Overclocking 3770K below 4.5GHz

[taq8ojh]

I am having problems finding stable settings for my CPU. I thought it was stable, but no: Prime95 throws out errors no matter what I do.

I have no idea what sort of voltages to expect/aim for around 4.2GHz (that's what I've been running at since I got the CPU, and what I aim for to be stable at right now).
Some numbers:
Auto setting or normal with zero offset results in 0.660V at idle or 1.080V under load. Those are the values CPU-Z tells me (I don't really care if they are not exactly perfect, they are numbers that basically never change, so I can work with them).
I tried various offsets all the way up to +0.060V, which results in 1.140V under load, but Prime95 still says no.

The other major voltages (VTT, PLL, IMC) are set to their defaults (1.050V, 1.8V, 0.925V)

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.

I am fairly sure it's not memory instability problem (using XMP profile DID cause fairly serious problems though, but I since started to set the memory up manually, which seems to have eliminated them), plus I am not overclocking that at all.

I welcome any suggestions at this point.
My specs are in my profile.

 

[tracerbullet]

If you take a look at the threads in this part of the forum you'll see tons of people talking about what voltages they need at 45 and other speeds to be stable. You'll see that there's a range and so it's dependent on your actual chip, but you'll get an idea. What you're running at, 1.140, obviously isn't enough and you'll find it's not really all that high anyhow.

Have a look around to see where you compare to the rest. Do a search, narrow it down to this part of the forum, look for 3770k in the titles only or maybe 3770k + voltage together in the threads. You'll find a lot of info.

 

[taq8ojh]

I did search a bit, but didn't find anything that could really help me. People are tossing around numbers like 1.3V and other unreal stuff.

Guess I will have to try higher. Apparently I have really bad piece

 

[TakeNoPrisoners]

What cooler and case do you have? Those temps are really high for 4.2Ghz.

 

[tracerbullet]

People are tossing around numbers like 1.3V and other unreal stuff.

I generally see people talking 1.200 - 1.250V or so. That's not a rule by any means, could be higher or lower, but that seems typical from what I've seen anyhow.

My 3570k wasn't totally stable even with 1.376V, I'd *love* to think that 1.140 was even close to stable. Different chip of course, but for comparison. I think if you were stable with 1.140 even with a 3770k you would actually be considered lucky, maybe not lucky but certainly I would think people would agree you didn't have a bad chip.

 

[2is]

+.060 isn't all that high really. Sounds like you need more voltage and better cooling to hit the speeds you want.

 

[Goros]

I run 4.38GHz at 1.224 under max load.

 

[Termie]

What motherboard are you using? For most boards, +0.060 doesn't result in 1.14v, it would be something higher..

Basically, 1.14v is almost certainly too low for 4.5GHz. I'd set it at 1.2v and work your way down. If that means an offset of +0.12v, than that's what it is, but your board is acting pretty strange, in my opinion. That idle voltage is seriously low, and would likely bluescreen a lot of systems.

 

[Owls]

Without knowing what board you have it looks like you got a combination of bad vdroop and a crappy chip. It also seems your chip is running really hot so it's possible you got a concave IHS and/or heatsink. Try reapplying TIM and reseating your heatsink and see if that helps. CPUz doesn't report temps very accurately so use RealTemp or hwinfo64 also.

In comparison my 3770k will do 4.8GHz with a +.145 offset and max temps under prime95 haven't exceeded 83C. And this is watercooled with a kraken x60

 

[2is]

What motherboard are you using? For most boards, +0.060 doesn't result in 1.14v, it would be something higher..

Basically, 1.14v is almost certainly too low for 4.5GHz. I'd set it at 1.2v and work your way down. If that means an offset of +0.12v, than that's what it is, but your board is acting pretty strange, in my opinion. That idle voltage is seriously low, and would likely bluescreen a lot of systems.

It's probably vdroop at work here. Probably need to add a bit of LLC for those clocks, but he'd still need better cooling since LLC is really nothing more than adding more voltage, except only when under load.

 

[taq8ojh]

One thing at a time.

I thought I had my specs listed in my profile, but then realized that feature apparently doesn't exist on these forums and I mistook it for some other place.
So, specs:
CPU cooling: CoolerMaster Hyper 212 EVO + Noctua NF-F12PWM
GA-Z77X-UD3H rev1.0, F19j BIOS
16GB of Corsair Vengeance memory (CMZ16GX3M2A1866C9)
Radeon 5850
Seasonic SS-550RM
Corsair Obsidian 550D case
That's about all the important stuff I'd say.

What in the heck is vdroop? I think it has something to do with the advanced voltage options in BIOS, but I am not messing with that at all. The famous Ivy Bridge overclocking guide says there's absolutely no reason to change those values unless overlocking way past 4.5GHz at all, and I believe it.

