• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

4770k overclocking problems

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

TurboJ

Registered
Joined
Dec 22, 2012
I have been doing stability tests on my 4770k and I'm getting inconsistant results.

I don't know how I can reliably stress test my CPU, because only ITB and Prime95 seem to make it BSOD - but they also cause the CPU to run silly hot. I don't see any point in stress testing max stable OC when Prime95 sees temps of 93-96C and games max out at 63.

I also used the Asus RealBench, but the stress test option on it causes consistent BSOD 0x000000C5 - even at stock CPU settings!
I believe I even saw the same BSOD code once when Metro Last Light crashed on me. So RealBench gives me this BSOD every time I run it, but now I don't see it anywhere else anymore. In normal use, all seems to be stable now.

I have been running Prime95 blend now to find an optimal Vcore value, and it's acting strange. Cores are synced at 42 and when I was running 1.14V, some workers just randomly stopped all the time and log showed a fatal error of a calculation done wrong.

But when I increased Vcore to 1.16 it ran stable for 30 minutes, then BSOD 0x00000124. Next run just to make sure: now 124 in just two minutes.
Vcore 1.17: BSOD 124 in 5 minutes.
I'm now trying 1.18 but temps are getting out of hand...

Prime is now running, max temp 94 but it isn't throttling. If I need still more voltage, I can't really. But I don't want to decrease core multiplier, as in normal use the temps aren't nearly as high.

So can someone help me with these inconsistent BSODs and maybe suggest a stress test that is reliable but allows me to run real life voltages?
 
Blend is for memory and CPU. If you want to isolate the CPU run small FFT.

What cooler are you running that that thing? That is awfully warm for the voltage.
 
Thermalright TrueSpirit 120m - I know it's a small heat sink, but it should be able to take this conservative OC. And there shouldn't be paste issues, I did many test fits on it to see how the paste acts.

And, well, it DOES - gaming or benchmarking temps never seem to reach 70C.
But Prime95 on some of its cycles goes up to 96 degress (still not throttling afaik). And, I'm running with six fans + side
panel removed...
 
That is small?? Looks decent, but who knows. Those temps seem quite high though even with that cooler... Those are essentially stock volts (a bit higher)... Doesn't make sense... Perhaps remount that cooler... Normally a gaming type load versus P95, the temp difference is around 10C, not 25+. So something isn't right there.
 
Reattached the cooler and applied new paste.
Did nothing to the temps. It's the P95 small exactly that makes the temp skyrocket.
The max so far has been 98 C.

I don't know how this can be.. A faulty CPU?

...Oh, and Aida64 CPU stress test is giving me a max temp of 58C.
That's about the same as on most games I play.

RealBench CPU stress: max temp 65C.

P95 small jumps to 98 in a minute.

WTF?
 
Last edited by a moderator:
Can someone help me with this?

Shouldn't the CPU throttle at 98C? It did not on my rig.
Now I'm getting 84C on P95 small FFT even with 3.9 GHz and 1.08V.

I have all BIOS settings stock, apart from Vcore and Core multi. Is there
any possibility that stock settings would put some other value so far off
that it would cause this heat issue?

Could someone please tell me the possible BIOS settings that might cause this overheating? I have an Asus Gryphon board.

And if settings are correct, the only fault possibility is the CPU itself.
A faulty mother board wouldn't do this, right?
 
Last edited:
A faulty CPU or mobo won't cause that, no.

I think these throttle around there, sure, maybe 100C? I do not recall. Run RealTemp and see if it hits the thermal limit (at the bottom it will let you know)

Run only the FPU test in AIDA64 and see what happens. But that said, the newest version of P95 adds instruction sets so perhaps it is testing all of that?
 
If the chip wont even run Prime95 small FFT at 100% stock, then i'd consider RMA, you may end up with a CPU from a much better batch.
 
New development: I had already taken the CPU cooler off and on again many times and tested two different thermal pastes.

Today I took it off again, and also removed and reattached the CPU itself.
I bought some Arctic MX2 thermal paste, and something has changed.

I did tests with 1.08V and 3.9 GHz. Before temps were 84 max, and after
the CPU refit and new paste temp maxed at 74 C! Idle temps dropped about 5-6 degrees as well.

I also noticed something else. One should not trust Asus's Thermal Radar 2 tempe readings. I was just assuming that the CPU cooler fan was running max rpm as that's how I had it set. BUT - I noticed that Thermal Radar 2 shows 13 C lower CPU temps than any other temp monitoring software! So the CPU fan was not running 100%.

Now, the Thermal Radar 2 is great in itself - I love how it allows me to run all fans off the mobo and control them all according to temperature. But the temp readings are off.

Right now I'm running P95 small FFT on the new fan profile and 4.1 GHz @ 1.12V. It has now been stable for 20 minutes and running. Avg temp is now 74 C although it is maxing at 80.

