i7 4790k - 86C too high?

[el_jefe]

Hey guys!

I've been putting together a Lightroom rig and went from 4771 and no plans to overclock to 4790k and a decent overclock within the last 10 days

I still don't plan to push the chip as far as it can go, because I need a rock solid production machine and I only need it to be overclocked for Lightroom anyways.

Currently I have a i7 4790k running at 4.7ghz on 2 cores and 4.6ghz on the other two. It took me 1 click and about 5 min. to OCed on Asus's Z97-PRO. The chip is cooled with a Noctua ND-D15. It's been running AIDA 64 for 6 hours now and it seems stable.

My question is about the temps: I am getting a max of 86C on one core, 85, 83 and 81 on the others. Isn't that kinda high for the NH-D15? There are three other fans in the case (a CM 690 II Advance): a NF-A14 front intake, NF-F12 top intake and NF-F12 rear exhaust and also no GPU btw...

Any thoughts?

 

[Techhog]

What are you doing when you get those temps? Also, what voltage are you running at?

Based on what you said and how fast you did it, I'm assuming you didn't touch voltages. In that case, the voltage was automatically set, which means that it's too high. If you're going to OC to such high clocks, you need to manually set the voltage and do stress testing. You're probably over 1.35V right now, which is NOT good for air cooling.

 

[el_jefe]

What are you doing when you get those temps? Also, what voltage are you running at?

I'm running AIDA64 stress test.

Based on what you said and how fast you did it, I'm assuming you didn't touch voltages. In that case, the voltage was automatically set, which means that it's too high.

Is that normal for a lot of onboard OC software? Because ASUS is very conservative with their automatic overclocks - they don't go over 1.275 V base.

If you're going to OC to such high clocks, you need to manually set the voltage and do stress testing. You're probably over 1.35V right now, which is NOT good for air cooling.

Is 4.6 - 4.7 ghz high? I thought people were getting 4.8 - 4.9ghz...

It's currently running at 1.307V according to CPU-Z while running the AIDA64 stress test. The temps haven't moved beyond the above mentioned maxes.

 

[Techhog]

Oh, during AIDA64... That might be fine, then. I also didn't consider the auto OC.

 

[Lepton87]

I don't know how much NH-D15 is better than D14 but I guess it's 4-5C at most which is actually very good. All high-end coolers clamped withing a few degrees north or south of D14. 86C is a good temp for HW but I don't know how well AIDA 64 stresses the CPU it would be better to test it with IBT making sure you have AVX enabled. Even 90C-94C is fine, you are not going to get the CPU anywhere near that temp during actual usage.

 

[Essence_of_War]

because I need a rock solid production machine and I only need it to be overclocked for Lightroom anyways.

These things are mutually incompatible.

OC'ing a production machine is just BEGGING for trouble. OC'ing a production machine that you NEED to be "rock solid" stable is looking Murphy in the eye and asking him to do his worst.

 

[el_jefe]

These things are mutually incompatible.

OC'ing a production machine is just BEGGING for trouble. OC'ing a production machine that you NEED to be "rock solid" stable is looking Murphy in the eye and asking him to do his worst.

Haha, you are most likely right I mean we have constant backups, RAID arrays etc. - the worst that would happen is a little running work lost, which could be annoying, granted, but not the end of the world in most cases.

Do you have experience with OCing production workstations? I am not very experienced at overclocking, but from what I understand if done correctly it should be stable 99% of the time.

 

[Essence_of_War]

Frankly, I think the worst that can happen is likely much worse than that.

Imagine what would happen if some poor overheated VRM blows on your MoBo and you're stuck until the RMA resolves. If LR is your work, you're talking about shelling our for a new mobo, or you're looking at lost work time = lost money.

The problem isn't that it won't be stable 99% of the time, the problem is how much annoyance that 1% is going to cost you in time/money/grey-hair.