Temps reading:
Actually CPU-Z doesn't show any temperatures at all I just use it to get rough idea of vcore (it might not be exactly correct, but it also never changes, so I can work with it). I started using hwinfo for everything recently after accidentally opening the sensor screen and seeing the program really came a long way. The temperatures reported also match those of Realtemp, so I am happy with it.

Voltage:
After some digging in hwinfo's sensor screen it seems CPU-Z is showing really wrong vcore. Right now at 4.2GHz with Prime95 running, it says 1.104V, but hwinfo says 1.171V.
The above numbers are without any offset btw. A bit high, aren't they? Temperatures across all the cores are within 75-80 range.

Instability:
I am still in heavy doubt about the memory.
The XMP profile has two greyed out values "profile DDR voltage" and "profile VTT voltage". The latter is 1.05V (which happens to be default CPU VTT value) when the profile is disabled, and 1.3 when enabled.
I can't seem to google up what exactly this means.
Since we are talking about memory here, the latter cannot mean I should set CPU VTT to that value. I mean wtf, Intel's max recommended VTT value is 1.08V only!
Anyone can comment on this?

Also, I have some weird results with Prime95:
When I run the Small FFTs test in Prime95, it can go on for 12+ hours without errors.
When I run the middle test, slightly changed so those sizes still fit in CPU cache, it freaks out on me the most.
The last, memory-heavy test still needs more testing before I can make conclusions.
There is something weird going on but I can't nail it down.

 

[coffeejunkee]

Vdroop means the voltage will be lower under heavier loads than under light loads. It sounds counterintuitive but it's normal and according to Intel specs. It happens at stock too but the difference between high and low increases as you up the clockspeed. Therefore you'll have to overcompensate a bit at higher freq.

LLC disables the vdroop mechanism giving steady vcore but it's not really necessary for 4.2 Ghz using offset, at least not for my chip. I use +0.020 which gives me 1.136 under Linx load. Owls is probably using highest llc setting which actually supplies more vcore than specified (relatively low +0.145 offset gives him 1.376 vcore which is quite high).

 

[Owls]

coffeejunkee is correct and I still think your issue has more to do with cooling and a bad overclocking chip. When you run the Small FFTs how hot does your CPU get?

The best way to test the theory is to keep bumping up the voltage until it can pass about 20minutes of Prime95 with the middle test. If it works with a ton of voltage then it's probably a bum chip.

 

[Goros]

I would also say that taking manual control of your VCCIO and VCCSA voltages is a must, since AUTO can cause them to run too high and damage your chip. You should be able to drop your CPU PLL from auto (1.8v) to 1.60 or 1.65v without any issues, and it will take your temps down a little bit helping you get stable.

Your CPU LLC is something you probably have to take control of as well to get this chip stable, since your vdroop is so bad. I have manual control over all my voltages and frequencies involving the CPU and memory and have done so 24/7 for over a year with 0 issues. Just stay within the safe zones on volts and temps and you'll be fine.

Don't trust your motherboard not to overvolt stuff when set to auto, it most definitely can, and might. Lots of people were cooking sb-e chips because of this when they first hit the market.

 

[taq8ojh]

I would also say that taking manual control of your VCCIO and VCCSA voltages is a must, since AUTO can cause them to run too high and damage your chip.

But I already do that! I set them all to default values. I am well aware of the cons (there are usually no pros anyway) of auto settings. This is not the first time I am playing around with overclocking btw.

...PLL from auto (1.8v) to 1.60 or 1.65v without any issues, and it will take your temps down a little bit helping you get stable.

Interesting. What does PLL affect then?
In fact, does anyone have a link with explanation of all the voltages specifically on IB platform?

Your CPU LLC is something you probably have to take control of as well to get this chip stable, since your vdroop is so bad.

That's the exact opposite of what "everyone" says about this board when it comes to power-related stuff.

edit:
I got some screenshots from BIOS here:

no idea what and why would I change here so I kept it on auto

This is what I refuse to touch considering my VERY light overclock and this guide.

coffeejunkee is correct and I still think your issue has more to do with cooling and a bad overclocking chip. When you run the Small FFTs how hot does your CPU get?

Just to have some base values, I switched to stock speeds for now (so, 3.7GHz at full load).
Temps are in the 65-70°C range, reported VID is 1.106V.

I honestly don't care about cooling at this point (as long as temps don't exceed 90°C), when the primary problem is instability. Plus pumping ridiculously high volts into the CPU is not something I would do anyway. I would rather accept the fact I have shit chip I can't do much about.

 

[Owls]

Increasing the voltage temporarily, even for an hour, won't do anything to kill the chip since you are only testing the upper bounds of it a little bit at a time. Still, I'd like to find out what is causing your instability if it's the board, cpu, or both. I was previously "stuck" at 4.5ghz with air cooling so anything is possible.

