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:20260522T150119Z
LOCATION:C140/142
DTSTART;TZID=America/Chicago:20181114T143000
DTEND;TZID=America/Chicago:20181114T150000
UID:submissions.supercomputing.org_SC18_sess204_pap133@linklings.com
SUMMARY:High-Performance Dense Tucker Decomposition on GPU Clusters
DESCRIPTION:Jee Choi (IBM), Xing Liu (Intel Corporation), and Venkatesan C
 hakaravarthy (IBM)\n\nThe Tucker decomposition method is one of the most p
 opular algorithms for analyzing and compressing data with multi-way relati
 onship. Its execution time is typically dominated by dense matrix multipli
 cation, which makes it well-suited for GPU acceleration. State-of-the-art 
 distributed dense Tucker implementations for CPU clusters adopt multi-dime
 nsional partitioning that optimizes for storage and communication. This, h
 owever, leads to smaller matrix dimensions that result in under-utilizing 
 the GPU. \n\nIn this paper, we present our optimized implementation and pe
 rformance analysis of dense Tucker decomposition on a multi-GPU cluster. W
 e propose three optimizations: a new partitioning strategy that improves G
 PU performance, a new tensor matricization layout that halves the number o
 f communication/matricization steps, and a variation of the randomized SVD
  algorithm to overcome the eigenvalue bottleneck that arises from the high
  speedups gained from GPU acceleration.  Our GPU implementation employing 
 all three optimizations achieves up to 11.8x speedup on 64 nodes over stat
 e-of-the-art TuckerMPI.\n\nTag: Algorithms, Applications, Computational Ph
 ysics, Scientific Computing\n\nRegistration Category: Tech Program Reg Pas
 s\n\nSession Chair: Erik Draeger (Lawrence Livermore National Laboratory (
 LLNL), Center for Applied Scientific Computing)\n\n
END:VEVENT
END:VCALENDAR