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:20230124T171524Z
LOCATION:C155
DTSTART;TZID=America/Chicago:20221114T144500
DTEND;TZID=America/Chicago:20221114T150000
UID:submissions.supercomputing.org_SC22_sess453_ws_pmbss110@linklings.com
SUMMARY:Time-Series ML-Regression on Graphcore IPU-M2000 and Nvidia A100
DESCRIPTION:Workshop\n\nTime-Series ML-Regression on Graphcore IPU-M2000 a
 nd Nvidia A100\n\nBalewski, Liu, Tsyplikhin, Roland, Bouchard\n\nWe compar
 e the ML-training performance of a Graphcore IPU-M2000-based system with N
 vidia A100 GPU-based system on the Perlmutter HPC machine at NERSC/LBL. Th
 e multivariate regression of time series data from a simulated biological 
 neuron was the scientific benchmark problem. The ML-model consisted of sev
 eral convolutional, batch normalization, and fully connected layers. The t
 raining data were distributed in CPUs memory to eliminate the system depen
 dent IO cost. The data-parallel training runs resulted in the same samples
  throughput on both GC200 IPUs and A100 GPUs for any choice of the number 
 of accelerators between 1 and 256. The achieved best MSE validation loss o
 n IPUs was only 10% to 20% larger. The aggregated energy use per 1 trainin
 g epoch was between 2.5 to 3 times smaller for the Graphcore-system in com
 parison to the Nvidia-system. This paper also discusses aspects of softwar
 e-hardware co-design to achieve highest efficiency on the IPU using PopTor
 ch.\n\nSession Format: Recorded\n\nTag: Applications, Architectures, Bench
 marking, Exascale Computing, Modeling and Simulation, Performance, Perform
 ance Portability\n\nRegistration Category: Workshop Reg Pass
END:VEVENT
END:VCALENDAR