Lattice QCD codes on AMD Opteron processor

Miho KOMA
DESY

Lattice QCD is a powerful method to study Quantum chromodynamics (QCD)
in a nonperturbative way.
As computer technology advances PC clusters can also be used for lattice
QCD simulations as well as a number of commercial supercomputers.
Since  lattice QCD simulations demand huge computer power, it is very
important to optimize the simulation codes so as to exploit the full
potential of the processor.
 
In this talk, we report our experience on the optimization of the
Lattice QCD codes with Streaming SIMD extensions (SSE) instruction sets
for the AMD Opteron processor.
We focus on the CPU dependence of the optimization by comparing Intel
Xeon with AMD Opteron processor. Performance test results for some
basic routines before and after optimization for both processors are
shown.