Overclocking 3770K below 4.5GHz

[jcniest5]

When you get a bad overclocker, you will kill it before you stabilize it. JJ from Asus suggests a max of 1.350v for anything up to 4.8Ghz. If that isn't stable, your chip is a bad overclocker. Might as well forget about overclocking at this point.

 

[Goros]

He should be able to get stable, he just never set his LLC like they told him to so his offset voltage is too low to stabilize under load. He's peaking at 1.15 and needs to be closer to 1.2.

His temps are getting too high to allow that type of voltage, so he's stuck until he fixes or replaces his cooler.

He probably has a fine overclocker, just not delivering it the voltage it needs to get there. 1.15 won't do 4, let alone 4.2+.

My settings would go like this - XMP profile #1 loaded, then take manual control of all the memory timings (take the first set of timings and set them to whatever spec is for your memory at that speed), then set the voltage to 1.5 and all the other memory voltages to auto.

I take manual control of all the other voltages like VCCSA (1.050), VCCIO (1.08125) and CPU PLL (1.65) while turning off CPU spread spectrum and setting LLC for extreme as well as CPU power phase and power duty control for extreme, but power response control to regular.

Leave the PCH voltage alone.

I'm not sure how these settings line up between brands, but the voltages I set are within safe limits.

 

[aarontpx]

My 3770 (not delidded) runs at 1.2v at 4500 mhz. The max overclock I can run prime95 stable at is 4700mhz at a voltage of 1.3. My temps reach a max of 70c during small fft test. I use an h100i.

 

[taq8ojh]

You were saying?

 

[Goros]

Just me, but those temp numbers seem kinda meaningless since the min and max temps are 3 degrees away from each other (69 to 72).

Looks like you turned it on to check temps right before taking the pic while the processor was at full load.

We also talked about temps because you mentioned how hot you were running in your OP.

Temperature-wise, at +0.060V the cores report 80-85°C at 22°C ambient. That means I don't really want to go much higher with voltage.

If you were unstable at 1.14v at 80-85c you would have been over 90c at 1.2-1.25 if I had to guess.

 

[akugami]

Goros, my i7-3770k is using 1.15v vcore and doing 4.2ghz. Using a Gigabyte GA-Z77X-UD5H. It's been through 3 hours of Lin-X as well as 24 hour Prime95. So yes, it is stable and it has been tested.

 

[Goros]

I believe it, but not every chip bins out that well (you are in the minority) and he has other issues to address besides voltage.

There is no reason for a chip that is running that voltage and speed to pop up to 85c under load with the cooler he is using.

He also didn't understand vdroop, offset voltages, or secondary voltages per his first post and subsequent. He was having stability issues at stock.

 

[Owls]

Can you post your event log? I have had stable clocks but tons of whea errors

 

[akugami]

Unless something is seriously wrong, at stock, no one should be having any stability issues.

The default temp is 1.2 according to the BIOS but I dropped it and still was able to overclock to 4.2ghz and acceptable temperatures so I'm extremely happy with the results. Was not a cherry picked CPU, just walked in a Microcenter and asked for one.

As for vdroop, offset voltages, etc, I usually research what the heck all of the BIOS options are before I even attempt to overclock a CPU. You can seriously (and permanently) screw up your system if you don't know what you're doing. Even though overclocking today is dead stupid simple, you still should know what you're doing. I actually didn't start overclocking my CPU until 2 weeks after I bought it because I wanted to make sure I know what all of the options I'm changing do.

 

[tracerbullet]

I have had stable clocks but tons of whea errors

Most would not call that stable.

 

[akugami]

Can you post your event log? I have had stable clocks but tons of whea errors

I did not think to check for WHEA errors. There's isn't a ton of them but I've been trying out different settings on the system so I'm not sure when the errors pop up. Also, I didn't log my minutes on what tools I was using at what times so some of the errors may have popped up when I was testing using Lin-X which puts an unrealistic strain on the CPU's.

Back to testing I go then. Should probably run Memtest86 to make sure it's not the RAM.

 

[2is]

Can you post your event log? I have had stable clocks but tons of whea errors

That's not stable. Hopefully you aren't continuing to run it that way or you're begging for data corruption/loss.

