What's the Culprit: CPU or MOBO?

jsrudd

Distinguished
Jan 16, 2009
927
0
19,060
I built my computer about two years ago using the following parts:

i7-920 w/ Scythe Mugen Cooler
Asus P6T
OCZ 1333mhz
Crossfire ASUS Radeon 4850
Corsair 650w PSU

Everything worked alright, except I ran the RAM at 1066mhz instead of 1333mhz. I ran it at 1066mhz because it was more stable. I thought that I just had some bad RAM at first, but even after RMA I couldn't get my RAM to run faster than 1066mhz reliably. I then read some posts about how terrible OCZ Gold memory was and chalked it up to that.

However, I recently I purchased Corsair XMS3 12GB 1600Mhz to replace the OCZ Gold that I had before. I installed the new sticks and set them up using XMP profiles. I booted up the system and all sticks were recognized and running alright. I then ran Prime 95 (torture test) to make sure everything was working.

Prime 95 crashed in less than 3 seconds. I changed some RAM settings including manually setting the timings and choosing the other XMP profile which had a higher QPI voltage. Even after all that I could never get Prime 95 (torture test) to run for more than a few seconds before my system crashed.

Thinking it was the memory I ran the Windows Memory Diagnostic and chose an extended scan from the advanced options (I didn't have any CD's lying around to run Memtest+). This completed two passes and didn't find any errors.

Then I booted up my machine again and ran Prime 95 on the setting that states it mostly tests CPU and not much RAM. I was able to run Prime 95 under this setup for about 10 minutes without any problem. I then switched Prime 95 to the "some RAM tested" setting and ran the test. My computer crashed after about 2 seconds.

The fact that my RAM was able to pass a memory diagnostic leads me to believe this is not the culprit, which leaves me with either the CPU or the motherboard. Any ideas how I can narrow it down or any other ideas as to what might be my problem?
 
Solution
Almost all RAM brands are made with little black chips that are reject chips from a very very small number of RAM manufacturers.

Those manufacturers test the chips heavily and the ones that fail get auctioned off to other companies who put their own name on it.

If the chips were high quality then the manufacturer would just sell it under its own brand name instead of auctioning it off to other companies.

The best of the generic brands (like Kingston, OCZ, Corsair, G.Skill, etc) do a lot of tests on those fail parts they bought and only use the best of them in their own sticks, but on some level they originally failed for a reason.

This is why I ask if you tried the new RAM underclocked. It is possible your new RAM is no better than...
RAM tests that run while booted into an OS can't test huge swaths of your RAM because of the way Windows works. You have to be pre-OS in order for a RAM test to actually test the entirety of the RAM.

That being said, RAM is a very touchy thing and even "name" brands fail a considerable percentage of the time. Kingston, OCZ, Corsair, G.Skill, none of those actually are OEM brands. They just buy chips auctioned off by real OEMs (the ones the OEMs make and don't want for some reason) and put them on circuit boards.

They do more tests than really no name brands, but they still fail pretty often.

Did you try running the torture tests with the old RAM underclocked?
 

jsrudd

Distinguished
Jan 16, 2009
927
0
19,060
I'm not overclocked at all. All my settings are set to auto in BIOS. I've also reflashed the BIOS just to make sure that's not the problem.

As to Windows Memory Diagnostic. It doesn't run while the OS is running. It reboots the system and runs the program. If Memtest+ is really more accurate I guess I'll go buy some CDs and try that.

I was able to run Prime95 torture test with the old RAM, but never at 1333mhz only at 1066mhz. However, that RAM was able to pass Memtest+ at 1333mhz.
 

jsrudd

Distinguished
Jan 16, 2009
927
0
19,060
I've manually set the timings to 9-9-9-24 2N, but that didn't really make a difference. I also tried increasing the DRAM voltage from 1.5 to 1.64 but that didn't seem to make a difference.

Is there any way I can test other components individually?
 

jsrudd

Distinguished
Jan 16, 2009
927
0
19,060


Everything is stock, which is why this is so strange.
 
Almost all RAM brands are made with little black chips that are reject chips from a very very small number of RAM manufacturers.

Those manufacturers test the chips heavily and the ones that fail get auctioned off to other companies who put their own name on it.

If the chips were high quality then the manufacturer would just sell it under its own brand name instead of auctioning it off to other companies.

The best of the generic brands (like Kingston, OCZ, Corsair, G.Skill, etc) do a lot of tests on those fail parts they bought and only use the best of them in their own sticks, but on some level they originally failed for a reason.

This is why I ask if you tried the new RAM underclocked. It is possible your new RAM is no better than the old and needs to be underclocked in order to be stable too.

Anyway, if the program you used to test them was run outside of an OS and it showed no problems then maybe you can search for other causes, but I would use Memtest if it were me.

Then again, I generally keep CDs and USB sticks laying around.

In any event, its hard for me to think your problem is anything other than RAM at this moment.
 
Solution