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:20250626T234542Z
LOCATION:B312-B313A
DTSTART;TZID=America/New_York:20241119T113000
DTEND;TZID=America/New_York:20241119T120000
UID:submissions.supercomputing.org_SC24_sess496_gb105@linklings.com
SUMMARY:Toward Capturing Genetic Epistasis from Multivariate Genome-Wide A
 ssociation Studies Using Mixed-Precision Kernel Ridge Regression
DESCRIPTION:Hatem Ltaief (King Abdullah University of Science and Technolo
 gy (KAUST)); Rabab Alomairy (Massachusetts Institute of Technology (MIT));
  Qinglei Cao (Saint Louis University); Jie Ren (King Abdullah University o
 f Science and Technology (KAUST)); Lotfi Slim, Thorsten Kurth, and Benedik
 t Dorschner (NVIDIA Corporation); Salim Bougouffa (King Abdullah Universit
 y of Science and Technology (KAUST)); Rached Abdelkhalek (NVIDIA Corporati
 on); and David E. Keyes (King Abdullah University of Science and Technolog
 y (KAUST), Columbia University)\n\nWe exploit the widening margin in tenso
 r-core performance between [FP64/FP32/FP16/INT8,FP64/FP32/FP16/FP8/INT8] o
 n NVIDIA [Ampere,Hopper] \nGPUs to boost the performance of output accurac
 y-preserving mixed-precision computation of Genome-Wide Association Studie
 s (GWAS) of 305,000 patients from the UK Biobank, the largest-ever GWAS co
 hort studied for genetic epistasis using a multivariate approach. Tile-cen
 tric adaptive-precision linear algebraic techniques motivated by reducing 
 data motion gain enhanced significance with low-precision GPU arithmetic. 
 At the core of Kernel Ridge Regression (KRR) techniques for GWAS lie compu
 te-bound cubic-complexity matrix operations that inhibit scaling to aspira
 tional dimensions of the population, genotypes, and phenotypes. We acceler
 ate KRR matrix generation by redesigning the computation for Euclidean dis
 tances to engage INT8 tensor cores while exploiting symmetry. We accelerat
 e solution of the regularized KRR systems by deploying a new four-precisio
 n Cholesky-based solver, which, at 1.805 mixed-precision ExaOp/s on a near
 ly full Alps system, outperforms the state-of-the-art CPU-only REGENIE GWA
 S software by five orders of magnitude.\n\nRegistration Category: Tech Pro
 gram Reg Pass\n\nSession Chair: Barbara Chapman (Hewlett Packard Enterpris
 e (HPE), Stony Brook University)\n\n
END:VEVENT
END:VCALENDAR