BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/New_York
X-LIC-LOCATION:America/New_York
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260422T143139Z
LOCATION:B304
DTSTART;TZID=America/New_York:20241118T095000
DTEND;TZID=America/New_York:20241118T100500
UID:submissions.supercomputing.org_SC24_sess749_ws_drbsd118@linklings.com
SUMMARY:FRSZ2 for In-Register Block Compression Inside GMRES on GPUs
DESCRIPTION:Thomas Grützmacher (Karlsruhe Institute of Technology (KIT), T
 echnical University of Munich); Robert Underwood, Sheng Di, and Franck Cap
 pello (Argonne National Laboratory (ANL), University of Chicago); and Hart
 wig Anzt (Technical University of Munich; University of Tennessee, Innovat
 ive Computing Laboratory (ICL))\n\nThe performance of the GMRES iterative 
 solver on GPUs is limited by the GPU main memory bandwidth. Compressed Bas
 is GMRES outperforms GMRES by storing the Krylov basis in low precision, t
 hereby reducing the memory access. An open question is whether compression
  techniques that are more sophisticated than casting to low precision can 
 enable large runtime savings while preserving the accuracy of the final re
 sults. This paper presents the lightweight in-register compressor \frsz th
 at can decompress at the bandwidth speed of a modern NVIDIA H100 GPU. In a
 n experimental evaluation, we demonstrate using \frsz instead of low preci
 sion for compression of the Krylov basis can bring larger runtime benefits
  without impacting final accuracy.\n\nTag: Data Compression, Data Movement
  and Memory, Middleware and System Software\n\nRegistration Category: Work
 shop Reg Pass\n\nSession Chairs: Sheng Di (Argonne National Laboratory (AN
 L), University of Chicago); Ana Gainaru (Oak Ridge National Laboratory (OR
 NL)); Sian Jin (Temple University); Xin Liang (Oregon State University); K
 ento Sato (RIKEN Center for Computational Science (R-CCS)); and Dingwen Ta
 o (Institute of Computing Technology, Chinese Academy of Sciences; Univers
 ity of Chinese Academy of Sciences)\n\n
END:VEVENT
END:VCALENDAR