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:20260422T143138Z
LOCATION:B303
DTSTART;TZID=America/New_York:20241118T160000
DTEND;TZID=America/New_York:20241118T163000
UID:submissions.supercomputing.org_SC24_sess748_ws_pmbsf101@linklings.com
SUMMARY:Assessing the GPU Offload Threshold of GEMM and GEMV Kernels on Mo
 dern Heterogeneous HPC Systems
DESCRIPTION:Finn Wilkinson, Alex Cockrean, Wei-Chen Lin, Simon McIntosh-Sm
 ith, and Tom Deakin (University of Bristol, England)\n\nWith an ever-growi
 ng compute advantage over CPUs, GPUs are often used in workloads with ampl
 e BLAS computation to improve performance. However, several factors, inclu
 ding data-to-compute ratio, amount of data re-use, and data structure, can
  all impact performance. Hence, using a GPU is not a guarantee of better B
 LAS performance. In this work, we introduce the GPU BLAS Offload Benchmark
  (GPU-BLOB), a novel and portable benchmark that measures CPU and GPU comp
 ute performance of different BLAS kernels and problem configurations. From
  the GPU offload threshold (a BLAS kernel’s minimum dimensions for a certa
 in configuration where using a GPU is guaranteed to yield improved perform
 ance), we evaluate the per-node performance of three in-production HPC sys
 tems. We show that the offload threshold for GEMM is highly dependent on p
 roblem shape and number of consecutive BLAS calls, and that, contrary to c
 onventional wisdom, GEMV can benefit from GPU acceleration, especially on 
 SoC-based systems.\n\nTag: Accelerators, Modeling and Simulation, Performa
 nce Evaluation and/or Optimization Tools\n\nRegistration Category: Worksho
 p Reg Pass\n\nSession Chairs: Simon Hammond (National Nuclear Security Adm
 inistration (NNSA)); Sascha Hunold (Technical University of Vienna); and S
 teven A. Wright (University of York, England)\n\n
END:VEVENT
END:VCALENDAR
