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:20211207T054812Z
LOCATION:227-228
DTSTART;TZID=America/Chicago:20211118T133000
DTEND;TZID=America/Chicago:20211118T140000
UID:submissions.supercomputing.org_SC21_sess179_pap247@linklings.com
SUMMARY:CAKE: Matrix Multiplication Using Constant-Bandwidth Blocks
DESCRIPTION:Paper\n\nCAKE: Matrix Multiplication Using Constant-Bandwidth 
 Blocks\n\nKung, Natesh, Sabot\n\nWe offer a novel approach to matrix-matri
 x multiplication computation on computing platforms with memory hierarchie
 s. Constant-bandwidth (CB) blocks improve computation throughput for archi
 tectures limited by external memory bandwidth. Configuring the shape and s
 ize of CB blocks operating from within any memory hierarchy level (e.g., i
 nternal SRAM), we achieve high throughput while holding external bandwidth
  (e.g., with DRAM) constant. We explain how, surprisingly, CB blocks can m
 aintain constant external bandwidth as computation throughput increases. A
 nalogous to partitioning a cake into pieces, we dub our CB-partitioned sys
 tem CAKE.\n\nWe show CAKE outperforms state-of-the-art libraries in comput
 ation time on real-world systems where external bandwidth represents a bot
 tleneck, demonstrating CAKE's ability to address the memory wall. CAKE ach
 ieves superior performance by directly using theoretically optimal CB-part
 itioned blocks in tiling and scheduling, obviating the need for extensive 
 design search.\n\nTag: Reproducibility Badge, Algorithms, Parallel Program
 ming Systems\n\nRegistration Category: Tech Program Reg Pass\n\nReproducib
 ility Badges: Artifact Available, Artifact Functional, Results Reproduced
END:VEVENT
END:VCALENDAR
