Ah ha! Check my thinking here.

[trents]

Just got my mobo back from gigabyte rma. They fixed the fried P4 power connector but they didn't check out the dual channel instability problem I had been having that I also mentioned in the RMA request. The problem is still there. I've emailed them numbers of times over the past several months about this but they keep saying "it must be the CPU IMC."

Well, yesterday after reinstalling the board I used just one memory stick and tested each of the four slots in turn for stability What I discovered confirms to me that there is a problem with the mobo. When I put the ram stick in either slot 1 or 3 it will not run prime95 at all. When I put it in slot 2 or 4 it is completely stable under stress. This makes sense because slots 2 and 4 will necessarily be used if you are running in dual channel mode. Is my thinking correct here. The problem is with the mobo, right?

 

[redduc900]

Those sticks are rated at 5-5-5-15 and 2.0 - 2.1 Vdimm? What's the current Vdimm, and with the sticks installed individually in slots 1 & 3 does Memtest86+ pass?

 

[trents]

Here is a screen shot of the SPD tab in CPU-Z. Note that at DDR2 800 speed the mfg. indicates the ram only needs 1.8v. At the time I tested it it was set to the DDR2 800 default speed and voltage. Interestingly enough though, both PC Health in bios and Speedfan in Windows showed the DIMM voltage to be 1.96 even though it was set to auto, which should have been 1.8v. This board has never had good voltage regulation.

At any rate, I'm not sure what your are driving at with your question about the Vdimm? I haven't tested it with Memtest86+ yet. If the memory is stable in slots 2 and 4 in Prime95 shouldn't it be stable in slots 1 and 3 at the same voltage?

 

[QuietIce]

MemTest is the only real way to check a memory problem. I understand your logic and it seems sound but it would be nice to have it confirmed by a MemTest run ...

 

[trents]

MemTest is the only real way to check a memory problem. I understand your logic and it seems sound but it would be nice to have it confirmed by a MemTest run ...

By "memory problem" do you refer to the ram itself or are you including the DIMM sockets in that category? I know Memtest86+ is the gold standard for ram but would it necessarily reveal a socket problem?

 

[QuietIce]

MemTest will pick up anything that interferes with RAM performance - and only RAM performance. That includes bad hardware or even an over-stressed IMC.


BTW - Your [Auto] result with the vDIMM is exactly the reason I always recommend manual entry of critical settings. The only exception I've found to that rule is ACC on a Phenom I. Even with hours of testing I could never get the ACC to balance better than [Auto] ...

 

[trents]

Ice, thanks for all your input and help.

Concerning an over-stressed IMC I think that would show up in all of the dimm slots if that were the problem, not just two of them. By the way, the two unstable slots constitute channel A on the board. Also, I have plugged this CPU into my wife's computer's mobo and it runs dual channel just fine, though her board only has 2 dimm slots. I don't know if having only two slots makes any difference or not. Concerning running at auto settings, I only did that because it wasn't stable in dual channel mode with manual settings. One weird thing is that when I set the dimm voltage to auto it should default to 1.8 volts when I run the memory at 200mhz speed. Yet, PC Health shows it actually running at 1.93. The fist two voltage bumps in bios are not even reflected in PC Health, it stays at 1.93. This board has always had crummy voltage regulation. I also think it has bios problems but gigabyte hasn't released a new one in a while for the board. I think they are writing it off. NewEgg only marketed it for a short time before dropping it for a version 2.

Anyway, I'm tired of messing with this mobo and I'm tired of dealing with gigabyte tech support. I just ordered a new board:
http://www.newegg.com/Product/Product.aspx?Item=N82E16813157149

By the way, its the same chipset (780G) so should I be able to install and run it without having to reinstall OS?

It has support for 140W CPU's so I'm thinking the electronics are a little stouter than the one I have been using which was only rated for 125W. It's my wife's birthday present to me.

 

[QuietIce]

I never reinstall the OS so I'm not one to ask about that. Oddly, I've never had blow-back from that either - at least nothing I couldn't fix. But on that note I recently changed from a BioStar to ASUS board, both 790GX chipsets. I didn't have any issues at all and didn't even change drivers ...

 

[trents]

Ice, just for the fun of it, ha!, ha! I did test ram with Memtest86+ as you suggested. I put both sticks in (dual channel mode, slot 1&2) and ran the test all night - 8 passes, no errors. The next day I switched the ram to slots 3&4 and ran Memtest all night again. No errors. Why can I run Memtest till the cows come home in dual channel mode but can't run Prime95 (same bios settings) in dual channel mode for even 2 minutes? Yet the computer is perfectly stable in single channel mode as long as I don't put the ram in slots 1 and 3? What are the possibilities here? Obviously, the ram itself is good and the ICM is good or it would have failed Memtest. Is this a bios issue or some quirky ram incompatibility issue that only shows in Windows when in dual channel mode or in single channel mode when slots 1 and 3 are populated?

 

[QuietIce]

How does OCCT test out at those settings? I and several others have mentioned a few times that Prime 95 is sometimes quirky with these new systems. I couldn't use it at all in blend on my 9950BE/M3A78-T rig - just wouldn't run right. It was VERY sluggish and never showed stable. OCCT ran for 16 hours and never a hint of problems. Since then I've been running SETI 24/7, now going on three months with no bad work units. As a testing program P95 has gone downhill ...

 

[trents]

Ice, thanks for the idea. I'm not familiar with OCCT. I'l have to check it out.

 

[QuietIce]

OCCT

Home: http://www.ocbase.com/perestroika_en/index.php
D/L: http://www.ocbase.com/perestroika_en/index.php?Download

 

[trents]

Hey, I like OCCT. You can set the test for a custom amount of time. It really heats up the CPU in short spurts when Linpack option is chosen. The OCCT test option seems to not work the memory very much.

Say, at certain points during the testing my desktop screen goes blank with maybe just the tool/task bar still viewable. Same thing when I run the stress test that comes with AMD overdrive. Is this normal?

 

[QuietIce]

Hmmm - mine doesn't go blank if that's what you're asking. Not sure what would cause that ...