BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/New_York
X-LIC-LOCATION:America/New_York
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260422T143139Z
LOCATION:B311
DTSTART;TZID=America/New_York:20241119T163000
DTEND;TZID=America/New_York:20241119T170000
UID:submissions.supercomputing.org_SC24_sess382_pap107@linklings.com
SUMMARY:Scaling Molecular Dynamics with ab initio Accuracy to 149 Nanoseco
 nds per Day
DESCRIPTION:Jianxiong Li, Boyang Li, Zhuoqiang Guo, MingZhen Li, and Enji 
 Li (Chinese Academy of Sciences, Institute of Computing Technology, State 
 Key Lab of Processors; University of Chinese Academy of Sciences, Beijing,
  China); Lijun Liu (Osaka University, Department of Mechanical Engineering
 , Graduate School of Engineering); and Guojun Yuan, Zhan Wang, Guangming T
 an, and Weile Jia (Chinese Academy of Sciences, Institute of Computing Tec
 hnology, State Key Lab of Processors; University of Chinese Academy of Sci
 ences, Beijing, China)\n\nPhysical phenomena such as chemical reactions, b
 ond-breaking, and phase transition require molecular dynamics (MD) simulat
 ion with ab initio accuracy ranging from milliseconds to microseconds. How
 ever, previous state-of-the-art neural network-based MD packages such as D
 eePMD-kit can only reach 4.7 nanoseconds per day on the Fugaku supercomput
 er. In this paper, we present a novel node-based parallelization scheme to
  reduce communication by 81%, then optimize the computationally intensive 
 kernels with sve-gemm and mixed precision. Finally, we implement intra-nod
 e load balance to further improve the overall performance. Numerical resul
 ts on the Fugaku supercomputer show that our work has significantly improv
 ed the time-to-solution of the DeePMD-kit by a factor of 31.7x, reaching 1
 49 nanoseconds per day on 12,000 computing nodes. This work has opened the
  door for millisecond simulation with ab initio accuracy within one week f
 or the first time.\n\nTag: Accelerators, Applications and Application Fram
 eworks, Graph Algorithms, Modeling and Simulation, Numerical Methods\n\nRe
 gistration Category: Tech Program Reg Pass\n\nSession Chair: Wenqian Dong 
 (Oregon State University)\n\n
END:VEVENT
END:VCALENDAR