 

[taq8ojh]

I have no idea either. There are so many variables in this case.
I decided to go back to defaults and test the shit out of memory first, because even with default CPU speeds, Prime95 throws out an error fairly soon if XMP profile is enabled - not the most memory-intensive test though, one that mostly stresses just the CPU, but utilizing most of its cache. That's a WTF situation by itself.

So what I have in mind for now is:
A) find out whether I can get 100% stable with default speeds and memory set up manually, but still with the same values defined by XMP (if not, the memory will be sold and replaced with different brand)
B) test various combinations of secondary voltages in order to find out whether it's even possible to get stable with XMP enabled at all (it could be more serious incompatibility with the board than I first thought then)

If A) works, will ramp up CPU speeds step by step and see what happens. It must be something really creepy, because I could pass hours of Small FFTs test at 4.2GHz with negative offset of like -0.050V.

 

[Owls]

Try prime95 at bios defaults with 1 stick of ram and see if that could be the cause? I have the same memory as you except 8gb density and it's running at 2000mhz at 1.65v

 

[Goros]

Start with memtest86+ 4.20 if you want to test your memory. Anything running inside windows is going to introduce variables that can color the testing results.
Prime95 is for the CPU, while it does test memory it's not the primary function.
All your memory voltage settings crank to auto except for the base memory voltage which you keep at 1.5v or run off the XMP profile.

CPU PLL of 1.5-1.7 is usually all you need to OC Ivy Bridge, more can cause issues. Look it up, for whatever reason IB likes lower PLL.

Your temps are way too high for stock volts and speed. Something is wrong. My idle at 24/7 4.4GHz only pulls 34c idle and 82c at max load. I don't use speedstep so my rig runs 4.4GHz full time.

I can't speak about your specific board because I use a P8Z77-WS. What I can tell you is I run overclocked RAM, overclocked CPU, 107 BLCK, and overclocked 580 Lightnings in SLI and I'm rock solid stable despite everyone telling me I shouldn't be.

LLC effects VDroop. If you don't want to tweak LLC, you just need to raise your offset until you see the target voltage you are going for and then test for stability. As long as you don't pass 1.4v you have lots of wiggle room. The post you linked specifically says to set your LLC but not your PWM settings here
"You will also want to set LLC which is under the 3D power menu, the LLC should be set to Turbo for a slight droop, or Extreme for no droop at all. The LLC on these boards is rock solid, what you set is what you get, and nothing other than that. If you want you can also mess with the other PWM settings, but that shouldn’t be needed as Ivy Bridge doesn’t pull enough power to warrant those changes under air cooling. I recommend a slight drop of voltage under load, this might help with temperatures.*"

My biggest concern are your temps. Something is very wrong, and it's probably effecting your stability (and the guide you linked states about 4x that temps are vital to stability of your overclock).

If you're going to keep referring to a guide, read it thoroughly and understand it and follow it exactly before telling people trying to help you that they are wrong.

 

[taq8ojh]

Prime95 is for the CPU, while it does test memory it's not the primary function.

Wrong. Neither is primary, it's reliable complex testing system.
Oh and just for the record, stress-testing is not the primary function of Prime95 at all

All your memory voltage settings crank to auto except for the base memory voltage which you keep at 1.5v or run off the XMP profile.

Already have that (minus DDR voltage which I will fix on next reboot) as you can see in the picture.

CPU PLL of 1.5-1.7 is usually all you need to OC Ivy Bridge, more can cause issues. Look it up, for whatever reason IB likes lower PLL.

Probably worth a shot. Why is the default value 1.8V then? That's pretty significant difference.

Your temps are way too high for stock volts and speed. Something is wrong. My idle at 24/7 4.4GHz only pulls 34c idle and 82c at max load. I don't use speedstep so my rig runs 4.4GHz full time.

I don't know. When I go back to the usual 4.2GHz, load temps are in 85-90 range, sometimes very slightly above that. I find it acceptable for the rather small heatsink I use.
I will try to reapply paste and reseat it though.
Delidding is also in the plans. Maybe I am unlucky with exceptionally shitty IHS with big gap inside.

 

[tweakboy]

I am having problems finding stable settings for my CPU. I thought it was stable, but no: Prime95 throws out errors no matter what I do.

I have no idea what sort of voltages to expect/aim for around 4.2GHz (that's what I've been running at since I got the CPU, and what I aim for to be stable at right now).
Some numbers:
Auto setting or normal with zero offset results in 0.660V at idle or 1.080V under load. Those are the values CPU-Z tells me (I don't really care if they are not exactly perfect, they are numbers that basically never change, so I can work with them).
I tried various offsets all the way up to +0.060V, which results in 1.140V under load, but Prime95 still says no.

