The school of 'high performance computing on accelerator board'
was held on December 7-9, 2009 at KEK Computing Research Center.
Many kinds of numerical computations in physics, such as
theoretical and experimental simulations, require enormous time
to get reliable results. People thus have studied how to
accelerate these computations in various ways. An accelerator
board has recently been drawing much attention because of
its good cost performance.
An accelerator board consists of many processors which can
execute the computation in parallel, and it can be attached
as an extension board to a personal computer. The processing
chip on the board has a variety; one developed for dedicated
purposes, for games, and for computer graphics, and so on.
Among them, the graphic boards, so-called GPUs, have become
a current trend because of not only its high performance,
but also convenient programming environment.
In this school, we focused on two types of the accelerator
boards: a commercial graphic board and GRAPE-DR. The latter
has been developed originally for the gravitational many-body
simulations at University of Tokyo and National Astronomical
Observatory of Japan.
The main purpose of this school is to learn about the hardware
features of the accelerator boards and programming methods so
as to make the best use of their performance. The school
consists of several lectures and practical training course
for beginners.
About twenty students and researchers participated from various
fields in physics as well as from computer science. In the
training, they enjoyed the speedup of computer programs by
GPU and GRAPE-DR. The school would have provided a first step
toward high performance computing on accelerator boards.
|