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:20211207T055347Z
LOCATION:Online
DTSTART;TZID=America/Chicago:20211115T093000
DTEND;TZID=America/Chicago:20211115T100000
UID:submissions.supercomputing.org_SC21_sess422_ws_pmbsf108@linklings.com
SUMMARY:Architectural Requirements for Deep Learning Workloads in HPC Envi
 ronments
DESCRIPTION:Workshop\n\nArchitectural Requirements for Deep Learning Workl
 oads in HPC Environments\n\nIbrahim, Nguyen, Nam, Bhimji, Farrell...\n\nSc
 ientific machine learning (SciML) promises to have a transformational impa
 ct on scientific exploration, by combining state-of-the-art AI methods wit
 h the latest generation of supercomputers. To efficiently leverage ML tech
 niques on high-performance computing (HPC) systems, however, it is critica
 l to understand the performance characteristics of the underlying algorith
 ms on modern computational systems.  In this work, we present a new method
 ology for developing a detailed performance understanding of ML benchmarks
 .  To demonstrate our approach we investigate two emerging SciML benchmark
  applications from cosmology and climate; ComsoFlow and DeepCAM; as well a
 s ResNet-50, a well-known image classification model.  We develop and vali
 date performance models that explore the key architectural artifacts, incl
 uding memory requirements, data reuse and performance efficiency across bo
 th single- and multiple-GPU computations. Our methodology focuses on the c
 omplexity of data-movement across storage and memory hierarchies, and leve
 rages our performance models to capture key components of runtime executio
 n while highlighting design tradeoffs.\n\nTag: Online Only, Accelerator-ba
 sed Architectures, Applications, Computational Science, Emerging Technolog
 ies, Extreme Scale Computing, File Systems and I/O, Heterogeneous Systems,
  Parallel Programming Languages and Models, Performance, Scientific Comput
 ing, Software Engineering\n\nRegistration Category: Workshop Reg Pass
END:VEVENT
END:VCALENDAR