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:20260202T201809Z
LOCATION:265
DTSTART;TZID=America/Chicago:20251116T105000
DTEND;TZID=America/Chicago:20251116T111000
UID:submissions.supercomputing.org_SC25_sess219_ws_resdis104@linklings.com
SUMMARY:TEGRA - Scaling Up Graph Processing with Disaggregated Computing
DESCRIPTION:William Shaddix and Mahyar Samani (University of California, D
 avis); Jason Lowe-Power (Google LLC; University of California, Davis); and
  Venkatesh Akella (University of California, Davis)\n\nGraph processing wo
 rkloads continue to grow in scale and complexity, demanding architectures 
 that can adapt to diverse compute and memory requirements. Traditional sca
 le-out accelerators couple compute and memory resources, resulting in reso
 urce underutilization when executing workloads with varying compute-to-mem
 ory intensities. In this paper, we present TEGRA, a composable, scale-up a
 rchitecture for large-scale graph processing. TEGRA leverages disaggregate
 d memory via CXL and a message-passing communication model to decouple com
 pute and memory, enabling independent scaling of each. Through detailed ev
 aluation using the gem5 simulator, we show that TEGRA improves memory band
 width utilization by up to 15\% over state-of-the-art accelerators by dyna
 mically provisioning compute based on workload demands. Our results demons
 trate that TEGRA provides a flexible and efficient foundation for supporti
 ng emerging graph analytics workloads across a wide range of arithmetic in
 tensities.\n\nRecording: Livestreamed, Recorded\n\nRegistration Category: 
 Technical Program Reg Pass, Workshop Reg Pass\n\nSession Chairs: Balazs Ge
 rofi (Intel Corporation, RIKEN Center for Computational Science (R-CCS)); 
 Christian Pinto (IBM); and John Shalf (Lawrence Berkeley National Laborato
 ry (LBNL))\n\n
END:VEVENT
END:VCALENDAR
