Offler

Physical problem fixed.

CAS latencies adjusted (8-7-7-18 instead of 8-6-6-15).

System stable. Passed 1500 LinX tests with 771mb problem size (hottest torture I know) while running TV tuner which writes timeshift buffer into memory.

Memory voltage set in bios 1.98v
Memory measured by bios and other software 1.88-1,95 and unstable...

vNB Stock

Ambient temperature 35-40 degrees of celsius.
CPU: 72-73
NB 55
System 55

Offler

For reference some successful settings

http://www.ocztechnologyforum.com/fo...d.-Post-1-Of-2

Don_Dan

Did you notice that he's running his NB at 1.63V? And he's only running a dual core with his memory at 720MHz... Your setup is way tougher on the NB.

Offler

He is also running on CAS7 which is totally insane.

Also CPU VTT on 1.200v is fewer i have.

Just testing now more volts on northbridge BUT

at 1,43 i have almost 60 degrees on NB.

Just to mention my current torture tests contains live TV DVB signal from two TV tuners, MPEG2 being decoded by graphic card, and in addition one DVD also in Mpeg2, linx running while one of TV tuner software is writing timeshift data into memory.

3 quite large data streams over PCI-E bus...

Two days ago I managed to have system stable with some config (CPU 1.36, VTT 1,29, NB 1,29), but the memory timings were loosened ( 1600 MHz 8-8-8-22, on 1,95v).

Today I set it to:
CPU 1,36v
VTT 1,265
MEM 1,95
NB 1,43

And timings are back to 8-6-6-22

Only TV tuner app was failing, so i will set VTT to 1,295-1,305v as was before.

Even if stable the NB temperatures under my tortures are just too high.

At all I find his configuration quite interesting. Using 333/1333 instead 400/1600 seems promising.

Currently any OC is limited more by CPU as by anything else. however the 100% stable configuration is still not completely reached.

Offler

So, after more tests still I have to run memory on 1.93v at least. Adding more volts to NB helped to prevent instant reboots and BSODs, so the system is stable now, but so much voltage on both NB and memory indicates that there is other device which is underpowered...

At all even 1.712v on memory and 1,43v on NB was not stable...

Offler

Currently system stable at:

400Mhz FSB, RAM 1600 cL 8-6-6-22 CR 1

1.3v CPU
1.93v RAM (measured values lower)
1.43v NB

Running those elpidas on 1.93v for some months on normal usage. Still alive... Runing linx from time to time...

Offler

I managed to lower memory voltage to 1,83v with some changes in Latencies.

Current latency is 8-6-6-23, while other latencies were set from auto to manual values. Also skews model and strenght is now 1-1-2-2.

At this setting system seems stable.

Therefore I tried lower memory voltage. First system was stable for some hours, then it was barely able to boot. It seems that capacitators on the board have some sort of "memory effect". The board must discharge completely before you can test its stability, because some energy still remains on the capacitators and that made system to look more stable as it should be.

It has also opposite effect. If are you raising the voltage system might be unstable for some time.

Usually all voltage changes require power off, removing power cord, pressing "on" button to discharge the system of all energy. Wait few minutes and then put it back on power.

Can be also caused by some kind of bios error when values are changed later as expected.

Offler

Later stress testing shows that if memory is stressed from both CPU and PCI-E it will fail on current settings.

Voltages have been raised again, so it seems that this configuration really needs more voltage on both NB and Memory.

Therefore, are there any memory chips with 2gb on dimm, with 1600mhz CAS 7-7-7-x at least able to handle 1,9v? Other choice is to run 1,9 on these Elpidas and wait...


Edit:

VNB 1,43
Mem: 1,85
1600 8-6-6-21

Stable. Possible to reduce latency to 8-6-6-18