SOLVED Perplexing temperatures with new(er) build

[THT]

It's been a very long time since I've built and overclocked a computer because my last machine carried me for so long, when I built my current one, it was like I was building my first computer all over again. My last machine was built about nine years ago and based on a Q6600 which I ran at 3.2GHz and a minor flood caused me to build a new machine. I have a rather odd temperature issue with this current build though. Before I go into it, here's what I'm using:

CPU: Intel i7-5820K @ 4.1GHz (125MHz BCLK and 1.25v)
Mobo: Asus X99-A USB 3.1
Memory: 4x 8GB Corsair DDR4 3000 (PC4 24000)
HSF: Noctua NH D15 with both 140mm fans installed

The issue I'm experiencing is abnormally high temperatures as reported by both HW Monitor and Asus' AI Suite 3. They both claim the CPU idle temp is around 40C with a motherboard temp of about 30C. When I run a small FFT torture test using Prime 95, my fans spin up as they should but my CPU temps hit 93C! The system is completely stable but that seems extremely high. Admittedly, I've been out of the computer building and overclocking game for quite some time but isn't 93C excessive and/or dangerous? Airflow is not an issue as I have tried this with the side of my case off and there's only a slight (1-3C) change in temperatures.

I pulled the HSF off tonight and reapplied Arctic Silver 5 hoping that I had screwed that up but nothing changed. I was under the impression that the Noctua HSF was one of the best air coolers available and that my overclock isn't extreme as most 5820K CPUs have no problem hitting 4.2GHz and higher with more voltage. What am I missing? Is there anything I should double check? Or am I wrong and this is normal for current CPUs?

 

[Robert17]

Do you have the latest MB BIOS? Otherwise, it sounds like regardless of removing the cooler and reTIMing you may still have not seaated it properly. I'd start there. If not, remove it all from the case and assemble it on top the MB box and allow yourself some elbow room to dig more easily.

 

[THT]

Yes, latest BIOS. I checked all of the mounting hardware when I pulled the HSF off to make sure it was installed properly and nothing was amiss. The Noctua mounting method makes it pretty idiotproof which was one of the selling points for me since I had been away from computer building for so long.

 

[EarthDog]

Do you need 1.25v for 4.1ghz? Try lowering the voltage as it seems to be too high for the clocks... I'd imagine 1.2V or less would be fine for 4.1ghz.

Confirm temps with realtemp for giggles...

What are your ambient temps?

 

[THT]

I haven't tried lowering the voltage yet. Realtemp showed similar (generally higher) temperatures than the other two programs. Ambient is 20C currently.

Googling a bit and I've uncovered a few people who have run into similar situations. One solved it by RMAing his CPU. But it looks like 93C is actually below the thermal limit for this CPU?

 

[EarthDog]

Lower the voutage to the least needed to be stable.

 

[THT]

Lower the voutage to the least needed to be stable.

I'm chasing voltage now. After more searching, it seems Asus boards provide a little extra voltage beyond what is set so my 1.25 was actually delivering 1.30 (confirmed by checking the voltage as reported by AI Suite 3). I've been ratcheting it down in .05 increments and running brief passes of Prime95 to check stability. I'm now on 1.00 and it's delivering 1.05. Temps stay at 65-66C with the case closed up and I'll let Prime95 run over night. If it's stable, I'll keep ratcheting it down until I find the minimum I need to run this speed. I did notice that my fans are no longer turning as fast either. They're running around 1,000-1,100 rpm whereas at 1.25, they were spinning up to 1,600 rpm.

 

[Tír na nÓg]

Mmmm... P95, last version on Haswell with 1.25v? Temps seem OK to me.

Try the 26.6 version of Prime.

FYI, the 5820K in sig hits [email protected]/1.26v with a H240X + 280mm rad.

 

[THT]

I think I have it solved. I was overvolting the bejesus out of it. It ran the small FFT Prime 95 torture test over night at 1.00 (effectively 1.05) and the temps never went past 66C. I tried bumping it to .975 and Prime wouldn't run for more than a minute before crashing so I'm probably at the minimum needed to run this speed. I probably have some decent headroom left if I wanted to push it more but my days of chasing every last MHz are behind me.

 

[JrClocker]

I was going to pipe in that your voltages may be too high.

When you are going through this, there are a few general relation to keep in the back of your mind:

1) CPU temperature scales directly with CPU power dissipation
--- the more power, the higher the temperature

2) Idle temperatures don't mean anything, it's ambient that is the important one
--- with normal air or water cooling, the LOWEST your CPU temperature can be is ambient
--- Your CPU temperature will be = ambient + CPU_POWER*Cooler_Efficiency
------ Cooler_Efficiency is measured in °C / Watt
--------- You will rarely get this specification from any CPU cooler manufacturer
--------- The CPU packaging has a thermal impedance (measured in °C / Watt) - you can lookup the socket style

3) CPU power scales linearly with CPU frequency
--- if you increase your frequency by 10%, your power dissipation will increase by 10%

4) CPU power scales as the square of CPU voltage
--- if you increase your CPU voltage by 10%, your power dissipation will increase by 21% (1.1 * 1.1 = 1.21)

For example, a if a 20% increase in frequency requires 15% more voltage, your CPU power will increase by approximately:
--- (1.2) * (1.15 * 1.15) = 1.587 = 58.7%

I say approximately, as the actual power depends on a lot of "stuff". But, if you have a baseline stability program that you are running, you will see these relationships hold true.