Can't figure reasons for CRC errors after copying-help plz

[jolofss]

Hi all!
I'm experiencing some weird problems with my machine. I've got two IDE PATA drives on separate cables. It worked flawlesly for about 3 years, now there are CRC errors after copy operation between partitions belonging to one disc, or between both drives. I mean, copied files differs from original - I use md5 sums to check it.
I removed all optical drives, replaced IDE cables, checked bios settings, checked temp of drives and cpu, checked surface of both drives using mhdd, checked smart attributes (there are 3 ultra ata crc errors shown on one drive, but none for other, all rest is ok), thoroughly tested ram with memtest+ (12 hours) , tested cpu with prime95 (12 hours). no errors found.
Both drives work good on other machine, no errors at all.
PSU is good, Enermax 560W, working flawlessly for about 5 years. I don't know how to check it, but I will try to replace it for a while - but first I need to get another one
There's no software problem - I've experienced bad checksums under both WinXP and Linux.

What else should I check to solve this issue? Is it possible for integrated IDE controller to fail "just so"? I do not have other ideas

Any help will be appreciated, thanks in advance.

Hardware: nForce4 Gigabyte K8-NF9, IBM/Hitachi and Caviar drives, athlon64, not overclocked.
My exact config looks as following:
1st channel: Master: IBM/Hitachi HDT722516DLAT80, no slave
2nd channel: Master: Caviar WD1200JB, slave: Toshiba DVD-ROM

and one Pioneer SATA DVD writer.

greetings from norway!
Jakob

[rchadwick]

I'd either test drive on another MB, or another drive on that MB. Likely all that is left is the controller, or the drive. You could also try a plugin controller card.

[BlackST]

I vote for Psu. The smart log show an err related, so in order : use separate branch for each hdd, leaving the rest disconnected, copy in pure dos and check. Otherwise, the mb controller or one of the drives is failing. P.S. Some drives when configuring require to be set if a SLAVE PRESENT.

[cgallery]

Could be bad RAM. Test w/ MEMTEST86.

Edit: Woops, I see you did.