The other major voltages (VTT, PLL, IMC) are set to their defaults (1.050V, 1.8V, 0.925V)

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.

I am fairly sure it's not memory instability problem (using XMP profile DID cause fairly serious problems though, but I since started to set the memory up manually, which seems to have eliminated them), plus I am not overclocking that at all.

I welcome any suggestions at this point.
My specs are in my profile.

You can OC to 4Ghz without touching any voltage or setting. Want to go past 4Ghz you have to raise the voltage thus creating more heat. Need good cooling if your gonna OC. let us know, gl

 

[Goros]

Wrong. Neither is primary, it's reliable complex testing system.
Oh and just for the record, stress-testing is not the primary function of Prime95 at all

I know, but if you want to narrow down the issues you should probably restrict testing to one system at a time and not your CPU, Memory and FSB all at once like Prime95 does. Test your memory with memtest86+ 4.20 and test stability with prime95 after the memory checks out as good. Too many variables to get a reliable answer and you'll be chasing your tail until you narrow down where the issues come from.

 

[2is]

Based on my own personal experiemce, Prime 95 blend is a MUCH better memory test than Memtest 86

 

[taq8ojh]

I might be onto something here.
Ramped up IMC voltage by 0.1V and passed 16 hours of the problematic test.
Then I switched back to default and got an error in two hours, which surprised me a lot, because normally it would happen within 30 minutes. The error itself didn't surprise me at all of course.

Next step was raising IMC again, this time only by 0.05V. So far four hours and still going.

 

[Redoitall]

One thing at a time.

I thought I had my specs listed in my profile, but then realized that feature apparently doesn't exist on these forums and I mistook it for some other place.
So, specs:
CPU cooling: CoolerMaster Hyper 212 EVO + Noctua NF-F12PWM
GA-Z77X-UD3H rev1.0, F19j BIOS
16GB of Corsair Vengeance memory (CMZ16GX3M2A1866C9)
Radeon 5850
Seasonic SS-550RM
Corsair Obsidian 550D case
That's about all the important stuff I'd say.

What in the heck is vdroop? I think it has something to do with the advanced voltage options in BIOS, but I am not messing with that at all. The famous Ivy Bridge overclocking guide says there's absolutely no reason to change those values unless overlocking way past 4.5GHz at all, and I believe it.

Temps reading:
Actually CPU-Z doesn't show any temperatures at all I just use it to get rough idea of vcore (it might not be exactly correct, but it also never changes, so I can work with it). I started using hwinfo for everything recently after accidentally opening the sensor screen and seeing the program really came a long way. The temperatures reported also match those of Realtemp, so I am happy with it.

Voltage:
After some digging in hwinfo's sensor screen it seems CPU-Z is showing really wrong vcore. Right now at 4.2GHz with Prime95 running, it says 1.104V, but hwinfo says 1.171V.
The above numbers are without any offset btw. A bit high, aren't they? Temperatures across all the cores are within 75-80 range.

Instability:
I am still in heavy doubt about the memory.
The XMP profile has two greyed out values "profile DDR voltage" and "profile VTT voltage". The latter is 1.05V (which happens to be default CPU VTT value) when the profile is disabled, and 1.3 when enabled.
I can't seem to google up what exactly this means.
Since we are talking about memory here, the latter cannot mean I should set CPU VTT to that value. I mean wtf, Intel's max recommended VTT value is 1.08V only!
Anyone can comment on this?

Also, I have some weird results with Prime95:
When I run the Small FFTs test in Prime95, it can go on for 12+ hours without errors.
When I run the middle test, slightly changed so those sizes still fit in CPU cache, it freaks out on me the most.
The last, memory-heavy test still needs more testing before I can make conclusions.
There is something weird going on but I can't nail it down.

Hello I would suggest you raise your voltages at least to 1.3 and start decreasing in increments of .05 If your chip is not stable at 1.3 then I would consider RMA because like someone else said here most of these chips run at 1.20 at 4.5> Imagine what you would have to do to get to 4.8 at that rate you would be at 1.6 . Don't get me wrong all chips are likea lottery and my 3770k is one of those that after 4.9 there is no stability. I could raise the frequency a notch but no matter how much voltage it would just fail. I use an H-100 but mine is delidded with LMP on die to IHS. Your temperatures may very well be a factor. Also try to lower the memory frequency to the lowest.Try also running on manual not XMP. I would also check windows event kernell WHEA. Believe me those errors logs help me a lot to achieve stable clocks at 4.7 which is what I run the chip when I am Gaming. I thought I was stable with lower voltages but found out that needed to raise from 1.315 to 1.328 @4.7 although I was able to go in every run 24 hours of prime 95 but still getting WHEA errors. I would also ask IDONTCARE he has shed some very helpfull info on the stability of these chips