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:20260422T143140Z
LOCATION:B302
DTSTART;TZID=America/New_York:20241117T162000
DTEND;TZID=America/New_York:20241117T164500
UID:submissions.supercomputing.org_SC24_sess744_ws_memo101@linklings.com
SUMMARY:Sum Reduction with OpenMP Offload on NVIDIA Grace-Hopper System
DESCRIPTION:Zheming Jin (Oak Ridge National Laboratory (ORNL))\n\nWe evalu
 ate the performance of the baseline and optimized reductions in OpenMP on 
 an NVIDIA Grace-Hopper system. We explore the impacts of the number of tea
 ms, the number of elements to sum per loop iteration, and simultaneous exe
 cution on the central-processing unit (CPU) and the GPU in the unified mem
 ory (UM) mode upon the reduction performance. The experimental results sho
 w that the optimized reductions are 6.120X to 20.906X faster than the base
 lines on the GPU, and their efficiency ranges from 89% to 95% of the theor
 etical GPU memory bandwidth. Depending on where an input array is allocate
 d in the program when co-running the reduction on the CPU and GPU in the U
 M mode, the average speedup over the GPU-only execution is approximately 2
 .484 or 1.067, and the speedup of the optimized reductions over the baseli
 ne reductions ranges from 0.996 to 10.654 or from 0.998 to 6.729.\n\nTag: 
 Data Movement and Memory, Emerging Technologies\n\nRegistration Category: 
 Workshop Reg Pass\n\nSession Chairs: Ron Brightwell (Sandia National Labor
 atories), Maya Gokhale (Lawrence Livermore National Laboratory (LLNL)), Ky
 le Hale (Oregon State University), and Ivy Peng (KTH Royal Institute of Te
 chnology)\n\n
END:VEVENT
END:VCALENDAR