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:20211207T054813Z
LOCATION:220-221
DTSTART;TZID=America/Chicago:20211118T153000
DTEND;TZID=America/Chicago:20211118T160000
UID:submissions.supercomputing.org_SC21_sess173_pap595@linklings.com
SUMMARY:ndzip-gpu: Efficient Lossless Compression of Scientific Floating-P
 oint Data on GPUs
DESCRIPTION:Paper\n\nndzip-gpu: Efficient Lossless Compression of Scientif
 ic Floating-Point Data on GPUs\n\nKnorr, Thoman, Fahringer\n\nLossless dat
 a compression is a promising software approach for reducing the bandwidth 
 requirements of scientific applications on accelerator clusters without in
 troducing approximation errors.  Suitable compressors must be able to effe
 ctively compact floating-point data while saturating the system interconne
 ct to avoid introducing unnecessary latencies.\n\nWe present ndzip-gpu, a 
 novel, highly-efficient GPU parallelization scheme for the block compresso
 r ndzip, which has recently set a new milestone in CPU floating-point comp
 ression speeds.\n\nThrough the combination of intra-block parallelism and 
 efficient memory access patterns, ndzip-gpu achieves high resource utiliza
 tion in multi-dimensional data decorrelation.  We further introduce an eff
 icient warp-cooperative primitive for vertical bit packing, providing a hi
 gh-throughput data reduction step.\n\nUsing a representative set of scient
 ific data, we demonstrate that ndzip-gpu consistently outperforms all othe
 r lossless floating-point compressors accessible to us on NVIDIA Volta and
  Ampere hardware in both throughput and compression ratio achieved.\n\nTag
 : Reproducibility Badge, Algorithms, Cloud and Distributed Computing, Data
  Management, Parallel Programming Languages and Models\n\nRegistration Cat
 egory: Tech Program Reg Pass\n\nReproducibility Badges: Artifact Available
 , Artifact Functional, Results Reproduced
END:VEVENT
END:VCALENDAR