Help Overclocking i7 5820k (AIDA64 test)

[Fraga500]

Hello guys, I'm fairly new to overclocking. I decided to follow instructions from a Linus' Tech Tips (I'm suchh a noob lol) overclocking video since his Mobo and CPU are identical to mine. I got an i7 5820k and a X99-A II, with a simple Hyper Evo 212. When he runs the AIDA64 stability test, he mentions that, if the test fails before 10 minutes, I should lower the multiplier OR, if my CPU low temperatures are in the comfort zone (80°C), turn up the CPU voltage a little bit.

The problem is: my test passes, but CPU temperatures go almost to a 100°C sometimes!

How should I proceed? Am I safe, since the test passed? Or should I do something else, like lowering the multiplier, as if the test hadn't passed?

Edit: video link: https://m.youtube.com/watch?v=Nj4ixZE7EWo

I'm "stuck" at 7:10.

 

[rhysiam]

Nope, 100 degrees is far too hot. Either the cooler isn't mounted properly, or you've hit your thermal limit.

What voltage are you running at?

 

[Fraga500]

Thanks for the quick reply, rhysiam.

Was running at 1.3 but now switched to 1.27. Should I lower the voltage even more? Ratio is at 34 (125MHz).
Should I keep testing? Is it safe to keep reaching 92°C peaks (CPU Core 3) when running the test for 15 minutes?
What should I tinker with to lower these peaks?
Thanks

 

[rhysiam]

Thanks for the quick reply, rhysiam.

Was running at 1.3 but now switched to 1.27. Should I lower the voltage even more? Ratio is at 34 (125MHz).
Should I keep testing? Is it safe to keep reaching 92°C peaks (CPU Core 3) when running the test for 15 minutes?
What should I tinker with to lower these peaks?
Thanks

92C is still too high really. It is a stress test, so approaching worst-case scenario, but that's still too high.

In terms of addressing it, do you understand what voltage does?

Obviously higher frequencies mean the processor is running faster = better performance. So you want your frequency as high as possible.
BUT, once the frequency gets too high, the transistors in the processor can't change state quickly enough and the system becomes unstable (tests fail, bluescreen/crashes/lockups, etc).
SO, if you want that frequency to be stable, you need to raise the voltage. Raising voltage effectively pulls the transistors to the correct state faster, making it possible to keep a stable system at higher frequencies.

The problem is that raising the voltage increases heat substantially. This is often misunderstood. Raising or lowering the frequency has a small impact on temperatures, but it's the voltage that has by far the bigger impact.

Your 5820K is a quite a power hungry (=hot) CPU, and you have a fairly entry level cooling solution. It's not surprising you're hitting thermal limits.

So back to your situation... you've already gone past the point Linus is talking about in his video. The CPU is too hot, which means the voltage is already too high.

Here's what I suggest, just do the procedure backwards...
What you need to do is lower the voltage and keep testing. Try ~1.22. Hopefully that will put your temps back into the safe zone. BUT, you need to see if your current frequency is still stable at the low voltage.
If NOT - you can either raise the voltage a little again (if you have some headroom on temperature, as in you're back under 80C), or you'll need to lower the frequency until it's stable.

The end goal you want is the highest possible frequency, with the lowest possible voltage (while still being stable), with temps under control.
And, just to reiterate, the two key levers you have to work with are:
Frequency gives you speed, but higher frequencies become unstable without higher voltages
Voltage increases stability, making it possible to hit higher frequencies, but at the cost of additional power and thus higher temperatures.

 

[Fraga500]

UPDATE: lowered ratio to 34 (125) - approx. 4.25GHz, and voltage to 1.23. Temps peak at 80-82 during stability test.
Is this too much? Do i need a better cooler?

 

[rhysiam]

UPDATE: lowered ratio to 34 (125) - approx. 4.25GHz, and voltage to 1.23. Temps peak at 80-82 during stability test.
Is this too much?

That's okay, but still kind of high-ish.

If you've tested thoroughly with AIDA64 and the system is still nice and stable with 4.25Ghz @ 1.23V, my question would be... can you get the voltage even lower and still maintain stability?
If you're prepared to put a bit more time in, keep dropping the voltage until you find the point where things get unstable. And then go a tick or two higher again. If you can get it down to or below 1.2V and everything's still stable, then your current cooling solution should be fine and you'll probably see temps under 75C most of the time.

Do i need a better cooler?

