This keeps coming up with IDE RAID - maybe we could get a mini-FAQ? Sandra is not a very reliable disk benchmark for some (at least) IDE RAID controllers. People with seemingly identical setups (including latency patches, 4-in-1's, etc.) can have differences of 20000+ points in Sandra's drive bench. By this point I've searched extensively all over the web on this topic, and no one has offered a satisfactory solution. It's not clear whether the problem is with the way Sandra calculates its score, some component of XP (seems to happen most there), a driver issue, a chipset problem, or even whether the performance really is any worse than it should be. And because people keep on using Sandra anyway, there's no collection of other benchmark scores to compare against. HDTach isn't much help, since the unregistered version doesn't test NTFS.
What to do? Well, ideally we'd use Iometer (available
here). This bench was recognized by Storage Review as the most reliable disk bench. But Iometer takes a lot of work to configure and standardize. I suggest ATTO Pro-Tools, which you can download after free registration
here. I'm going to make a post later tonight asking for ATTO benches - I'm getting tired of this uncertainty. Until then, try ATTO and compare it to my results (dual RAID 0 60GXP's on Promise onboard): max read/write peak speeds levelled off at 43k-50k after the ~64k file size. Some other ATTO benches, SCSI and IDE, are linked to
here.