Stability / boot loop issue with tight loadline calibration settings

[tehownt]

I posted this on other forums but didn't get any answers, I hope someone here has some insight on the issue...

So about a month ago I get all of the pieces for my new build up and running, upgrade the UEFI to F6 and then get started on O/Cing it with the original PSU I had.

Fairly quickly I get the CPU really stable at 4.5Ghz with the following settings : 100mhz Core speed, 45x mult., 1.190v Vcore, "Extreme" Vcore Current Protection and "Extreme" Vcore Loadline Calibration (i.e. no vdroop), disabled the C3/6 state and forced the "Turbo" mode to 45.
This setup allows me to run prime95 stable for hours, CPU reaching temps of max 80-82°Celsius (one core), while other cores are between 70° and 78° (the 10° difference between cores really surprises me).

All seemed good and well but then I wanted to see how the temperature of the GPU would affect the CPU's in order to gauge the case's cooling efficiency. That's when things turned ugly.

As soon as I started Furmark I got a BSOD.

Then the UEFI wouldn't POST normally, it would try to POST twice, stopping on error code 15 (Northbridge init.) and then reboot get back to the config screen telling me either that it had a boot failure or that I could revert to "optimized defaults".

Immediately I tried to find out whether it was a video card problem so I ran Furmark alone again and to my big surprise it didn't crash for about an hour (temperatures are fairly reasonnable, solely reaching more than 70°).

I figured that maybe the video card was pumping too much juice on the rail, so I tried switching rails, switched to another video card (8600GT), yet the same thing happenned all over. It would crash and then boot loop.

Between all the tests, I noticed that whenever I shut down the computer completely and cut the power for a couple of minutes, I was able to POST with the OC settings w/o problems or boot loops (as long as it hadn't already tried
to POST and failed, otherwise it sortof continue failing until I saw the UEFI "boot problem" screen).

I thus upgraded the UEFI to F7, put back the 6850 and did the test again but to no avail, it would still fail. I tried raising the vcore to 1.2-1.25, nothing would make any sort of difference, it would crash and then boot loop.

I also tried changing the core speed to 100.1 or 101, although it would help some of the POSTs, I still could not run both burn-in at the same time.

The next option was to look at the PSU, so I got a new one, single rail (just to make sure that wasn't an issue), and almost immediately started the CPU+GPU bench at 1.190v vcore.

The surprising thing is that it worked for about 20min, but then the same issue happenned (actually I had no BSOD but prime95 just crashed) and when I rebooted this time I got an even worse boot loop, it would do it forever.

I upgraded the UEFI to F8D just for the sake of it, then unplugged everything, waited a bit then got it to POST and I decided I would be changing some other UEFI settings.
Since it looked like a voltage problem, I lowered the vcore protection and vcore calibration to "high" instead of "extreme" and went back to try and find the stable vcore voltage for my 4.5ghz OC.

After being able to run prime95 stable for hours using a 1.24 vcore (vdrooping to 1.20 under load), I decided to run Furmark in parallel and successfully got it to run for about 45mins before prime95 crashed.

So now, I guess my only options left are to lower the vcore voltage tightening and raise the actual vcore but to me it seems a bit stupid for the board to provide me with super-tight voltage reg that aren't able to sustand high loads...

TL;DR:
Boot loop/instability with "extreme/turbo/high" Vcore loadline calibration when both CPU(OC'd)+GPU burn-in eventhough CPU (OC'd) and GPU stable on their own. Needs a cold boot (PSU OFF) to come back to normal.

Does it seem like some voltage regulators might be defective (should I RMA the board) ?

[Hyper530]

What is the model name of your mobo and CPU RAM etc

[drotaru]

There is allready a new bios for thr UD5H , try with that one and see if there are any more issues

[profJim]

GIGABYTE Latest Beta BIOS often has beta bios versions that are not on the Gigabyte website.

Latest Overclocking Programs, System Info, Benchmarking, & Stability Tools is an excellent resource.

There are other stickies at the top of each web page in the Gigabyte forum.

Welcome to the forum.

[tehownt]

What is the model name of your mobo and CPU RAM etc

Sorry I didn't include them in the article since they are present in the "dropdown menu" just under my nick...

I will try the new BIOS hoping that it helps but so far none from F6 to the F8d did any real difference. What helped a bit was : 101mhz core speed, loadline calib to "high" max, vcore protection to the same setting...

It's a pain since instead of having the real voltage all the time I have to take vdroop into account and run at 1.24 -> 1.20 under load but it can last a longer GPU+CPU burn-in time.

[boondocks]

Have you tried using offsets (DVID) for overclocking? Vdroop isn't a bad thing, and the LLC will make up the extra voltage when needed.
But it could be your problem isn't related to your OC at all. Or could be you're just not providing enough vcore. Not every cpu is going to do 4.5 at 1.2v, mine for sure won't (stable).

[tehownt]

Have you tried using offsets (DVID) for overclocking? Vdroop isn't a bad thing, and the LLC will make up the extra voltage when needed.
But it could be your problem isn't related to your OC at all. Or could be you're just not providing enough vcore. Not every cpu is going to do 4.5 at 1.2v, mine for sure won't (stable).

I'll look into DVID for OC but as I mentionned, the CPU was incredibly stable on its own using "extreme" vcore loadline calibration at 1.19v/4.5ghz (I ran prime95 for hours) it's really only when I shoot a GPU bench at the same time that it crashes... (The GPU isn't OC'd and runs Furmark w/o issue for hours).

[Conners]

+1 on Vcore