Memory mystery, at least a logical puzzle

Gustavus

Golden Member
Oct 9, 1999
1,840
0
0
I have two machines built around ABIT IC7 G motherboards -- one in use and one on the bench. One has a P4 Northwood 3.0 GHz CPU, the other has a Prescott P4 3.0 GHz CPU. Both machines are running Windows XP with SP2 installed, although that fact is irrelevant to the problem I have encountered.

The Northwood machine was failing a Prime95 test so I ran MemTest on it and the memory showed lots of errors. The memory was OCZ PC3200 EL Platinum Rev 2, with latency specs of 2,2,2,5. Thinking the memory had gone bad, I opened a new matched pair of Corsair XMS PC3200, also with specs of 2,2,2,5, and replaced the memory. Still lots of errors. The memory addresses at which errors occur are random.

The bench machine with the Prescott is overclocked to 3.8 GHz. The FSB is 253 MHz, The memory divider is set to 5:4 so the memory is running at close to spec (202 MHz instead of 200 Mhz). The timings are set in the BIOS to 2,3,2,5 -- CAS,Act to Precharge, RAS to CAS and RAS precharge. I swapped in both the OCZ and the Corsair memory that was showing errors on the Northwood machine and no errors were shown in MemTest for either set after many hours of testing on the bench machine. The conclusion is that both memory sets are good.

I have gone back to the Northwood machine and set the memory divider so that the clock to the memory is only 180MHz. Both machines are running the memory at 2.8v, the highest the IC7 G has available. Both the OCZ and the Corsair still show errors in Memtest -- even though both are now running at under spec. I even relaxed the timing in the Northwood machine to 2,3,2,7 without getting rid of the MemTest errors.

The bottom line is that memory which works flawlessly in one IC7 G motherboard at nominal spec. (202 MHz, 2,3,2,5) fails miserably in another IC7 G motherboard at below spec (180 MHz, 2,3,2,7)

Can anyone even suggest a possible explanation for this anomolous behaviour?

The reason for the 3 setting instead of the spec 2 is that the SPD setting results in a latency setting of 2,2.5,2,5 and Eric Ryder at OCZ informed me that the OCZ memory does not like the 2.5 setting. He suggested it be set in the BIOS to 3.

Incidentally the Prescott machine -- overclocked to 3.8 GHz -- can and has run Prime 95 for 24 hours at a time. Rock stable.
 

Lemon law

Lifer
Nov 6, 2005
20,984
3
0
The first thing you might research is memory voltage, you could be under volting the ram, so check ram manufacturers specifications against what your bios shows. Or use freeware programs like speedfan or everest2.2, or PC wizard 2008 to monitor your ram voltages. And you may also have to try, on a trial and error basis, all kinds of memory timings. I doubt you will lose must speed if you go up to cas 3.

Then you can use the various benchmarks in something like PC wizard to see how much you have gone up hill or downhill in terms of performance.

But if your memory is making errors, you can screw up the integrity of your windows install on your HDD.

Other warning, if your computer crashes, your bios may reset, so check your bios for changes before you reboot.
 

Gustavus

Golden Member
Oct 9, 1999
1,840
0
0
Lemon law

Thanks for the reply. Both memory sets would benefit from a slight bit more voltage, but 2.8v is the highest the ABIT IC7 G can provide. The puzzle is that at the same voltage both sets of memory run indefinitely without erros (in MemTest) on the bench machine and produce errors on the other machine. I have built, and often overclocked, many machines, but can't even guess as to why the memory is behaving so differently in two very similiar machines.

The OCZ memory is especially good. I have run it in other machines at 250 MHz FSB. That is why I thought the memory must have failed when I began getting memory probles. Luckily I had a couple of new packages of the PC3200 2,2,2,5 Corsair matched modules and could replace the OCZ modules. When that failed in the same way I was left with the puzzle that prompted this post.

Thanks again for the response to my question.
 

Fardringle

Diamond Member
Oct 23, 2000
9,200
765
126
If known good ram fails in the second machine, that would seem to indicate a problem with the memory controller on the motherboard. It could be an issue with the CPU as well but that's less likely.
 

Lemon law

Lifer
Nov 6, 2005
20,984
3
0
With all due respects to Fardringle who may be right, then the remaining option is to play around with memory timings or go to 2T if you are not there yet. Its a lot of trial and error but if you find a setting where they stay stable, stick with it. I also assume you have something like cpu-Z to cross check what the memory is actually running at. What you set is not always what you get. I also note there are other bios setting for running ram in synchronous or asynchronous mode that may be part of the trial and error process.

I also note you could try to swap cpu's to see if the problem lies in the cpu.
 

Gustavus

Golden Member
Oct 9, 1999
1,840
0
0
Lemon law

I have run further tests of the memory and both the OCZ and the Corsair can run MemTest for hours at 2,2,2,5 at well above the rated 200MHz FSB. So the memory is not only OK, it is exceptional.

I may have wasted money, but I just purchased an SL7WK Northwood on e-Bay a few minutes ago and will swap out the CPU presently in the machine. If replacing the CPU does not correct the problem, I will give up since it makes no sense to replace the aging socket 478 IC7 G motherboard -- which surprisingly still sells for premium on e-Bay
 

KGB

Diamond Member
May 11, 2000
3,042
0
0
As a "last ditch" effort, have you tried clearing the CMOS on the affected MB?

I know a few people will scoff at this suggestion but I have seen stranger things.
 

Gustavus

Golden Member
Oct 9, 1999
1,840
0
0
KGBMAN
Thanks for the reply and suggestion. As a matter of fact I was compelled to clear CMOS as a result of corruption during one of the many tests I have run on the machine. Still have the same problem.

I have one of the 30 capacitor SL7WK Northwoods on the way (shipped only hours after I paid for it!) and will hold off doing anything until it gets here.
 

KGB

Diamond Member
May 11, 2000
3,042
0
0
Originally posted by: Gustavus
KGBMAN
Thanks for the reply and suggestion. As a matter of fact I was compelled to clear CMOS as a result of corruption during one of the many tests I have run on the machine. Still have the same problem.

I have one of the 30 capacitor SL7WK Northwoods on the way (shipped only hours after I paid for it!) and will hold off doing anything until it gets here.

Good luck and please report back with results. :thumbsup:

 

Gustavus

Golden Member
Oct 9, 1999
1,840
0
0
wjal,
Thanks for the response. Yes both machines have the latest ABIT BIOS -- version 28.
The line at the bottom of the boot startup screen is

12/28/2004-i875P-W83627H-6A79BA19C-28

where the last two digits identify the BIOS version.

I have never had occasion to check chipset drivers and will have to look up on net how to do that.