Perfomance Tests on INTEL Core2 Duo and IBM POWER5+(R6000)

[cyc99]

I compiled vasp on INTEL Core2 Duo using ifort and mkl.
(core clock speed is 1.6 GHz and L2 cache size is 2MB per core)
Below Options and Libraries were used.

FC= ifort
OFLAGS= -O3 -xW -tpp7
BLAS= -L/.../ -lmkl_em64t -lguide -lpthread
LAPACK= -L../vasp.4.lib/lapack_double.o
LINK = -L/.../ -lsvml
FFTW= fftw3d.o fft3dlib.o libfftw3.a (compiled using ifort)

Binary file worked well, but I'm wonder whether I can drastically further optimize or not.

Actually, I also compiled vasp on IBM(POWER5+) system, whose core clock speed is 1.9 GHz and cache size is about ten times larger than above system, using optimized libraries written in makefiel.rs6000.

Then, I ran test jobs on both machines.
In the results, the elapsed time were 1100sec and 22sec for Intel Core2Duo and IBM systems, respectively.

Is it normal results?
I think 50 times difference is too much.
So I wanna to know whether I can further optimize or not on Intel Core2Duo.

Thanks.

[cyc99]

Sorry for my mistake. I tested different job on each machine. So, I tested the same jobs on both machines, then the elapsed time was 60sec and 22sec for Intel Core2Duo and IBM systems, respectively.
Anyway, I hope this banchmark will help vasp users. Thanks.

[admin]

dear colleague, thank you for the update of your message!