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:20260522T150115Z
LOCATION:D220
DTSTART;TZID=America/Chicago:20181112T143000
DTEND;TZID=America/Chicago:20181112T150000
UID:submissions.supercomputing.org_SC18_sess172_ws_phpsc104@linklings.com
SUMMARY:Performance, Power, and Scalability Analysis of the Horovod Implem
 entation of the CANDLE NT3 Benchmark on the Cray XC40 Theta
DESCRIPTION:Xingfu Wu and Valerie Taylor (Argonne National Laboratory, Uni
 versity of Chicago) and Justin M. Wozniak, Rick Stevens, Thomas Brettin, a
 nd Fangfang Xia (Argonne National Laboratory)\n\nTraining scientific deep 
 learning models requires the large amount of computing power provided by H
 PC systems. In this paper, we use the distributed deep learning framework 
 Horovod to parallelize NT3, a Python benchmark from the exploratory resear
 ch project CANDLE (Cancer Distributed Learning Environment). We analyze NT
 3's scalability, performance, and power characteristics with different bat
 ch sizes and learning rates under two memory modes, cache and flat, on the
  DOE pre-exascale production system Cray XC40 Theta at Argonne National La
 boratory. Our experimental results indicate that the power profiles for th
 e node, CPU, and memory are  useful in showing how the Horovod NT3 benchma
 rk behaves on the underlying system. Using the communication timeline of t
 his benchmark, we found that the Horovod communication overhead in NT3 inc
 reases significantly with the number of nodes although Horovod has the abi
 lity to scale up.\n\nThe benchmark leads to smaller runtime and lower powe
 r consumption for the node and CPU under the cache mode than under the fla
 t mode. Furthermore, increasing the batch size leads to a runtime decrease
  and slightly impacts the power. Increasing the learning rate results in a
   slight decrease in runtime and node power and an increase in accuracy. S
 everal issues raised by the Horovod NT3 benchmark results are discussed, a
 nd suggestions are proposed for further work.\n\nTag: Parallel Application
  Frameworks, Reproducibility, Scientific Computing\n\nRegistration Categor
 y: Workshop Reg Pass\n\n
END:VEVENT
END:VCALENDAR