BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20191206T142453Z
LOCATION:Substanz 1\, 2
DTSTART;TZID=Europe/Stockholm:20180626T083000
DTEND;TZID=Europe/Stockholm:20180626T100000
UID:isc_hpc_ISC High Performance 2018_sess113_post116@linklings.com
SUMMARY:(RP10) Deep Learning Hardware Accelerates Fused Discontinuous Gale
 rkin Simulations
DESCRIPTION:Research Poster\n\n(RP10) Deep Learning Hardware Accelerates F
 used Discontinuous Galerkin Simulations\n\nHeinecke, Breuer, Qiu, Cui\n\nI
 n recent years the compute/memory balance of processors has been continuou
 sly shifting towards compute. Especially the rise of Deep Learning, which 
 is based on matrix multiplication, accelerated this path, especially in te
 rms of single precision (FP32) and lower precision compute. An important r
 esearch question is if this development can be leveraged for traditional H
 PC. In this work we demonstrate that a high order discontinuous Galerkin s
 olver for seismic wave equations can execute in single precision without a
 ny loss of accuracy when running application scenarios. Additionally, we e
 xtended its kernels to support the Intel Knights Mill CPU with 14 TFLOPS o
 f FP32 deep-learning performance. Such a performance is achieved by chaini
 ng FMA units and implementing a M=16, N=1, K=4 matrix multiplication in ha
 rdware. This works has two main contributions: a) we executed an extensive
  study employing several wave propagation benchmarks of the seismic commun
 ity. We present misfit norms which prove that the solver can be executed i
 n FP32 while producing negligible differences compared to double precision
  (FP64) b) we sketch a runtime code generator to leverage deep-learning or
 iented instructions (targeting large dense matrix multiplication) to run s
 mall sparse times dense matrix multiplications for several right hand side
 s, called fused simulations. One Knights Mill chip can obtain the same app
 lication performance as the latest top-bin dual socket Intel Xeon Platinum
  HPC server, while consuming significantly lower power. Compared to the HP
 C-focused Knights Landing processor speed-ups of up to 1.6x are possible d
 epending on the scenario.\n\nPasses: Conference Pass, AI/Machine Learning/
 Deep Learning, Compilers, Math Library Design, Parallel Algorithms, Scient
 ific Software Development\n\nTag: Conference Pass, AI/Machine Learning/Dee
 p Learning, Compilers, Math Library Design, Parallel Algorithms, Scientifi
 c Software Development
URL:https://2018.isc-program.com/presentation/?id=post116&sess=sess113
END:VEVENT
END:VCALENDAR