Posted by
Michael Schmid on
Jan 23, 2008; 6:27pm
URL: http://imagej.273.s1.nabble.com/ImageJ-benchmarks-tp3697397p3697405.html
Hi Albert,
your OSX result is very strange!
On a similar machine (also 2 year sold) I got 3.5 seconds:
OSX 10.4.10, 2.1GHz PowerPC G5, 1GB DDR2 SDRAM, ATI Radeon X600 XT
(128 MB)
256 MB reserved for ImageJ and a lot of other applications open.
I have Java 1.5.0_07.
Maybe your Mac has a very slow graphics card? OSX performance
is known to depend significantly on the graphics performance,
and Benchmark.java contains a lot of updateScreen(imp);
commands.
On my OSX G5 machine, as an external plugin with
boolean showUpdates=false;
i.e. no image updates, it needs 2.6 s.
Concerning the remark of Jon about the similarity of Core Duo
and Quad, none of the number crunching routines called by the
benchmark is multithreaded and can take advantage of multi-
core CPUs.
Currently, ImageJ plugins use multithreading only for
processing stack slices (if the PARALLELIZE_STACKS flag is set -
it is for the majority of ImageJ built-in filter operations).
Michael
________________________________________________________________
On 23 Jan 2008, at 19:03, Jon Harman wrote:
> Windows XP Intel Core 2 Quad 2.4 Ghz .735 sec
>
> I'm surprised at the similarity between the core 2 quads and duos.
> Perhaps the mandrill image is too small. Image display may be a
> big part of the benchmark
On 23 Jan 2008, at 18:27, Albert Cardona wrote:
> ImageJ's Plugins - Utilities Benchmark on the Mandril sample image
> (RGB 512x512):
>
> (Best of 3 runs, ImageJ freshly restarted)
>
>
> MacOSX 10.4 java 1.6.0-beta 32-bit on a powerpc G5 2.0 Ghz 1.5 Gb
> SDRAM: 22.8 seconds
>
>
> Ubuntu 7.10 java 1.6.0 64-bit on an intel core 2 quad 2.4 Ghz 4 Gb
> DDR3: 0.741 seconds
>
>
> Ubuntu 7.04 java 1.6.0 32-bit on an intel core 2 duo 2.2 Ghz 2 GB
> DDR2: 0.802 seconds
>
>
>
> The difference is staggering. That G5 is about 2 years old only.
>
> I would appreciate if people posts benchmarks on their machines,
> to get an idea on what is ImageJ (and TrakEM2) going to be run on.
>
> Thanks!
>
> Albert
>
> --
> Albert Cardona
>
http://www.mcdb.ucla.edu/Research/Hartenstein/acardona