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:20260422T143138Z
LOCATION:B312
DTSTART;TZID=America/New_York:20241117T163000
DTEND;TZID=America/New_York:20241117T165000
UID:submissions.supercomputing.org_SC24_sess739_ws_ss106@linklings.com
SUMMARY:Input-Dependent Power Usage in GPUs
DESCRIPTION:Theo Gregersen (University of Washington, Carnegie Mellon Univ
 ersity); Pratyush Patel (University of Washington); and Esha Choukse (Micr
 osoft Corporation)\n\nGPUs are known to be power-hungry, and due to the bo
 om in artificial intelligence, they are the major contributors to the high
  power demands of datacenters. Most GPU usage in these popular workloads c
 onsists of large general matrix-matrix multiplications (GEMMs), which have
  therefore been optimized to achieve high utilization of hardware resource
 s.\n\nWe show that modifying the input data to GEMMs, while maintaining th
 e matrix shapes and sizes can notably change the power consumption of thes
 e kernels. We experiment with four kinds of input variations: value distri
 bution, bit similarity, placement, and sparsity, across different data typ
 es. Our findings indicate that these variations can change the GPU power u
 sage during GEMM by almost 40%.\n\nWe hypothesize that input-dependent pow
 er usage variations occur due to changes in the number of bit flips in the
  GPUs. We propose leveraging this property through compiler and scheduler 
 optimizations to manage power and reduce energy consumption.\n\nTag: Energ
 y Efficiency, HPC Infrastructure, Sustainability\n\nRegistration Category:
  Workshop Reg Pass\n\nSession Chairs: Cate Berard (US Department of Energy
 ); James H. Rogers (Oak Ridge National Laboratory (ORNL)); Fumiyoshi Shoji
  (RIKEN Center for Computational Science (R-CCS), Center for Computational
  Science); Michèle Weiland (EPCC, The University of Edinburgh; The Univers
 ity of Edinburgh); and Mike Woodacre (Hewlett Packard Enterprise (HPE))\n\
 n
END:VEVENT
END:VCALENDAR