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:20181221T160731Z
LOCATION:C140/142
DTSTART;TZID=America/Chicago:20181114T153000
DTEND;TZID=America/Chicago:20181114T160000
UID:submissions.supercomputing.org_SC18_sess210_pap277@linklings.com
SUMMARY:The Design, Deployment, and Evaluation of the CORAL Pre-Exascale S
 ystems
DESCRIPTION:Paper\nArchitectures, Networks, Performance, Scientific Comput
 ing, State of the Practice, Tools, Tech Program Reg Pass\n\nThe Design, De
 ployment, and Evaluation of the CORAL Pre-Exascale Systems\n\nVazhkudai, d
 e Supinski, Bland, Geist, Sexton...\n\nCORAL, the Collaboration of Oak Rid
 ge, Argonne and Livermore, is fielding two similar IBM systems, Summit and
  Sierra, with NVIDIA GPUs that will replace the existing Titan and Sequoia
  systems. Summit and Sierra are currently ranked No. 1 and No. 3, respecti
 vely, on the Top500 list. We discuss the design and key differences of the
  systems. Our evaluation of the systems highlights the following. Applicat
 ions that fit in HBM see the most benefit and may prefer more GPUs; howeve
 r, for some applications, the CPU-GPU bandwidth is more important than the
  number of GPUs. The node-local burst buffer scales linearly, and can achi
 eve a 4X improvement over the parallel file system for large jobs; smaller
  jobs, however, may benefit from writing directly to the PFS. Finally, sev
 eral CPU, network and memory bound analytics and GPU-bound deep learning c
 odes achieve up to a 11X and 79X speedup/node, respectively over Titan.
URL:https://sc18.supercomputing.org/presentation/?id=pap277&sess=sess210
END:VEVENT
END:VCALENDAR