Stability gone on my 3770K

[Octopuss]

I just ran Prime95 memory test and it crashed within an hour, which I found really weird, considering everything was perfectly ok a few months ago when I built this PC.
I have all BIOS settings related to memory on auto, performance enhance is set to turbo.
CPU voltage is, I believe, set to normal with -0.045V offset.

Is there anything I could try in advanced voltage settings that could bring memory stability back? I have to point out the Small FFTs test doesn't give any errors at all.

 

[Octopuss]

I have to slightly correcy myself: I did have Performance Enhance set to Extreme (which obviously did work in past though). After swithing to Normal, Prime passed well over an hour without problems. I have to test it further, because it still crashes when I use Turbo, which is thedefault value. This is not right.

 

[brandonwh64]

Make sure you do not use the predefined overclocking modes that the board comes with. Each chip and board is different so it may do things like ramp up the voltage here and there thinking the CPU needs it when it does not.

 

[Octopuss]

I didn't touch all that much. My original goal was some reasonable overclock from default speed at stock voltages (or lower), which I kind of achieved after lenghty discussions with cadaveca (thanks again). I extensively tested it and it was stable. Weird...

Attachments pretty much sum all the important stuff I changed up.

 

[cadaveca]

many possibilities as to what is causing the issue. Could even just be new software. You'll need to test a bit more and get some crashes, and provide the info about those crashes, before we can really assess what's going on.

You aren't running really high clocks, or voltages, and sometimes things have a tendency to "settle in" a bit after first OC as the system gets "used to it", so what you are reporting is relatively normal.

 

[Octopuss]

That's a possibility I thought about, some sort of "burn in".

BTW, are there any specific settings that could let me increase the voltage offset? Something in the advanced voltage options or whatever.
(of course provided I get back to rock stability again)

 

[Octopuss]

This is seriously weird.
Prime95's "In-place large FFTs" keeps crashing in under an hour (usually much faster, like 30 mins or so), while Small FFTs (which as I understand is stressing CPU the most) can happily pass a few hours.

The original DVID offset was -0.045V, and the Small FFTs test could go without errors even at like -0.055V (wouldn't think it stable at all, but it's interesting). I've been repeating the former test all the way down to -0.020V and then stopped, because the problem must lie elsewhere. But where?
All the other major voltages are set manually to their defaults (VTT=1.050V, PLL=1.800V, IMC=0.925V) just like always. I also tried to set BCLK and multipliers manually (to the same values though), and it didn't do anything as expected.

One more thing I am trying right now is going back to the original offset, but having Performance Enhance option set to normal instead of turbo (what in the hell does it do anyway?). After that I am completely out of ideas. (edit: fail in 6 minutes. fking hell...)

 

[Octopuss]

To add to the weirdness.
DVID offset doesn't seem to do anything. I compared min and max values of VCORE in CPU-Z during idle and load, and then switched from -0.045V to pure auto setting, and the voltage was absolutely identical. Just wtf...

 

[cadaveca]

Re-flash BIOS to previous version, and then flash to latest. Sounds like you've got the UEFI bug that's making the rounds lately.

 

[Octopuss]

Oh? Can you elaborate further?
I haven't flashed anythhing since... october. Whoa.

Is that something that happens out of the blue?

edit: the weird thing is positive offset seems to work (or increase VCORE somehow at least).

 

[cadaveca]

Hmm, I dunno, I have found some bugs with UEFI caused by Windows crashes, I think. Changes in settings have no effect. A couple of brands have this problem right now.

But if you can change some settings, that's odd, kinda similar, but not quite.

And yes, out of the blue, after a crash that is software- OR hardware-related.

 

[Octopuss]

Ok I reflashed it as you suggested... and exactly the same numbers. This is seriously weird. Especially the reported VCORE under load.

And Prime95 is crashing, even with positive offset - tried +0.010V so far. I don't believe I'd need to increase anything for just 4.2GHz though???!

 

[Hood]

UEFI bug?

Just earlier tonight I had similar problems with my Asus P8Z77-V; after replacing the stock TIM on my H100 with Arctic Silver 5, I noticed that changing the multiplier in BIOS didn't change the clock speed in Windows. After several attempts using every setting available, still the same result - zilch. I had updated to 1805 when it came out a few weeks ago. I remembered reading another post about someone with the same problem who re-flashed the previous BIOS and solved the problem, so I flashed 1708 using USB flashback and now I can change settings.
I think I'll just keep this BIOS until a new one comes out. I never had any problem with 1708 and also my temps seem to be lower.