So something did change. I don't know if it was the MX2 paste, or if the CPU wasn't getting 100% good contact on the socket before.

As for voltage, I think it is not bad really to get at least close to stability at 4.1 & 1.12V
But even now, I think 80 max on P95 is still a bit much for 1.12V.
BUT, I do see that this latest P95 with the "improved" AVX commands is very, very taxing on temps.

So how do these new volts and temps look to you?
 
We always say to get your temps off Realtemp, Coretemp or Hwmonitor... MX2 is good paste, so I bet, as was mentioned earlier, it was something in the mount of the cooler.

The CPU not getting good contact should not cause a temperature increase.
 
Yeah, well I wasn't using Thermal Radar for monitoring, but as it was controlling the fans according to its own sensors, well, you see what happened...

So does this function all now seem normal or what do you guys say?
 
We always say to get your temps off Realtemp, Coretemp or Hwmonitor... MX2 is good paste, so I bet, as was mentioned earlier, it was something in the mount of the cooler.

The CPU not getting good contact should not cause a temperature increase.

I have found that open hardware monitor is decent too, its like HWmonitor thou. Both work, i just use open hardware since i can import information into the aqua computing fan controller.
 
Yeah, well I wasn't using Thermal Radar for monitoring, but as it was controlling the fans according to its own sensors, well, you see what happened...

So does this function all now seem normal or what do you guys say?
One thing you could do for the fan profile is to do a hard step. Say, keep it nice, low, and quiet up to about 65* or so (with whatever offset you know should be set for the Asus software) then just ramp up to 100%

That way when you're stress testing, gaming, folding, whatever the fan will more or less be full bore, but when more or less idle, it'll be quiet.

Just a thought. Glad it seems you got your temps straightened out.
 
I'm using a fan profile like the one you described :)
It works but my rig is far from silent because of overclocked reference graphics cards..
This is a sort of temporary build, I'm saving up for some water cooling equipment to do a full custom loop or the CPU and GPUs.

Now I can keep the temps mostly under 80 on Prime95 when I use 1.12V voltage, but the problem is that voltage isn't available on the Adaptive Mode. Adaptive seems to have a minimum of 1.16V on turbo mode, and because of this I have to run manual volts if I want to go below 1.16, which I need to do for stress testing.

Is there a reliable stability test that does not produce such extreme heat as the P95?
I believe I could push the voltage up a bit if I didn't use P95 or IBT.
It seems kind of stupid to have to do stress tests on software that runs the CPU 30 degrees hotter than the apps I actually use.

Like for instance: P95 max temp is now 82 C when running small FFT for 70 minutes, and it's stable. Games, however, only see a max of 53 C.

So can I test stability on higher voltage without the extreme temps of P95 or ITB?
Or how would it be if I run P95 for two hours and temps go as high as 92 C? That would be only for stress testing, but
would it be safe to run the CPU so hot for two hours straight? In normal use with same volts it would be 30 C cooler.
 
Is there a reliable stability test that does not produce such extreme heat as the P95?
I believe I could push the voltage up a bit if I didn't use P95 or IBT.
It seems kind of stupid to have to do stress tests on software that runs the CPU 30 degrees hotter than the apps I actually use.
That would defeat the purpose really.

I would stick to P95 versus IBT or applications of the like...Normally, P95 only goes 10C over gaming type loads.. Weird. Perhaps the updated version beats the hell out of it worse than the version I have. I will have to test that out...
 
Prime 95 max temp is now 82 C when running small FFT for 70 minutes.

Sounds still a high temp for just 1,12V, does it not?
Since the temp difference between P95 and any normal app is still huge, I am wondering
if it would be safe to let the CPU run at 95 C just for the 1-2 hour stress test duration.
Because at such voltage, I could get higher clocks and games would still only run at 62-65 degrees C.
 
I guess I have found a solution to the question which stress test I should be using.

I did a test run on P95 2.79 and 2.81 respectively, both running the same amount of time on small FFT and I did a max temp logging run using CoreTemp.
Starting temps were controlled.

Prime95 v.2.79 max. heat 72 C
Prime95 v.2.81 max. heat 82 C

...there you have it.

I wondered how it is possible my CPU is running so hot compared to every one else's, but apparently it is not. Using the 2.79 version of P95, I can run the same voltages as other people using a mig range air cooler. How cool is that! (pun inteded)

Do you guys see any problems in using the v.2.79 to test stability on Haswell? At least this would allow me to get a little closer to real life temps.
And what would be a safe max temp on a 2-hour stability run? Since that heat would only be present under that very stress test...
 
...there you have it.
Yep...

Perhaps the updated version beats the hell out of it worse than the version I have. I will have to test that out...
Thanks for testing!

I would use the newest version as it is testing out more/all of the instruction sets. WHile part of the point is the heat, the other part is stability. If the older version was not using some instruction sets, hence the lower temps, then you could not be stable. Id stick with the latest version personally.
 
Back