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:20211207T054807Z
LOCATION:220-221
DTSTART;TZID=America/Chicago:20211117T143000
DTEND;TZID=America/Chicago:20211117T150000
UID:submissions.supercomputing.org_SC21_sess147_pap263@linklings.com
SUMMARY:G-SEPM: Building an Accurate and Efficient Soft Error Prediction M
 odel for GPGPUs
DESCRIPTION:Paper\n\nG-SEPM: Building an Accurate and Efficient Soft Error
  Prediction Model for GPGPUs\n\nYue, Wei, Li, Zhao, Jiang...\n\nAs GPUs be
 come ubiquitous in large-scale HPC systems, ensuring the reliable executio
 n of such systems in the presence of soft errors is increasingly essential
 . To assess GPGPU programs' resilience toward soft errors, researchers rel
 y on Random Fault Injection (FI) method. However, it is prohibitively expe
 nsive to obtain a statistically significant resilience profile and not sui
 table for identifying all the critical bits of GPGPU programs.\n\nTo addre
 ss these challenges, in this work, we build a GPGPU-based Soft Error Predi
 ction Model (G-SEPM) to estimate fault site resiliency. We observe that th
 e instruction-type, bit-position, bit-flip direction, and error propagatio
 n chain have capabilities to characterize fault site resiliency. Leveragin
 g these heuristic features, G-SEPM drives out the machine learning model t
 o reveal the hidden interactions among fault site resiliency and our propo
 sed features. Experimental results demonstrate that G-SEPM achieves high a
 ccuracy for fault site error estimation and critical bit identification wh
 ile introducing negligible overhead.\n\nTag: Performance\n\nRegistration C
 ategory: Tech Program Reg Pass
END:VEVENT
END:VCALENDAR