The rise of GPU
Graphical Processing Units (Graphics Cards) have become a useful tool
for fast and efficient numerical computing. In certain areas they are able
to compete with or surpass the previously used GRAPE hardware or reconfigurable
FPGA hardware. In the GRACE project we are currently researching this.
-
The benchmark plot
shows a selected benchmark comparing MPRACE-1, with an
NVIDIA
GeForce GTS8800
.
for the highly specialized and complex pipeline used in Smoothed
Particle Hydrodynamics (SPH) simulations.
From the technical data of
the new MPRACE2 board
we expect it to be faster than the GPU for this application, but this is
subject to further testing. This plot is part of a paper in preparation for
a special issue on Computational Astrophysics
in CSE (Computing in Science and Engineering) for which
the GRACE team has been invited to submit a paper for review.
Our team is a member of the
SPH European Research Interest Community , and has provided
review talks in 2007 and will provide another one in 2008.
-
We are currently involved in
the FRONTIER Project which includes to build a compute cluster using
GPUs - this cluster and project will provide us with a further test and
production environment for the comparative study of applications using
FPGA and GPU, respectively.