[ADMB Users] Benchmarks

[Arni Magnusson]

I've performed some additional benchmarks, this time using the 'catage' 
example. Here's how many seconds it took the model to converge and run 
100 thousand MCMC cycles, all on the same machine (Intel Core 2 Duo 
T7500, 4GB RAM):

12 linux64-gcc346-9065, compiled with gcc424
19 win32-gcc345-90652, compiled with gcc345
20 win32-gcc432-802, compiled with gcc421
24 win32-vc6-501, compiled with vc6
32 win32-bcc551-9065, compiled with bcc551
38 win32-bcc52-502, compiled with bcc52
38 win32-gcc295-501, compiled with gcc2952

Like in the previous benchmark, linux64 uses both cores and runs much 
faster than the win32 (WinXP) models. Unlike the 'truncreg' benchmark, 
the GCC 345/432 models run faster than Borland 52/55 and Microsoft VC6. 
It looks like GCC is better at MCMCing 'catage', but Borland and 
Microsoft are better at handling the large 'truncreg' dataset.

I don't have a more recent version of a Microsoft compiler. Feel free to 
send me win32-vc7-9065 compiled versions of 'simple', 'truncreg', and 
'catage' if you're interested in adding it to this benchmark.

For the sake of completeness, the commands I've used in the benchmarks 
are "truncreg -ams 1000000" and "catage -mcmc 100000 -mcsave 100".

Arni