Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
That makes sense if you are using IBT without manually updating to the latest binaries that Intel releases on a regular basis...
My personal testing method is as follows:
1) Set desired CPU speed and attempt to boot into Windows. If successful go to Step 2. If not, add Vcore and try again..
2) I open IBT and using the "Extreme" preset, and all available ram, run 5 passes. If successful, go to Step 3. IF not, reboot and add Vcore and try again.
3) After 5 IBT runs, I go to 20, again using the extreme preset. (I also load my GPU to 100% to maximize temperatures in my loop). If successful, I go to Step 4. If not, well you know the drill by now.
4) After 20 IBT runs, I go into Event Viewer and make sure there are ZERO WHEA ERRORS* showing. If there is, I reboot and add Vcore. If not, I go to Step 5.
5) If everything above is good to go, I then go to a 24hr Prime 95 Torture Test, using Custom Preset..I set Prime to use all available threads, and every MB of memory. I then allow it to run for 24 hrs. If I can't get a stable 24hr run, I retweak and try again..
This! I don't consider any OC stable if it doesn't pass this kind of testing.
I would do as Dan suggests - use manual voltage to stress test. Later you could dial the same voltage with adaptive w/o the outlier tests.
The loop used for his rig must be insane.
On my H100 running full bore @ 1.255V I hit 100C instantly with IBT.
I dont do that much stress testing. I pretty much use OCCT for everything. I run a 4 hour session of OCCT and if it passes thats, its stable. I havent had a single BSOD or crash on an overclock that passes 4 hours of OCCT so for what I use my computer for, thats all the stress testing I need.
I'm at 4.5ghz at 1.32V ...I've been gaming pretty much all day on BF4 NFS Rivals etc during the holiday weekend I never crashed no problems...that's my stress test.
A good 4770K wouldn't need that much voltage for 4.5GHz. That kind of voltage would get you 4.7+ if you can keep it cool enough.
Damn. Looks like air isn't viable for this chip then. I'm at stock volts and it'll get too hot. But what does it say about prime (or Intel) when factory speeds and a factory cooler can't keep up..
Is that with fixed or variable voltage?
For shits and giggle I dropped my clock to 4.0 with 1.17v last night and ran P95 27.9. Temps spiked to 97c in about 15 minutes. There's no fucking way that software is properly loading the CPU.
Blowing up CPU by a game that uses similar power requirenments as prime. That's a worthwhile goal.
Am I the only one amused at the fact this 50+ post thread is basically a mirror of the thread two threads down in this same forum and was started 4 days earlier? I won't even start on the fact this isn't even in the overclocking and cooling forum and both threads are specifically about overclocking and heat issues, lol.
Thanks for that well thought out, constructive contribution to the thread. This solves all our issues!
For shits and giggle I dropped my clock to 4.0 with 1.17v last night and ran P95 27.9. Temps spiked to 97c in about 15 minutes. There's no fucking way that software is properly loading the CPU.
Yeah, okay. I tried Prime95 28.1 (small in-place FFT's) and my system failed in like a few seconds. I jacked up the voltage to 1.31 and it ran for about a minute and then bluescreened, but I looked at the sensors and I was honestly afraid of this monster. It managed to make my processor go to 99C like near instantly. What the heck? I was honestly surprised I wasn't smelling burning in there, that's ridiculous.
Apparently for everything I'm doing with my system, 1.305 is fine and my temperatures stay good. Heck I even did a handbrake conversion just now, went through just fine. I don't understand what kind of stability this tool is measuring... I don't think I could remain stable under any settings if this thing makes me go over 99C INSTANTLY at this overclock. o_0
Not everyone has the time (or shamelessness in a way) to go around cherry picking all of their processors until they get one that actually hits what you reviewers do. Unlike you guys, this stuff ain't my life, it's just a side hobby.
Yes, I know the CPU on the test bench is cherry picked. That was done for a reason. We didn't want a poor CPU limiting what the motherboards could do. We wanted to give each motherboard a chance to reach high levels of performance and try and see what if any difference the board made. So our CPU was tested at 4.8GHz before we got it. And actually, it's capable of more with more extreme cooling than what I'm using. And in any case, I didn't mean to suggest you should cherry pick your CPUs. I'm simply stating that your using a lot of voltage for relatively small gains. This isn't abnormal necessarily as a large percentage of these CPUs fall about where yours is. I'm just providing information.
And as I explained earlier, the problem with these CPUs is the IVR. The CPU decides what amount of voltage is needed for the current clock speed it's running at based on application demands. Newer editions of Prime95 are using AVX instructions and Unclewebb said. This is creating a load on the CPU which is unrealistic. It's synthetic. Granted if it's Prime stable it's stable, but don't expect to see insane overclocks on these CPUs. Most of them simply can't sustain 1.3v or more which is often required to break 4.5GHz or so. Even if the CPU is stable in the short term, if you can't dissipate the heat, heat soak will cause the temps to slowly creep up until failure.
You can use manual voltage and ensure the CPU can't use anymore power than you specified. The problem with that is that the CPU will never use less voltage than that even when it doesn't need it. So the heat soak you get will be worse, as is power consumption and general thermal conditions for the board and system. We test in this manner because it's easy and we are just going for broke to hit the maximum results achievable. The problem with Aida64 and some other benchmark utilities is that they don't create conditions in which the IVR behaves the way it would in actual applications. Prime95 is too harsh and Aida64 is too soft.
