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:20260202T201247Z
LOCATION:Second Floor Atrium
DTSTART;TZID=America/Chicago:20251120T080000
DTEND;TZID=America/Chicago:20251120T170000
UID:submissions.supercomputing.org_SC25_sess533_post272@linklings.com
SUMMARY:Scaling Singular Values Beyond GPU Memory Limits: Out-of-Core, GPU
 -Accelerated, and Unified Across Data Precision and Hardware
DESCRIPTION:Evelyne Ringoot (Massachusetts Institute of Technology (MIT))\
 n\nWe present a unified, out-of-core, GPU-accelerated singular value solve
 r that achieves performance portability across diverse hardware platforms 
 and data precisions for datasets exceeding GPU memory. The singular value 
 decomposition (SVD) is fundamental for processing large-scale datasets, ye
 t the diversity of computing architectures and the proliferation of precis
 ion formats pose significant challenges in heterogeneous environments. Tra
 ditional HPC libraries require separate implementations for each architect
 ure and precision, limiting scalability and usability. Building on our pre
 vious work, where we developed an open-source unified solver achieving per
 formance comparable to vendor-optimized libraries across multiple precisio
 ns and GPU platforms, we extend this capability to handle larger-than-memo
 ry datasets. We adapt a QR-based communication-hiding strategy to improve 
 the compute-to-communication ratio and leverage Julia's multiple-dispatch 
 for seamless backend integration. Our implementation significantly outperf
 orms CPU-based LAPACK and remains only 3–5× slower than GPU-resident solve
 rs across different hardware and data precision configurations.\n\nTag: Re
 search & ACM SRC Posters\n\nRegistration Category: Technical Program Reg P
 ass\n\nSession Chairs: Kento Sato (RIKEN Center for Computational Science 
 (R-CCS)); Chris Schlipalius (Pawsey Supercomputing Research Centre; Common
 wealth Scientific and Industrial Research Organisation (CSIRO), Australia)
 ; and Anja Gerbes (Georg-August-Universität Göttingen)\n\n
END:VEVENT
END:VCALENDAR