I have a very puzzling situation.
On one machine (a Mac Pro dual Xeon dual-core (rev 1.1)), model
estimation runs in a jiffy. On two much newer, much beefier machines
(a Mac Pro dual Xeon quad-core (rev 3.1)), it runs like a dog... we're
talking 8 to 10 times slower, totally unusable.
I really IS the hardware. I've physcially moved the hard drives from
machine to machine, so all tests were conducted with exactly the same
operating system, matlab version, spm version, data files, etc. The
only thing that's changed is the hardware.
Using the profiler, I was able to track it down to slowdowns in
spm_sample_vol from within spm_get_data. On the old machine each call
to spm_get_data takes 0.1 msec, whereas on the newer machine it takes
over 4.0 msec. A 40x slowdown in data access could well explain the 8
slowdown in total processing time.
Does anyone have any idea what's going on? (Note that the slowness
exists on *two* new machines, so don't go accusing me of having bad
ram or a bum disk I/O.)
Any help at all would be greatly appreciated,
G Corrado PhD
|