Actual applications, and how much voltage they'll cause the CPU to use varies wildly. You should use adaptive voltages for the best thermals, etc. but if you don't use manual voltage with Prime95, your temps will hit excessive thermals in mere seconds. The CPU may call for as much as 1.45v or so with the CPU clocked at 4.6GHz+ while running Prime95 as an example. That's more than Handbrake will use, more than your games will use and more than most anything will use. It's just how Prime95 loads the system. Intel's own Burn in Test is pretty bad about this too. Intel's IVR rules basically have a voltage curve relative to the clock speed it is currently running at and for stability sake it's pretty greedy about voltage and how much it wants. It likes a lot of voltage sometimes. The problem is, that this is bad from a thermal standpoint. To make matters worse these CPUs have a lot of variances in their IMCs. So one CPU can do 4.6GHz at 2400MHz memory and others will struggle to do anything over DDR3 1600MHz. Haswell is an interesting architecture but overclocking is now more luck of the draw than it has been in over a decade.
One final note, be careful with your voltages and thermals. We've seen CPUs that seem to degrade in a short time. They'll run with DDR3 1866MHz memory speeds or so, or hit 4.8GHz at first and then over time they'll do less and less. Our sample size isn't large enough to state this conclusively, but it's something that we've "noticed" on one or more CPUs. There are other variables possibly at play, so again your mileage may vary. I'd hate to see people roast their CPUs when they don't have the luxury of simply having a new one sent to them. So be careful. Remember, at 4.4GHz Haswell is roughly as fast as Ivy Bridge was at 4.6GHz and Sandy Bridge at 4.8GHz.
Thanks for the very detailed response. I did kind of take what you were saying the wrong way.
So, you think this CPU will slowly thermally degrade at 1.305 vcore? I know you said you don't have any real proof for it, but it's an interesting mention that I haven't heard of before. Most people on other sites are under the impression that under something like 1.35 Vcore is fine.
They didn't bother to read Intel's manuals. Actually from what I seen Asus engineers didn't either, or there is a typo in the Intel's manual.
That might explain why my g41 is capped at 1.30. One inaccuracy I found in that article is that the Intel Thermal compound is the reason to delid; while using a better compound helps, the real gains are from decreasing the distance from the cpu to the ihs by removing the glue.
Yes, I know the CPU on the test bench is cherry picked...
What volts, version of Prime, and how long are you testing?
With that version of Prime, I can do 4.5ghz under a 212+ for hours with low temps. It all depends on what stress test you choose. With Aida64, I had to turn it down to 4.2.
Thanks for the very detailed response. I did kind of take what you were saying the wrong way.
So, you think this CPU will slowly thermally degrade at 1.305 vcore? I know you said you don't have any real proof for it, but it's an interesting mention that I haven't heard of before. Most people on other sites are under the impression that under something like 1.35 Vcore is fine. If you're actually correct then it would be nice to share that information with some of these other places. And you are correct, I could probably lower down to 4.4Ghz without too much issue and go down to 1.2x VCore without too much of a performance hit. I was doing 4.2 on 1.15 earlier.
I've been doing LOTS of research. Good info here:Interesting. I want to delid at some point,,, my 4770k is crappy enough that I plan to practice on it once I find one that seems like a winner.
I would have guessed that the glue is causing a heat island effect of the air inside the space between. But your explanation makes more sense I think. I'm sure someone has the right answer and that's as plausible as any I've heard. I haven't tried hard to find out more about it, either. I am curious if temps around throttling (boiling water ish) has any effect on the glue itself.
I like the part where Intel forgot what "K" meant on the cpu model number myself. The average user isn't going to have the stuff you need to keep this cpu cool. Yes you can do 4.2 or 4.3GHZ fine But it boost to almost 4.0GHZ to begin with. Kinda defeats the purpose then. But it is what it is and Intel knows what they did.
My auto overclock set the board to 4.2 immediately @ something like 1.15V so if I'm going to bother OCing manually I would like to beat that by 400-500 MHz at least...
Yah there's a lot of good info in this one reply. Until like a few days ago I was under the impression that even if you had a manual voltage dialed in, there was a BIOS option for Haswell that would cause CPU states to include voltage degradation. I'm still not fully clear on this point, but a prog like HWMonitor (CPUID) will only display your Vcore going up and down like it should if you have an adaptive voltage set up for it. I'm guessing AI Suite (asus software) was throwing me off... it likes to display .000 voltage often and that's gotta be wrong.
I'm getting a little ill with Prime95 and the 4770k I'm trying to overclock for a customer.. Prime keeps spiking in temps be fine at 52c or so and then out of no where spike to 75c. or higher never had that happen with prime before. Always had it stay at a pretty constant temp with good cooling.. Prime95 is running hotter then Intel burn in test and I have never seen that happen before with prime95 .Running but in test right now and it's over 10 passes and not failing or getting over 72c.. Just makes no sense. Usally make sure a system is 24 hours of prime stable but now I have no Idea whats the best thing to check to make sure it's stable now..