[ADMB Users] Benchmarks
arnima at hafro.is
Wed Dec 10 09:31:43 PST 2008
Holy Moly. Jan Jaap is right; the linux64 model in my benchmarks is only
using one core, just like the win32 models. I misinterpreted what 100%
CPU load means, as reported by the 'top' utility, but now I'm certain
it's just using one core.
This makes the linux64 performance twice as impressive. What could be
the main reason for such a speed difference between my linux64 and win32
Maybe someone with a win32/linux32 dual boot machine could run similar
benchmarks, to see whether the main reason is win vs. linux. Otherwise,
the main reason might be 32 vs. 64 bits. A win64/linux64 dual boot
benchmark would also be useful here.
The similar performance of win32-gcc345 and win32-gcc432 implies that
gcc3* vs. gcc4* is not relevant.
P.S. I've also run simple benchmarks in R and my own test.cpp to compare
the performance in Windows XP vs. Kubuntu 8.04-64, where linux64 shows a
similar performance advantage as in the ADMB benchmarks. ADMB users
would benefit from knowing whether the reason is win-linux or 32-64.
More information about the Users