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:20210916T132448Z
LOCATION:Louis Favre
DTSTART;TZID=Europe/Stockholm:20210705T163000
DTEND;TZID=Europe/Stockholm:20210705T170000
UID:submissions.pasc-conference.org_PASC21_sess111_msa185@linklings.com
SUMMARY:A Multi-GPU Supercomputing Framework for 3D High-Resolution Forwar
d and Inverse Iterative Solvers in Julia
DESCRIPTION:Minisymposium\n\nA Multi-GPU Supercomputing Framework for 3D H
igh-Resolution Forward and Inverse Iterative Solvers in Julia\n\nRĂ¤ss, Reu
ber, Holbach, Omlin\n\nAlongside fault zones, porous rocks exhibit efficie
nt pathways for fluids to migrate from depth to surface. Nonlinear hydrome
chanical processes suggest that buoyant fluid pockets trapped within creep
-activated porous host rocks can accumulate and migrate to the surface as
solitary waves of porosity. The porosity distribution of such systems is m
ostly unknown while highly relevant for carbon dioxide sequestration or oi
l and gas operations. Seismic imaging methods unveil vertical chimneys by
inverting for an effective seismic wave speed. However, these seismic chim
neys may not directly be mapped to a porosity distribution. Here, we inver
t for the 3-D porosity field calculating the pointwise gradients of the ve
rtical fluid flux at the surface with respect to the porosity using an adj
oint method. The forward and adjoint problems are solved using an implicit
finite difference scheme and a pseudo-transient solver. The iterative and
matrix-free solver only requires local information and thus fully exploit
s the potential of GPUs. We present a Julia-based 3-D MPI-GPU framework to
solve the forward and adjoint problems as well as the gradient equation.
We assess the performance of the 3-D memory-bounded solvers using a simple
metric. Our development relies on the ParallelStencil.jl and ImplicitGoba
lGrid.jl packages enabling high-performance stencil-based calculations and
optimal distributed memory parallelisation.\n\nDomain: CS and Math, Clima
te and Weather, Physics, Solid Earth Dynamics
END:VEVENT
END:VCALENDAR