[ADMB Users] Benchmarks
arnima at hafro.is
Tue Dec 2 14:03:15 PST 2008
I've only performed a crude benchmark using the 'truncreg' example.
Here's how many seconds it took the model to converge and calculate the
Hessian, all compiled and run on the same computer (Intel Core 2 Duo
T7500, 4GB RAM):
4 linux64 gcc424(admb9)
8 win32 bcc52(admb5)
10 win32 vc6(admb5)
11 win32 gcc345(admb9)
13 win32 gcc295(admb5)
The results are quite striking, the main factor being that linux64 runs
on 100% CPU load, using both cores, while WinXP uses only one core.
Further benchmarking would be worthwhile.
On Tue, 2 Dec 2008, Ian Taylor wrote:
> Thanks for your posts to the ADMB Users list. This new era in which we
> can voice issues about ADMB and expect to see results is definitely an
> exciting one.
> Would you be willing to share some of the results of your linux vs
> windows speed comparison--how much gain to you find?
> I've tried to put together a linux vs. windows speed test with ADMB,
> but never had a pair of identical computers to do it on. Using dual
> boot on an old 32 bit single-processor computer (not a good test), the
> results were mixed (with minimization slightly faster on linux, but
> MCMC slower). I did find linux 64 to be faster than linux 32 on a
> different computer. We didn't seem to be a benefit of running the
> windows 32 bit executable on a windows 64 bit machine, but maybe
> they'll release ADMB for windows 64 one of these days. Ultimately,
> however, it seems that nothing makes as big a difference as buying a
> faster computer.
> Also, do you have access to Microsoft Visual C, which seemed to
> provide a definite speed boost over the Borland compiler for windows,
> and might make up any speed deficit of windows-borland vs linux-gcc.
More information about the Users