Offler

After playing around with ATX Power connector I am sure that part of problem was here - mechanical connection was good, but the electromechanical characteristics were poor. The trick with the spray helped with more boards, althougt some people dont believe me in this.

Also another part of problem - occt crashing - was purely SW problem. OCCT graphics test can reboot the machine. MSI kombustor works well however i fear that it can ignite fire in my computer . 85 degrees on each GPU core, 90 degrees on GPU0 power rail, 150 degrees on GPU1 power rail. I almost burned mi finger when i touched the HD3870X2

I did only minor OC to the CPU, but the computer is extreme to the core

I did not measure the ATX connector with multimeter. Howerer the DFIs tool measure same values as OCCT and Everest do. Therefore i suspect that the measurement is bit screwed on hardware layer.

eva2000

nah software via DFI SG is inaccurate for psu rails trust me, you need to measure with multimeter

Offler

Well after few days of normal usage i had only one problem - oblivion is crashing, but when the oblivion was running fine?

Now to OC. Its much harder than before. Dont know real reason of this.

Offler

3d mark2003 is particulary very good tool in case you are trying to find any errors with your system.

During loading of "Battle of proxycon" the system paused for a minute and then it ran fine. Since during this the disk was churming (SCSI disk) the problem was surely located somewhere on Southbridge. Althought the problem with ATA Jmicron adapter has been sucessfully identified, and it solved the problem i was working on it for a while.

Results are very good. System now boots with PCI-E spread spectrum on and JMicron enabled, and the problem with the pausing of 3dmark was solved too. Have just set "Host slew rates" to "Strong".

Since many trouble with OC has been later referenced to SB/PCI instability this might solve the OCing as well.

The Cooling: temperature at 76 degrees on CPU has been indentified as problem temperature where LinX can measure first error due overheating. Need to check cooling system, however this temperature was reached only when ambient temperature was 35 degrees of celsius.

Also i have one X25-V already ordered. Should be enough for the OS or gaming.

Offler

I have set the LinX to very special setting - to 768mb of ram. At this setting the CPU is utilized as much as possible, highest GFlops are reached and highest temperature too.

Now i have 300x pass of this test which is the hardest ever i did on my machine.

The problem with errors was both on low CPU VTT voltage (adjusted again to 1.285v) and CPU Vcore. I turned the Auto VID control on again and it works.

Now have to just check whether stability was improved even in real environment.

Offler

Yesterday was very hot day... 40 degrees ambient temperature. The machine gone crazy again

Today i am gonna test the memory again. At 1600 the system might fail at higher temperature, on lower memory frequency most likely will work normally.

I ran Prime95 Torture test while using only CPU cache. The test was OK, so i think that most trouble are outside the CPU, and while CPU can run fine even clockgen cannot cause trouble.

Memory controller hub...

Since i tested more DIMMs on the machine i am sure that trouble are not caused by Nanya neither OCZ modules. Seems more that the Nb does not handle properly the full gas mode, while it cannot run on its standard 1,9v.

Offler

After few tests and various MCH settings i removed the OCZ memory and placed back the Nanya Elixir dimms.

First attempt to gain good stability was sucessfull.

"Performance level" is set to Auto, which means that value is 10.

System works flawlessly even after long gameplay i was unable to cause error or BSOD related to the memory, also LinX always pass.

Will test same settings with OCZs but i fear that the problem is in overheating of those DIMMS.

Offler

For now at basic settings the memory issue has been detected only once - when the 3.3v rail dropped below 3.13v. ATX cable needed some fixing again. Since then single crash of testing game - The Witcher after 5 hours of playing, and seems to be of software cause (the game crashed and took the system with it, or maybe the modded X-Fi did this ).

Also the system passed some hours of playing the witcher on core 1 and 4, while on cores 2 and 3 was running LinX. (Quite impressive, because the game runs smoothly same as with two cores unused.)

And small correction - Performance level is set to 8 on channel A and to 9 on channel B. Bit strange, but possible. Will set it to 9.

For now it seems that the issues with memory are caused neither by clockgen, but partially by low CPU cooling and cooling at all, and partially by high load on NB which is caused memory.

Cross reference testing of OCZ and Nanya shows similar problems related to temperature, voltage and BIOS settings. Althought the DIMMs are differrent the root cause of the problem is not the memory itself. Clockgen voltage settings affect more Southbridge and PCI devices than memory or CPU (at least on 400Mhz FSB), but fact that on lower memory frequency both dimms work fine shows that NBridge is bit overloaded by 4 CPUs, 2 GPUs and highest memory multiplier.

This can be tested - by modifying the NB voltage or the Memory voltage. The MCH should run at 1,9v at 1600MHz... OCZs are unable to accept such high voltage, Nanyas will run.