BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260421T090513Z
LOCATION:Bldg. 6 - Room 103
DTSTART;TZID=Europe/Stockholm:20260629T160000
DTEND;TZID=Europe/Stockholm:20260629T163000
UID:submissions.pasc-conference.org_PASC26_sess158_msa168@linklings.com
SUMMARY:Co-design and heterogeneous acceleration of molecular dynamics sim
 ulation in GROMACS using HIP
DESCRIPTION:Erik Lindahl (KTH Royal Institute of Technology; National Acad
 emic Infrastructure of Supercomputing in Sweden, Linköping University)\n\n
 Molecular dynamics (MD) is a cornerstone of computational biology and an i
 ncreasingly demanding workload for heterogeneous HPC systems. In this talk
 , I describe how GROMACS is being co-designed with modern accelerator arch
 itectures to deliver scalable, high-performance MD across diverse GPU plat
 forms. I focus in particular on our recent HIP port, which enables first-c
 lass GPU acceleration on AMD hardware while preserving a unified, performa
 nce-oriented code base. Through strict modularization, developers across v
 endors can contribute to and strengthen the same code base.\nI will discus
 s the architectural decisions behind the HIP port, including kernel restru
 cturing, memory-layout design, and abstraction boundaries that balance per
 formance portability with low-level control. These choices allow GROMACS t
 o offload the dominant MD workloads, short-range nonbonded interactions, b
 onded terms, and long-range electrostatics, to GPUs while maintaining stro
 ng scaling through hybrid MPI and OpenMP parallelism. I will also address 
 how heterogeneous acceleration reshapes traditional MD bottlenecks, shifti
 ng attention toward communication overlap, GPU-resident data paths, and la
 tency-sensitive execution.\nAlthough native ports increase porting effort 
 and long-term maintenance, they offer major advantages: deeper hardware op
 timization, a more robust vendor-neutral code base, better cost efficiency
 , stronger driver ecosystems, improved numerical validation, and access to
  the world’s largest HPC resources for scientific discovery.\n\nDomain: En
 gineering, Life Sciences, Computational Methods and Applied Mathematics\n\
 nSession Chairs: Gagandeep Singh (AMD), Sriranjani Sitaraman (AMD), Bertil
  Schmidt (Johannes Gutenberg University Mainz), Kristof Denolf (AMD), Mitt
 ul Singh (AMD), and Yijie Xu (AMD)\n\n
END:VEVENT
END:VCALENDAR