BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20210916T132452Z
LOCATION:Mère Royaume
DTSTART;TZID=Europe/Stockholm:20210707T150000
DTEND;TZID=Europe/Stockholm:20210707T153000
UID:submissions.pasc-conference.org_PASC21_sess155_msa311@linklings.com
SUMMARY:Opportunities for Approximate vs Transprecision Computing in Spars
 e Linear Solvers for GPUs
DESCRIPTION:Minisymposium\n\nOpportunities for Approximate vs Transprecisi
 on Computing in Sparse Linear Solvers for GPUs\n\nQuintana-Orti\n\nThe con
 vention in scientific computing is to employ IEEE double-precision (64-bit
 ) arithmetic for all computations involving floating-point data. Nonethele
 ss, appealing benefits from the adoption of mixed precision schemes have b
 een reported for the solution of dense and sparse linear systems on graphi
 cs processing units (GPUs) via iterative refinement. In this talk, we will
  illustrate the benefits of a generalization of the mixed precision strate
 gy, known as Transprecision Computing (TC), in terms of execution time and
  energy efficiency. For this purpose, we will employ several case studies 
 arising in the iterative solution of sparse linear systems on GPUs, with c
 odes currently integrated the Ginkgo library (https://ginkgoproject.github
 .io <https://ginkgoproject.github.io>). In some detail, this research effo
 rt exploits the fact that, for sparse linear algebra operations, the cost 
 is dominated by the memory accesses while the arithmetic is largely irrele
 vant. To leverage this property, the Ginkgo solvers store certain parts of
  the data in reduced precision in memory, but operate in "full" 64-bit pre
 cision in order to bound the accumulation of rounding errors. Reduced-prec
 ision storage can be leveraged to maintain approximation operators, such a
 s a preconditioner, or in a solver that gradually augments the precision o
 f the operands as the iteration converges to the solution.\n\nDomain: CS a
 nd Math
END:VEVENT
END:VCALENDAR