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:20210916T132528Z
LOCATION:Mère Royaume
DTSTART;TZID=Europe/Stockholm:20210707T140000
DTEND;TZID=Europe/Stockholm:20210707T160000
UID:submissions.pasc-conference.org_PASC21_sess155@linklings.com
SUMMARY:Multiprecision Numerics in Scientific High Performance Computing, 
 Part I
DESCRIPTION:Minisymposium\n\nRecently, hardware manufacturers are respondi
 ng to an increasing request for low precision functionality such as FP16 b
 y integrating special low-precision functional units, e.g., NVIDIA Tensor 
 cores. These, however, remain unused even for compute-intensive applicatio
 ns if high precision is employed for all arithmetic operations. At the sam
 e time, communication-intensive applications suffer from the memory bandwi
 dth of architectures growing at a much slower pace than the arithmetic per
 formance. In both cases, a promising strategy is to abandon the high-preci
 sion standard (typically fp64), and employ lower or non-standard precision
  for arithmetic computations or memory operations whenever possible. While
  employing formats other than working precision can render attractive perf
 ormance improvements, it also requires careful consideration of the numeri
 cal effects. On the other end of the spectrum, precision formats with high
 er accuracy than the hardware-supported fp64 can be effective in improving
  the robustness and accuracy of numerical methods. With this breakout mini
 symposium, we aim to create a platform where those working with multipreci
 sion or interested in using multiprecision technology come together and sh
 are their expertise and experience.\n\nAddressing the Memory Wall: Designi
 ng a Multiprecision Ecosystem\n\nAnzt\n\nThe performance of sparse linear 
 algebra is to a large extent constrained by the communication bandwidth, m
 otivating the recent investigation of sophisticated techniques to avoid, r
 educe, and/or hide data transfers in-between processors and between proces
 sors and main memory. One promising strategy ...\n\n---------------------\
 nProperties of GMRES with Iterative Refinement on GPUs\n\nLoe, Glusa, Yama
 zaki, Boman, Rajamanickam\n\nAlgorithms to solve large, sparse linear syst
 ems Ax=b are crucial to many applications.  The popular GMRES algorithm is
  typically memory bound on modern parallel computers.  Multiprecision algo
 rithms reduce the cost of data movement by storing some data in a lower pr
 ecision format.  We implement GMR...\n\n---------------------\nMixed Preci
 sion s-step Lanczos and Conjugate Gradient Algorithms\n\nCarson\n\nCompare
 d to the classical Lanczos algorithm, the s-step Lanczos variant has the p
 otential to improve performance by asymptotically decreasing the synchroni
 zation cost per iteration. However, this comes at a cost. Despite being ma
 thematically equivalent, the s-step variant is known to behave quite di...
 \n\n---------------------\nOpportunities for Approximate vs Transprecision
  Computing in Sparse Linear Solvers for GPUs\n\nQuintana-Orti\n\nThe conve
 ntion in scientific computing is to employ IEEE double-precision (64-bit) 
 arithmetic for all computations involving floating-point data. Nonetheless
 , appealing benefits from the adoption of mixed precision schemes have bee
 n reported for the solution of dense and sparse linear systems on gra...\n
 \n\nDomain: CS and Math
END:VEVENT
END:VCALENDAR