This section should probably be titled "make your (own) mind up time" rather than conclusions, since it would be inappropriate for me to make a judgement on which machine is best for SPM-ing on the basis of numbers alone. However, these data do reveal that all machines are not created equal, and having a speedy processor and lots of RAM will get you to the top of the benchmarking league.

When examining these results one must be aware that the benchmark used is not particularly sophisticated and only gives a rough indication of a machine's capabilities. For instance "bench" does not give any indication of disk I/O performance (which is a major determinant of "real-world" performance). Thus these results do not reflect the increased performance of SCSI over IDE disks.

Multiple CPUs and a multiprocessor aware operating system (Windows2000/XP, Linux, Solaris etc.) will enable you to run mutliple SPM analyses without a significant impact on performance, making these systems ideal for central processing servers which share resources between several users (particularly when combined with fast SCSI disk storage). However, for individual researchers who wish to obtain results as quickly as possible, a single fast CPU system will probably be ideal for your needs.

The situation for single users may change when The Mathworks make Matlab multithreaded and thus able to take advantage of multiple CPUs.

People have commented on reliability of a particular platform being more of a concern than absolute speed when choosing which system to buy, and this seems to be the basis of the divide between the PC (Windows) and Sun camps. Undoubtably Suns are more reliable than PCs, but this is achieved at a significant price premium. Given improvements in PC technology, the balance between cost and reliability seems to be leaning towards the PC.