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:20181221T160727Z
LOCATION:D165
DTSTART;TZID=America/Chicago:20181112T103000
DTEND;TZID=America/Chicago:20181112T110000
UID:submissions.supercomputing.org_SC18_sess161_ws_pmbsf106@linklings.com
SUMMARY:Deep Learning at Scale on Nvidia V100 Accelerators
DESCRIPTION:Workshop\nBenchmarks, Parallel Programming Languages, Librarie
 s, and Models, Performance, Simulation, Workshop Reg Pass\n\nDeep Learning
  at Scale on Nvidia V100 Accelerators\n\nXu, Han, Ta\n\nThe recent explosi
 on in the popularity of Deep Learning (DL) is due to a combination of impr
 oved algorithms, access to large datasets and increased computational powe
 r. This had led to a plethora of open-source DL frameworks, each with vary
 ing characteristics and capabilities. End users are then left with the dif
 ficult task of determining software and hardware configuration
 s to get optimal performance from each framework. \n\nWe share our experie
 nces and develop best practices for DL training with TensorFlow, MXNet, an
 d Caffe2. The paper also looks at DL inferencing with TensorRT on Nvidia V
 100 “Volta” GPUs. It focuses on one of the more prominent neural network a
 rchitectures, Resnet50, combined with Imagenet dataset. We quantify the im
 pact of hardware attributes on DL workloads such as the usage of PCIe vs N
 VLink GPUs, performance past a single worker node, effect of high speed in
 terconnect such as InfiniBand EDR on training, and the implication o
 f utilizing a network attached storage and its advantages.
URL:https://sc18.supercomputing.org/presentation/?id=ws_pmbsf106&sess=sess
 161
END:VEVENT
END:VCALENDAR