BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Chicago
X-LIC-LOCATION:America/Chicago
BEGIN:DAYLIGHT
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
TZNAME:CDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
TZNAME:CST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20211207T055411Z
LOCATION:Online
DTSTART;TZID=America/Chicago:20211119T105000
DTEND;TZID=America/Chicago:20211119T111000
UID:submissions.supercomputing.org_SC21_sess329_ws_pawatm106@linklings.com
SUMMARY:Evaluation of Two Topology-Aware Heuristics on Level-3 BLAS Librar
 y for Multi-GPU Platforms
DESCRIPTION:Workshop\n\nEvaluation of Two Topology-Aware Heuristics on Lev
 el-3 BLAS Library for Multi-GPU Platforms\n\nGautier, Ferreira Lima\n\nNow
 adays GPUs have dominated the market considering the computing/power metri
 c and numerous research works have provided Basic Linear Algebra Subprogra
 ms implementations accelerated on GPUs. Several software libraries have be
 en developed for exploiting performance of systems with accelerators, but 
 the real performance may be far from the platform peak performance with mu
 ltiple GPUs.  This paper presents two runtime heuristics to gain in perfor
 mance when task based programs are performed on heterogeneous architecture
  such as multi-GPU systems. The first is a topology-aware policy to takes 
 into account the heterogeneity of the high speed links that interconnect G
 PUs.  The second is an optimistic heuristic that favors communication betw
 een devices.  These have been implemented in the XKBLAS library BLAS-3 lib
 rary.  We made experiments on a NVIDIA DGX-1 with up to 8 GPUs V100 on a s
 et of Basic Linear Algebra Subroutines.  Experimental results on kernels s
 howed that XKBlas outperformed most implementations including the overhead
  of creation and scheduling of dynamic tasks.\n\nTag: Online Only, Applica
 tions, Extreme Scale Computing, Heterogeneous Systems, Parallel Programmin
 g Languages and Models, Software Engineering\n\nRegistration Category: Wor
 kshop Reg Pass
END:VEVENT
END:VCALENDAR