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:20260522T150117Z
LOCATION:C141/143/149
DTSTART;TZID=America/Chicago:20181115T143000
DTEND;TZID=America/Chicago:20181115T150000
UID:submissions.supercomputing.org_SC18_sess192_pap430@linklings.com
SUMMARY:PRISM: Predicting Resilience of GPU Applications Using Statistical
  Methods
DESCRIPTION:Charu Kalra, Fritz Previlon, and Xiangyu Li (Northeastern Univ
 ersity); Norman Rubin (Nvidia Corporation); and David Kaeli (Northeastern 
 University)\n\nAs Graphics Processing Units (GPUs) become more pervasive i
 n HPC and safety-critical domains, ensuring that GPU applications can be p
 rotected from data corruption grows in importance. Despite prior efforts t
 o mitigate errors, we still lack a clear understanding of how resilient th
 ese applications are in the presence of transient faults.  Due to the rand
 om nature of these faults, predicting whether they will alter the program 
 output is a challenging problem. In this paper, we build a framework named
  PRISM, which uses a systematic approach to predict failures in GPU progra
 ms. PRISM extracts micro-architecture agnostic features to characterize pr
 ogram resiliency, which serve as predictors in our statistical model. PRIS
 M enables us to predict failures in applications without running exhaustiv
 e fault-injection campaigns on a GPU, thereby reducing the error estimatio
 n effort. PRISM can also be used to gain insight into potential architectu
 ral support required to improve the reliability of GPU applications.\n\nTa
 g: Algorithms, Architectures, GPUs, Linear Algebra, Networks, Resiliency\n
 \nRegistration Category: Tech Program Reg Pass\n\nSession Chair: Steven A.
  Wright (University of York, England)\n\n
END:VEVENT
END:VCALENDAR