreases with increasing beam energy. △ Less

Submitted 2 March, 2001; v1 submitted 14 March, 2000; originally announced March 2000.

Comments: 9 pages, 3 figures, 1 table replaced version has figures rotated

Journal ref: Phys.Rev.Lett.86:1970-1973,2001

Do statistically significant correlations exist between the Homestake solar neutrino data and sunspots?

Authors: J. Boger, R. L. Hahn, J. B. Cumming

Abstract: It has been suggested by various authors that a significant anticorrelation exists between the Homestake solar neutrino data and the sunspot cycle. Some of these claims rest on smoothing the data by taking running averages, a method that has recently undergone criticism. We demonstrate that no significant anticorrelation can be found in the Homestake, data, or in standard 2- and 4-point averages… ▽ More It has been suggested by various authors that a significant anticorrelation exists between the Homestake solar neutrino data and the sunspot cycle. Some of these claims rest on smoothing the data by taking running averages, a method that has recently undergone criticism. We demonstrate that no significant anticorrelation can be found in the Homestake, data, or in standard 2- and 4-point averages of that data. However, when 3-, 5-, and 7-point running averages are taken, an anticorrelation seems to emerge whose significance grows as the number of points in the average increases. Our analysis indicates that the apparently high significance of these anticorrelations is an artifact of the failure to consider the loss of independence introduced in the running average process. When this is considered, the significance is reduced to that of the unaveraged data. Furthermore, when evaluated via parametric subsampling, no statistically significant anticorrelation is found. We conclude that the Homestake data can not be used to substantiate any claim of an anticorrelation with the sunspot cycle. △ Less

Submitted 27 January, 2000; originally announced January 2000.

Comments: 17 pages, 5 figures, AasTeX

Report number: BNL-66906

Journal ref: Astrophys.J. 537 (2000) 1080-1085