 

[el_jefe]

Imagine what would happen if some poor overheated VRM blows on your MoBo and you're stuck until the RMA resolves. If LR is your work, you're talking about shelling our for a new mobo, or you're looking at lost work time = lost money.

Yeah, I know what you mean I have another machine though and while it would definitely be a huge pain in the behind if that happens it won't be the end of the world either.

That said, as soon as I am sure the system is as stable and as cool as it gets I'll test it OCed and stock on a couple of typical jobs I do to see how much time the OC can potentially save me. If it's anything less than 20% real production time saved, the OC is probably not worth it, all things considered.

...especially considering Adobe can easily speed up the damn thing 50% if they only did some simple optimization of that mess

 

[el_jefe]

...but also it seems both Intel and the mobo vendors encourage overclocking - what with all the overclocking parties, people using liquid nitrogen to cool their chips... I don't see who'd buy the Z97-Pro from Asus unless they were planing overclock their CPU. So I would expect them to have made the VRMs "sturdy" enough ;-) provided you use proper cooling etc. of course...

 

[Techhog]

Yeah, I know what you mean I have another machine though and while it would definitely be a huge pain in the behind if that happens it won't be the end of the world either.

That said, as soon as I am sure the system is as stable and as cool as it gets I'll test it OCed and stock on a couple of typical jobs I do to see how much time the OC can potentially save me. If it's anything less than 20% real production time saved, the OC is probably not worth it, all things considered.

...especially considering Adobe can easily speed up the damn thing 50% if they only did some simple optimization of that mess

Well, a 10% OC obviously won't increase performance by 20%. Assuming perfect scaling, you'd need a 5.1GHz OC to get 20% more performance, and scaling won't be perfect.

The 4790K is worth it over any other quad core even if you don't OC, though.

 

[BonzaiDuck]

I can agree with TechHog on at least two of his posts.

Another member here -- Z51CAM -- has a Sandy Bridge i7-2700K he says can be OC'd higher than 5Ghz with water-cooling. I don't doubt it; I just wouldn't push my 2600K's voltage as high as his. He maintains his load temperatures don't ride higher than 50C.

I've seen a couple "serious" efforts to OC the 4790K. One, with water-cooling, de-lidding (and I hope with replacement, as I'll explain ), use of CLU TIM and custom-water-cooling, got it to 4.7Ghz with core temp (assumed to be average across the cores) at around 69C. I think the voltage had been pushed to 1.4+V, and my "plus" indicates more than 10 mV to my vague recollection -- of which I'm trying to be forthright.

Without IHS replacement - a bare die application -- that temperature seems too high for a custom-water rig, especially with CLU application. You wonder what it would be without the delidding. And it would be even higher with IHS replacement.

I also think the voltage is way too high for any serious 24/7 use. It's a Haswell, and it has Haswell lithography. Even if the temperature is "acceptable" with custom water cooling, the voltage is not.

I think as Intel continues to reduce thermal wattage and the lithography, the overclocking potential will still be attenuated. This seems to be what happened with the 4770K: I saw sustainable OC results around 4.4 or 4.5.

The air-cooling even with a D15 is only likely to show worse results.

So I think that "after Sandy Bridge" and especially with this Haswell refresh, we should have lower expectations about overclocking.

As for overclocking machines used for "serious" work. Rule of thumb: don't do it. You can "take a walk on the wild side" I suppose -- if you can guarantee total stability, but then you'd have a modest OC target -- like 10%. But 10% of 4.4Ghz would put the target for 4790K at 4.8, which -- after what I've said already -- seems insane.

So with that, I'd take the 4790K at stock and call it a day. And I'd still use an NH-D15, just for the expense of the motherboard and the processor. But using the stock cooler (if the 4790K even HAS one) -- should still work.

Now someone else might say my speculations are bool-s***. Maybe they are -- maybe they aren't. But if the 4790K is any harbinger of a Haswell E, I'll have to reconsider whether that project is worth the trouble -- or if overclocking it is worth the trouble.