BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Denver
X-LIC-LOCATION:America/Denver
BEGIN:DAYLIGHT
TZOFFSETFROM:-0700
TZOFFSETTO:-0600
TZNAME:MDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0600
TZOFFSETTO:-0700
TZNAME:MST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260422T000713Z
LOCATION:505
DTSTART;TZID=America/Denver:20231115T163300
DTEND;TZID=America/Denver:20231115T164200
UID:submissions.supercomputing.org_SC23_sess308_spostu109@linklings.com
SUMMARY:Near-Optimal Reduce on the Cerebras Wafer-Scale Engine
DESCRIPTION:Piotr Luczynski (ETH Zürich)\n\nEfficient reduce and allreduce
  communication collectives are crucial building blocks in many workloads, 
 including deep learning training, and have been optimized for various arch
 itectures. We provide the first systematic investigation of the reduce ope
 ration on the Cerebras Wafer-Scale Engine (WSE) using the Cerebras SDK. We
  improve upon existing reduce implementations by up to 5x in certain setti
 ngs. We show that using at most three different implementations we can ach
 ieve performance at most 1.38x slower than an optimal reduction tree. Fina
 lly, we provide an allreduce that outperforms patterns like ring or butter
 fly by up to 2x. \n\nWe will (a) cover unique features of the Cerebras WSE
 , (b) introduce a model to accurately predict performance on the hardware,
  (c) discuss different reduce implementations, (d) analyze the results of 
 running them using an accurate simulator and compare them against an optim
 al reduction tree, (e) show how to extend them to an efficient allreduce.\
 n\nRegistration Category: Tech Program Reg Pass\n\nSession Chair: Ana Gain
 aru (Oak Ridge National Laboratory (ORNL))\n\n
END:VEVENT
END:VCALENDAR