Well, a better cooler will drop temps on your current OC, and also allow you to run higher voltages without the temps getting out of control. If 4.25Ghz is stable at 1.23V (you need to run a long stress-test to confirm this), then you have a good shot of hitting 4.5Ghz to 4.6Ghz at or a little over 1.3V. Better cooling could handle that voltage making that (probably) a usable 24/7 OC. But in the end you're getting around 5%-8% more performance. So maybe not worth it.

 

[Fraga500]

UPDATE: lowered ratio to 34 (125) - approx. 4.25GHz, and voltage to 1.23. Temps peak at 80-82 during stability test.
Is this too much?

That's okay, but still kind of high-ish.

If you've tested thoroughly with AIDA64 and the system is still nice and stable with 4.25Ghz @ 1.23V, my question would be... can you get the voltage even lower and still maintain stability?
If you're prepared to put a bit more time in, keep dropping the voltage until you find the point where things get unstable. And then go a tick or two higher again. If you can get it down to or below 1.2V and everything's still stable, then your current cooling solution should be fine and you'll probably see temps under 75C most of the time.

Do i need a better cooler?

Well, a better cooler will drop temps on your current OC, and also allow you to run higher voltages without the temps getting out of control. If 4.25Ghz is stable at 1.23V (you need to run a long stress-test to confirm this), then you have a good shot of hitting 4.5Ghz to 4.6Ghz at or a little over 1.3V. Better cooling could handle that voltage making that (probably) a usable 24/7 OC. But in the end you're getting around 5%-8% more performance. So maybe not worth it.

Thanks for the reply!
I thought the temp values were pretty high so I lowered the voltage even further. I want to keep it safe while my cooling is not optimal. I also live in a very hot country (Brazil) so my PC temperatures tend to get even higher.

Right now Ratio is still 34 at 125MHz (4.25GHz), but voltage is 1.18. What do you think? Should I try increasing ratio while keeping this voltage?
And what are the cons of lowering the voltage? The PC could eventually crash since it wouldn't hold the ratio?

 

[fedpul]

Hi Fraga500, I suggest you to test stability with prime95 for 24 hours to be sure at what voltage you are stable. If you are not OCing your memory I suggest you run In place FFTs in a custom blend test 8-4096K (15 min per FFT) it will took around 21 hs to complete all the FFTs. If you pass without errors after 21 hs you are 100% stable. Also have in mind that if you set priority to 10 your computer will be unusable for the time you run the test, first you should monitor the temps for 15 min running the torture test Small FFTs at priority 1 to check temps. If everything is OK with temps then you can run the 24 hs test.

 

[rhysiam]

Hi Fraga500, I suggest you to test stability with prime95 for 24 hours to be sure at what voltage you are stable. If you are not OCing your memory I suggest you run In place FFTs in a custom blend test 8-4096K (15 min per FFT) it will took around 21 hs to complete all the FFTs. If you pass without errors after 21 hs you are 100% stable. Also have in mind that if you set priority to 10 your computer will be unusable for the time you run the test, first you should monitor the temps for 15 min running the torture test Small FFTs at priority 1 to check temps. If everything is OK with temps then you can run the 24 hs test.

Just bear in mind that you have Haswell CPU with AVX 2.0 feature set, current versions of Prime95 are going to generate massive thermal loads on that CPU. Some people would argue that to be "stable" an OC must maintain stability and keep temps in check under any workload, however we've seen consistently for years now that loading up AVX 2.0 workloads generate heat far beyond any "normal" use-case.
For that reason I personally would NOT suggest using Prime 95 version 28.5 or later for stress testing a Haswell OC.
Others argue differently.
But I can tell you that if you're pushing high temps under AIDA, Prime95 using AVX will hit much higher temps.

Right now Ratio is still 34 at 125MHz (4.25GHz), but voltage is 1.18. What do you think? Should I try increasing ratio while keeping this voltage?
And what are the cons of lowering the voltage? The PC could eventually crash since it wouldn't hold the ratio?

It depends how much time you're willing to spend. The only con of lowering the voltage is potential stability issues. When you start failing tests/crashing/freezing, it's a good sign that the voltage is too low for the given frequency. You won't cause any damage with this, in fact lowering the voltage actually increases component life, so if you have the time and patience, it's worth seeing how low you can get the voltage.

Personally, I like to know where the limit is, and then go a few ticks higher (not just one or two), to give me a nice stable OC. If I were you I'd keep dropping it and stress testing until I get a crash, and then go back up a bit.