Hello,
since I somehow initiated the debate, I feel obliged to partly correct
and elaborate my statements about the performance of different
compilers:
I am still doing tests with the good old linpack benchmark, just to
compare present day machines and old computers. And this benchmark
gives results quite contrary to the figures at Polyhedron:
For n=200 (dimension of problem) the DVF6.0 compiler displays 75-98
mflops on a 450PIII, while for the linux compilers I get
55-61 mflops for the Fujitsu compiler
and 81-86 mflops for the NAG f95 compiler.
Increasing n to 1024 gives over 200 mflops for the DVF compiler, but
numerical problems (underflow and overflow) for the Linux compilers.
Similar behaviour is observed for the LAPACK timing routines, where
the best results are obtained by DVF.
Best regards
Alois
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|