 

[taq8ojh]

He also didn't understand vdroop, offset voltages, or secondary voltages per his first post and subsequent. He was having stability issues at stock.

1) I don't need to understand vdroop because I don't need it, the power supply of this particular board is rock stable
2) I do understand them. + means it goes up (from whatever the CPU thinks is default for spefic resulting frequency) both under idle and load, - means it goes down.
3) I couldn't google any definite answers to my questions regarding secondary voltages, and every guide I found said something else, or was written for SB, so there's not much to work with, unfortunately.
4) Stability problems, yes, but not CPU-related. My memory is picky, obviously. This is now 90% sorted and I am still exploring other possibilities how to go around this.

P.S. WHEA errors are serious stuff. Usually I would say they are MUCH more serious than simple instability: last time I had them, at least one of the cores was completely fried. The thing that sucks the most is you don't normally check event log for this sort of stuff unless someone specifically tells you to do so, because you just don't expect to find anything related to overclocking there. I didn't even know what a WHEA error was before that.

 

[Goros]

1) I don't need to understand vdroop because I don't need it, the power supply of this particular board is rock stable

Vdroop has nothing to do with the power supply or stability of the motherboard. All boards and cpu's suffer vdroop. You adjust your LLC to give the power you tell the mobo to send to the CPU actually to the CPU. When you set voltages manually in your system (with speedstep and c states off), not using offset, and you type in 1.26v but actually only get 1.225 or some-odd volts to your processor and then under load get 1.21 and it causes errors, that's vdroop and it's normal. You adjust LLC (like every IB OC guide tells you to) until you find a setting where when you type in 1.22 you get 1.22 actually delivered and it stays within .01 of that under load. It takes a mess of going back and forth between the bios and prime95 to figure out, but once you do you can switch to offset mode, adjust your speedstep and c-states and try to find the offset voltage that works best for you, once you have determined the proper level of LLC.

2) I do understand them. + means it goes up (from whatever the CPU thinks is default for spefic resulting frequency) both under idle and load, - means it goes down.

If you understood offset, you wouldn't argue that .060 is too much when you only get 1.15v delivered to the CPU. Total voltage under load voltage adjustment is what offset is there for, and you need to adjust your LLC which will change how much that offset number actually delivers to the CPU under load. Offset is your max additional power under load, and c-states with speedstep adjust the volts delivered to your system when it's idle. You literally have to continue messing with offset until it gets you where you want for max volts under load. Adjusting your LLC will keep you from getting into the red zone (where AUTO would start cooking other parts of the CPU). For example, a LLC of extreme with offset adjust of +.010 on my mobo results in a voltage of 1.22, but with LLC on auto or regular it needs almost +.055 to do the same thing. LLC keeps you safe, not the other way around.

3) I couldn't google any definite answers to my questions regarding secondary voltages, and every guide I found said something else, or was written for SB, so there's not much to work with, unfortunately.

There are 2 or 3 really good guides on OCing IB out there, mostly at forums dedicated to OCing as their primary function.

4) Stability problems, yes, but not CPU-related. My memory is picky, obviously. This is now 90% sorted and I am still exploring other possibilities how to go around this.

Memory is tricky because when you overclock it you change the frequency it presents to the VCCIO and it can cause stability issues in both depending on your mobo's binnings. Most memory is just 1333 that was binned higher and tested out at a better speed than it should have. I always buy memory with a ton of headroom so I can either increase the frequency or tighten the timings for my overclock - but you pay for it.

I'd adjust your VCCIO a little bit - just keep it under 1.1v, and you will probably have to raise VCCSA as well - gigabyte boards are notorious for needing both within a few steps of each other in actual presented voltage to get stable - might be worth trying lowering the VCCIO and raising the VCCSA if you don't just want to set them both to their max safe numbers.

 

[taq8ojh]

So far I am rock stable (21 hours of priming) with VCCSA at 1.025V, or in other words, +0.1V bump. This is with 1.080V vcore under load btw.
This is unlikely to be the lowest possible value, though. 0.970V said big FU after 18 hours (almost stable), but after that was instant fail all the way up to and including 0.990V.
Then I decided to give it a blind try as I said in the first sentence and it worked.
Now I am going to experiment with VCCIO to see what happens.


