My home computer has been pretty unstable lately: occasional BSODs under Windows XP, lockups, and spontaneous reboots. A couple of weeks ago Windows started running CHKDSK on my applications volume (on a Hitachi Deskstar, 160 GB, HDS722516VLSA80) on every startup. This happened about half a dozen times, and then the filesystem became unreadable. The drive letter still appeared in Windows, but the volume appeared to be corrupt.
I ran some recovery software on it - all the files were still there, but the MFT was apparently corrupt - so I got all the files onto another drive, and then set about testing the drive. The Hitachi drive diagnostics showed no problems with the drive. I reformatted the partition and copied the recovered files back.
However, the problems continued. Windows would often reboot during startup, and then have to check the filesystem again. It would boot into safe mode OK, so I started wondering if the nVidia SATA drivers (the drive is on an nForce3 250 Gb motherboard) might be to blame, although I didn't get as far as uninstalling them. I also noticed (after normal Windows bootup - not safe mode) that the drivers had downgraded the controller to PIO mode, so I wondered if maybe the cable might be faulty.
So, I replaced the cable with another, and that seems to have fixed it! No lockups, reboots, bluescreens, or DMA downgrades so far... I should have checked the drive's SMART parameters earlier, but I notice now that the Ultra DMA CRC Error Count has a data value of 108. I assume the Hitachi diagnostics would have reported any SMART parameters outside the normal range.
FYI, the bad cabling was COMAX E173433-D, 30 AWG, 50 cm length, with a right-angled connector at one end. I'd used the right-angle connector at the motherboard end because it routed under the video card more easily. The new cable uses COPARTNER E119932 26 AWG cable (and, as it happens, quite a bit longer). Noticeably thicker cabling.
Thanks for reading - hope this might be useful for others.