Bizzare CPU/Mem issue, help required!! ;-)

[K22]

Evening all!

I am currently experiencing a bizarre issue which I am unable to resolve... time to consult the knowledge that is Hexus!

In a nutshell...

I have a DFI Lanparty NF4 SLI-DR with 406 bios, 4 x 512Mb Mushkin PC4400 Black, and naked Opteron 144 CABNE. PSU is an OCZ Modstream 520, GPU 7900GTX, 1 Spinpoint HDD, 1 Pioneer DVD-RW, OS XP32 SP2.

The rig is under high end water, with load temps @ 2.9Ghz/1.65v of 30 Centigrade. The chip will sit at load all day long crunching with prime, no problem. This is with the memory @ a max of about 235Mhz.

Now, here comes the issue... with the CPU @ default, the memory will crunch prime all day long @ 285Mhz. I have also run memtest for 12 hours @ 285Mhz, CPU @ default with no problems... However, try to run them together, mem @ 285, cpu @ 2.9 and instability ensues. CPU @ 2.8, mem @ 280 is still unstable and the memory only becomes stable at around 235 with the CPU overclocked. What is baffling is that individually they're happy to run at these speeds.

I do run the memory on a divider as the FSB is normally run anywhere between 300 - 400 in the search for stability. Low FSB's are not really optional as the max multi of the CPU is 9, though I have succesfully run memtest @ a straight 285 FSB with a 6 - 7 multi and stock volts on the CPU to verify the issue is not FSB/Divider related.

Memory timings are a lenient 8-4-4-2.5. The board will happily run the CPU @ anything up to 475 FSB, so I have tried all FSB/Multi combinations below that to try to stabilise the memory but to no avail.

My initial thoughts are the memory controller on the chip causing the problems, and therefore an unresolvable issue, but your thoughts on this would be greatly appreciated. Many times in the past I have overlooked the obvious and I am hoping this is another classic example...

Regards, K22

[Jay]

what PSU do you have? I wonder if the OC is stressing it?

ps

forget that, your "My system" tells me and that should be fine.

[peterb]

Memtest is doing a LOT of memory accesses - running something like prime on top may be just pushing it that bit too far and you are getting some memory read/write errors - although I would expect Memtest to show that - unless it is a critical one that just crashes the PC. Only a theory though!

[Lucio]

Ok, firstly, have you checked the obvious, is your Southbridge being cooled properly? You mention water cooling which can play havoc with airflow inside the case, if you've got extra voltage going through that, it's gonna get hot quickly. Also, are the RAM chips themselves getting proper cooling.

Secondly, try your timings on "Auto", rather than choosing what you think the memory should run at, sometimes you can get a more stable response from a good motherboard.

[Hunain]

Id second Lucio

[K22]

Thanks for the comments guys I can clarify on all of the points raised and hopefully get closer to a solution.

Originally Posted by peterb

Memtest is doing a LOT of memory accesses - running something like prime on top may be just pushing it that bit too far and you are getting some memory read/write errors - although I would expect Memtest to show that - unless it is a critical one that just crashes the PC. Only a theory though!

I use Memtest86+ which is not Windows based... it's a standalone app loaded via boot disc.

Originally Posted by Lucio

Ok, firstly, have you checked the obvious, is your Southbridge being cooled properly? You mention water cooling which can play havoc with airflow inside the case, if you've got extra voltage going through that, it's gonna get hot quickly. Also, are the RAM chips themselves getting proper cooling.

The Southbridge has a Thermalright HR-05 SLI with 80mm Yate strapped to its back. The case has also been modified and now has Scythe Slipstream 68CFM fans front and back. Due to the layout of the DFI, the memory running hoirizontally across the top of the board, the memory is in the best possible position as far as airflow is concerned in the case.

Originally Posted by Lucio

Secondly, try your timings on "Auto", rather than choosing what you think the memory should run at, sometimes you can get a more stable response from a good motherboard.

This is a interesting suggestion and something which I have not yet tried, will give it a go and report back!

Thanks to you all, K22

[K22]

Well... I just tried auto settings, but the board runs them tighter than my manual settings and it barely gets into Windows @ 285. I am now trying a technique posted by another user, whereby I'll loosen every timing to a ridiculous level, then prime, tighten one a notch, re-prime etc.

This is similar to the method used to find the stable 285 in the first place, I was hoping to avoid a second run as it is SO time consuming. I reckon I'll be going for a week or so, but if it works the results will be worth it!

On a side note, it's snowing today, and having put the rad on the window sill, I'm priming 2840Mhz with mem @ 284Mhz, full load at 20c/68f... This is the coolest I have ever seen it

K22