In our experience with SUN, Linux and PC there are three factors that
determine speed in SPM with fMRI data. In order of importance they are RAM,
Processor, HD. We run SPM on a dual processor Sun Ultra 60 with 2GB of
RAM and we have experienced a drastic improvement over our previous 500MB.
The idea is to avoid swapping at all costs. The processor has an obvious
impact on performance but many ignore a common bottleneck in storage. We
use Ultra wide LVD SCSI 10,000 RPM hard drives. We found that due to the
large number of files and the size of our volumes the I/O is also a
crucial part of performance.
The bottom line is: be sure that your system is well balanced. Investing
large amounts in one of these three areas while ignoring the others will
have minimal impact. Although, if you can only go for one area, an increase
in RAM will almost always cut your processing time. (this is an
oversimplification, there are situations where an increase in RAM will have
no effect on your speed. To get the full picture you need to take into
account your specific studies and paradigms, file sizes, existing ram,
whether you will do group analyses, etc...)
Dear Joseph
Could you provide us with a typical processing time for statistical
inference (specifying with the size of you data, and the number of
covariates) using your faster computer ?
In our limited experience, RAM > 256 Mo essentially increase the
statistical computation, not or at least poorly the pre processing steps.
Did you experience the same ?
Thanks in advance
Jack
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|