Ok, let's start from the beginning...
MSI P6N SLI-Platinum (650i SLi)
4G (4x 1G) G.Skill D9GMH Micron DDR2-800
Intel Core 2 Duo Conroe E6600
After unplugging an IDE drive from my mainboard, the machine refused to POST. I removed half of the memory, and it would run. It didn't matter which half I removed -- of the 4 modules I had, it would work and pass memtest with any 2 modules, in any 2 slots, but any 3 modules in any 3 slots, it wouldn't POST. I assumed it was a bad memory controller, and I replaced the board.
Replacement: GByte EP45C-DS3R
The new mainboard was an entirely different chipset. It, too, would have problems with my memory. Only this time, any 3 modules in any 3 slots was fine, it only refused to POST when I put in the fourth. I run it like this for a while on 3 memory modules, but then I started having a lot of stability issues, read-write errors, and disk-related bluescreens. I try to run hard drive diagnostics. 2 of my 4 drives fail MHDD, the third won't even be seen by the utility, and the fourth drops out of the array. I assume the drives are bad, I replace them with 2 more.
The machine stops POSTing again, now working only with 2 of the 4 DIMMS in, like the first had been. One of the new drives fails diagnostics. I get laid off and can't afford to throw more money at the problem.
Fast forward 3 months, during which time my computer has pretty much just sat there. I finally bite the bullet and replace all my memory and the failing hard drive.
Replacement: 4GB (2x2GB) OCZ DDR2 1066
I fire up the PC with the new memory and test the hard drive. It passes, but only marginally. Then I try to run memtest86+ 2.11 on HBCD, and the machine hangs. I reset the BIOS to optimized defaults, and run again. It hangs. I put in QuickTech. It goes until the ATS memory test, then spits out 2 errors. I try a memtest86 CD, it runs, and passes. I try it without the memory in dual-channel--no difference. I try mixing up the types of memory--no difference. I find the BIOS is undervolting the memory, so I set the voltages manually to 2.1v, and the timings to 5-6-6-18 (the old memory takes 4-4-4-12, the new 5-6-6-18)
I pack it up and take it in to work to run more quicktech and memtest diagnostics. Here are my (strange) findings:
Both of the new modules pass memtest and quicktech, in any memory slot.
If you combine the two modules, either in single or dual-channel, they pass memtest, but fail quicktech ATS.
one of the old modules fails memtest86 on its own, but passes if combined with any other module. I'll assume this module is bad.
another of the old memory modules passes memtest in slot 0, 1, or 4, but if you put it in slot 3, the board won't POST.
the other two old modules pass memtest and quicktech, in any slot, and also together, in single or dual channel.
Adding either one of the two new modules to the mix while the two good old modules are in it, causes it to fail quicktech ATS.
The ATS failures are 100% consistent, and occur at between 70 and 90% of the way through the test. Mixing one of the new memory modules with the old ones will result in a slightly different pattern offset than the other. Adding the two new modules together produces the same pattern offset as the one module does in conjuncton with the older ones, and even the same as with all 4.
What's goin' on eh?
MSI P6N SLI-Platinum (650i SLi)
4G (4x 1G) G.Skill D9GMH Micron DDR2-800
Intel Core 2 Duo Conroe E6600
After unplugging an IDE drive from my mainboard, the machine refused to POST. I removed half of the memory, and it would run. It didn't matter which half I removed -- of the 4 modules I had, it would work and pass memtest with any 2 modules, in any 2 slots, but any 3 modules in any 3 slots, it wouldn't POST. I assumed it was a bad memory controller, and I replaced the board.
Replacement: GByte EP45C-DS3R
The new mainboard was an entirely different chipset. It, too, would have problems with my memory. Only this time, any 3 modules in any 3 slots was fine, it only refused to POST when I put in the fourth. I run it like this for a while on 3 memory modules, but then I started having a lot of stability issues, read-write errors, and disk-related bluescreens. I try to run hard drive diagnostics. 2 of my 4 drives fail MHDD, the third won't even be seen by the utility, and the fourth drops out of the array. I assume the drives are bad, I replace them with 2 more.
The machine stops POSTing again, now working only with 2 of the 4 DIMMS in, like the first had been. One of the new drives fails diagnostics. I get laid off and can't afford to throw more money at the problem.
Fast forward 3 months, during which time my computer has pretty much just sat there. I finally bite the bullet and replace all my memory and the failing hard drive.
Replacement: 4GB (2x2GB) OCZ DDR2 1066
I fire up the PC with the new memory and test the hard drive. It passes, but only marginally. Then I try to run memtest86+ 2.11 on HBCD, and the machine hangs. I reset the BIOS to optimized defaults, and run again. It hangs. I put in QuickTech. It goes until the ATS memory test, then spits out 2 errors. I try a memtest86 CD, it runs, and passes. I try it without the memory in dual-channel--no difference. I try mixing up the types of memory--no difference. I find the BIOS is undervolting the memory, so I set the voltages manually to 2.1v, and the timings to 5-6-6-18 (the old memory takes 4-4-4-12, the new 5-6-6-18)
I pack it up and take it in to work to run more quicktech and memtest diagnostics. Here are my (strange) findings:
Both of the new modules pass memtest and quicktech, in any memory slot.
If you combine the two modules, either in single or dual-channel, they pass memtest, but fail quicktech ATS.
one of the old modules fails memtest86 on its own, but passes if combined with any other module. I'll assume this module is bad.
another of the old memory modules passes memtest in slot 0, 1, or 4, but if you put it in slot 3, the board won't POST.
the other two old modules pass memtest and quicktech, in any slot, and also together, in single or dual channel.
Adding either one of the two new modules to the mix while the two good old modules are in it, causes it to fail quicktech ATS.
The ATS failures are 100% consistent, and occur at between 70 and 90% of the way through the test. Mixing one of the new memory modules with the old ones will result in a slightly different pattern offset than the other. Adding the two new modules together produces the same pattern offset as the one module does in conjuncton with the older ones, and even the same as with all 4.
What's goin' on eh?