P.S. Idle temperatures are uniformly 30°C. it's rather cool weather these days, though. Load temperatures today are more or less 70-75°C, that's at 4.2GHz.

 

[Unoid]

Good. Now can this thread die?

Edit: sorry forgot this wasn't OT

 

[taq8ojh]

No, but trolls can go away.

 

[taq8ojh]

Look what we have here - my old thread.
I am still on the quest for stability, only this time it's 4.4GHz. Not much luck thus far.
Just to recap some basic stuff: board is GA-Z77X-UD3H which has 7 LLC levels: normal, standard, low, medium, high, turbo, extreme.

I tried various combinations of lowest LLC and increasing offset, and turbo with negative offset.

With lowest LLC I was getting WHEA errors up to and including +0,035V.
Interesting thing happened when I switched LLC from lowest to 2nd lowest while sticking with +0,040V, and got WHEA error after 13 hours. wtf?

lowest LLC, +0,040V: Prime95 crapped out on me after 21 and 17 hours - so far the best result of them all. Vcore was in 1,116-1,152V range, with 1,129V average.
When I jumped to +0,045V, I got an error after being shy of 10 hours. Weird.

Then I started messing with LLC on turbo. Due to the amount of time it takes to test just one offset, I only went through a few so far, starting at -0,045V, and stopping at -0,035V due to temperatures getting out of hand.
-0,045V happened to give the best result with error appearing after 17 hours. Average vcore was 1,128V (I didn't note down the range, but it wasn't big).

What do you suggest I try next? While someone might say I am stable, for me it's 24 hours without errors or nothing.

P.S. My original goal to get 4.2GHz stable was achieved with zero offset and +0,1V IMC (=1,025V).

 

[taq8ojh]

Hm, nailed it. For some weird reason, temperature drop made huge difference.

Got Prime95 error after 31 hours, which I finally am willing to take for stable.
Lowest LLC level, +0,040V offset.
Average vcore during that interval: 1,133V (1,128-1,152).

Not sure if I want to stop there or look for possibly lower vcore just for the hell of it (or 72 stable hours just for fun).

 

[taq8ojh]

WTFLOL update on the whole subject.
Damn memory modules were faulty from the very beginning. How do you call it, dead on arrival?
I just never noticed and thought Prime95 giving me errors had something to do with CPU instability.
It also got slightly better if I raised IMC voltage a LOT, confusing me even more, making me think my board was picky instead.
The dumbest thing was not ever bothering with Memtest, assuming the memory couldn't be bad since it was completely new, blaming the board/CPU.

I also RMAd the board twice lately, and now I have newer revision, which I hope can generally help things a little bit too.

So yesterday I just set BIOS like before, only I left the IMC at its default (0.925V, 1.050V before), set turbo to 4.4GHz, and so far 21 hours passed. I also did a "quick" 10 hours memory-intensive test (Prime95). No errors either.

 

[hawtdawg]

mine does 4.5@1.3v. At first the temps were close to 90C with an NH-D14, so I delided that sumbitch, and now I'm in the 60's. I can run it as high as 4.8 without temps being a problem, but I have to give it a lot of voltage, so I just leave it at 4.5

 

[taq8ojh]

I am currently sitting at 4.4@1.130V average under load.

 

[guskline]

I am currently sitting at 4.4@1.130V average under load.

Great news. My 3770k rig below is set to 4.4Ghz with no change to the vcore. I run a little higher at 1.23v when running Intel Burn test etc but overall real solid. I downloaded Intel's Overclocking and testing software and my rig was stable after an hour at 4.5Ghz but the temps were a bit too high for me.

I'm glad you found the problem. Of course all of the attention will be on the Haswell chip. However, your 3770k is so close in performance you probably won't notice any difference. Have fun.

 

[taq8ojh]

Yea I don't see any reason to upgrade especially since I'd have to get new board, which is not gonna happen so long as I keep my sanity

 

[2is]

mine does 4.5@1.3v. At first the temps were close to 90C with an NH-D14, so I delided that sumbitch, and now I'm in the 60's. I can run it as high as 4.8 without temps being a problem, but I have to give it a lot of voltage, so I just leave it at 4.5

Are you running bare die or did you reapply the lid after reducing the gap and swapping compounds?