 

[Octopuss]

I switched memory completely for some 2GB piece of shit, and it passed like 9 hours. Could be coincidence, but after plugging my regular memory back in (in different slots now just to give it a go), Prime95 crashed in 22 minutes. I will test this once more. Interesting fact is MemTest passed without problems. Could it be some weird incompatibility?

 

[Octopuss]

I think I am fucked. Flashed a lot newer BIOS (which says improves some compatibility as well) and sure enough, Prime95 crashed in minutes.
There's no way in hell I can RMA this since no store gives a flying fuck about what program a random hippie uses for stability testing. They'd throw Memtest at it at most and would charge me for RMAing working product.

 

[MxPhenom 216]

I think I am fucked. Flashed a lot newer BIOS (which says improves some compatibility as well) and sure enough, Prime95 crashed in minutes.
There's no way in hell I can RMA this since no store gives a flying fuck about what program a random hippie uses for stability testing. They'd throw Memtest at it at most and would charge me for RMAing working product.

Christ,

You do not RMA to the store! You call Asus support or whatever manufacture and get an RMA ticket, and send it in. They process the RMA, and they send a replacement back.

 

[BiggieShady]

I switched memory completely for some 2GB piece of shit, and it passed like 9 hours.

This and the fact MemTest runs fine ... hmm .... could it be that Prime95 heats up the cpu and it's huge heatsink which can be close to huge fins corsair vengeance has?

 

[Octopuss]

Yeah but there are serial numbers written on the invoice from the store I bought the stuff in. I would value that warranty over anything.
What you say is definitely a possibility as well, though.

UPDATE:

I should have friggin' guessed - the dumbass XMP profile! I'll have to set timings manually and experiment with values, but manually setting 1866MHz and having timings on auto doesn't seem to give me any errors.

 

[Octopuss]

And it works. Who would have though. Hope Gigabyte will fix this with future BIOS somehow.

Btw, I even managed to make the memory run at 2133MHz.

The big question:
Is going from 1866@9-10-9-27 1.5V to 2133@9-13-13-36 1.65V worth the pretty big jump in voltage?

 

[jsfitz54]

Nobody mentioned this here but best practice for updating Bios is to set Bios to Default settings before attempting any sort of update/flash. This means no overclock settings.

Since I ran in to a problem once with a UEFI flash, I have taken to disconnecting all HDs and SSDs for this step and setting up Bios or UEFI by resetting after the flash the AHCI, XMP, etc. then rebooting. I then tweak memory and voltage settings, rebooting to see that settings are working, shut down, then reattach the drives. Seems like more work than it should be but it appears to make certain that the flash and settings work without corruptions.

 

[Octopuss]

Never heard about that before. I do however load optimized defaults after each flash before I start changing stuff.

 

[jsfitz54]

I do however load optimized defaults after each flash before I start changing stuff.

Loading optimized defaults after the flash is the problem area. It needs to be done before you flash.

Once you load optimized defaults after the flash (second time), then you need to reboot again and then adjust settings. This assures a clean deck.

Steps would be:

System off, disconnect HDs and SSDs.
boot to UEFI/Bios
load defaults, set primary source drive to optical or usb drive (flash source), save, exit
boot with flash program, flash gives ok to restart
boot
load default values again, save, exit
boot
program AHCI, XMP, etc
power off
reconnect drives
boot to Windows.

As a side note I have been using a USB drive to flash from. The .bin file you would scroll to (pick from list), if there are more than just the .bin on the USB drive. Some motherboards may want a dedicated/clean source for file and by that I mean a USB with only the .bin file.

I also use YUMI http://www.pendrivelinux.com/yumi-multiboot-usb-creator/ to load ISO files for updating SSD Firmware. You can also use YUMI to load Windows, Ubuntu, etc... any ISO file.

 

[Octopuss]

Like I said, never ever heard I should load default before flash.

Once you load optimized defaults after the flash (second time), then you need to reboot again and then adjust settings.

That's what I do.

 

[LAN_deRf_HA]

Idk if you have to disconnect drives. This is what Asus says to do:

1. Reset your UEFI to Optimized Defaults, reboot, enter UEFI and then flash to the new UEFI build.
2. After flash and restart, enter Windows/Linux, then exit and shutdown.
3. Remover the power cord from the power supply, wait thirty seconds, press Power On button to clear any remaining charge, clear CMOS for five seconds, reattach the power cord, power on the system, enter UEFI, set to Optimized Defaults, save and exit.
4. Enter UEFI, set your system to customized settings based on your system configuration (RAID, OC, etc), save and exit.

 

[Octopuss]

That sounds pretty stupid to be honest. Similar to those "regtweaks" which in fact don't do anything but have been circulating the internets for ten years.