High load temps with Noctua NH-U14S

[SithSolo1]

Problem with new build.

Parts:
Define R5 with 2x intake, 1x outtake
Intel i7-4790k, stock clocks
Gigabyte GA-Z97X Gaming 7
Noctua U14S

At idle the core temps are in the mid-20Cs but as soon as any load is applied they jump up.

E.g.
Running windows update: upper 30s to low 40s
Gaming: low to mid 60s
Prime95 Large FFTs @ 1min: mid to upper 60s
Prime95 Small FFTs < 1min: low to mid 90s

- The fan is running and it does increase speed under load

- Both HWmonitor and RealTemp are reporting similar temps. I manually set the vCore to 1.2v and according to CPU-Z it didn't budge.

- I've re-mounted the heatsink 3 times. Twice with the included NH-T1 and once with Arctic MX-2. All temps were roughly the same and contact/spread of the paste appeared good. Between each mounting I cleaned the CPU and heatsink base with ArctiClean 1&2.

I plan to try the stock cooler later this week but I'm open to any input or suggestions.

 

[BonzaiDuck]

Problem with new build.

Parts:
Define R5 with 2x intake, 1x outtake
Intel i7-4790k, stock clocks
Gigabyte GA-Z97X Gaming 7
Noctua U14S

At idle the core temps are in the mid-20Cs but as soon as any load is applied they jump up.

E.g.
Running windows update: upper 30s to low 40s
Gaming: low to mid 60s
Prime95 Large FFTs @ 1min: mid to upper 60s
Prime95 Small FFTs < 1min: low to mid 90s

- The fan is running and it does increase speed under load

- Both HWmonitor and RealTemp are reporting similar temps. I manually set the vCore to 1.2v and according to CPU-Z it didn't budge.

- I've re-mounted the heatsink 3 times. Twice with the included NH-T1 and once with Arctic MX-2. All temps were roughly the same and contact/spread of the paste appeared good. Between each mounting I cleaned the CPU and heatsink base with ArctiClean 1&2.

I plan to try the stock cooler later this week but I'm open to any input or suggestions.

I would have recommended a different heatpipe cooler, although the U14S should perform about 2C warmer than a stock NH D14.

The gaming and Large FFT temperatures would seem about right.

Did you try something like OCCT's "CPU" test? The 4790K can be a bit toasty depending on the stress-test. But we'd seen 70C under stress-load with an AiO cooler that wouldn't perform much better than your U14S. And that was @ 4.6Ghz, while you're running at stock settings.

Maybe someone else has more insight to this . . .

In the meantime . . . You can trim a couple C off the peak temperature with IC Diamond paste TIM. You can trim a few more degrees by lapping all the nickel off your processor cap to bare copper and voiding your warranty. Ducting the cooler to the rear exhaust and using a fan slightly beefier than the stock should get you another 5C. The lapping: That's your decision if you even consider it.

IF you replace the U14S with an EVGA ACX cooler, you can trim another 7C to 8C off the total.

Also, I would investigate how your motherboard is volting the processor. I think we'd seen how these DC/Haswell-refresh chips will run stable at lower voltage settings, and the auto settings are probably putting the VCORE too high. I've seen many cases where Haswell users reduced their load voltage, kept the same clock speed, and lowered the temperatures.

 

[Essence_of_War]

In the meantime . . . You can trim a couple C off the peak temperature with IC Diamond paste TIM. You can trim a few more degrees by lapping all the nickel off your processor cap to bare copper and voiding your warranty. Ducting the cooler to the rear exhaust and using a fan slightly beefier than the stock should get you another 5C. The lapping: That's your decision if you even consider it.

IF you replace the U14S with an EVGA ACX cooler, you can trim another 7C to 8C off the total.

I think all of this is irrelevant. The OP isn't asking for recommendations on how to lower their temps, they think there is a problem with their temps.

OP,
Could you try changing to stock voltage configuration to go along with your stock clocks, and see how that does for a reference point? 1.2V seems quite reasonable, but it might be nice to be sure.

 

[SithSolo1]

I'm not sure what the stock voltage should be. Reading around on the net I saw a lot of references to 1.2v so I set it to that. The auto setting had it at 1.228v

I won't be back to my brother's place until Friday but I'll be sure to try and see if I can lower the vcore more.

 

[Essence_of_War]

The auto setting had it at 1.228v

Was that under small FFT load?

Another thing to try would be OCCT's stress test, and Intel's Extreme Tuning Utility stress test, to see if those give comparable temps to p95 small FFT.

 

[SithSolo1]

Yeah that was under small FFT load. I'll add those other stress test to my to do list.

 

[Essence_of_War]

One more thing to note for Haswell, recent versions of prime95 add AVX optimizations for small fft that add +100mV and I don't know if that shows up in vcore measurements of CPU-Z.

I would check it myself but I don't have a Haswell chip

My guess is that this is what you're seeing if you're running a recent version of p95, Vcore of 1.2 that is in reality 1.3! That is enough to get an i7-4790k mighty toasty. It is certainly consistent with your very reasonable large fft temps and crazy high small fft temps. I don't believe OCCT's stress test or Intel Extreme Tuning Utility's stress test use AVX, so that would be a good cross check.

 

[Ketchup]

Those temps look pretty good. A few things to consider:
1. On auto, the voltage will fluctuate above and below "stock" as the CPU speed ranges from idle to load. I know on mine that the auto voltage setting was extremely close the the voltage needed at 4.5 when I set it to manual, for example.
2. The new Prime95 will push your CPU to higher temps than any other software out there (including games).
3. My chip starts hitting the 100 C throttle-start between the 4.4 Turbo frequency and 4.5 Ghz, so at least so you have overclocking headroom if you wanted it.

 

[SithSolo1]

Well for kicks I ran the Intel Extreme Tuning Utility for 5mins on my current system.

stock i7-4770k and Noctua NH-C12(larger aftermarket fan) and temps were bouncing between 58C and 63C which is about 10C lower than P95 for the same period. I also noticed XTU ran at 3.7ghz but P95 ran at 3.5ghz. Could that be AVX related?

Apples and Oranges I guess but I just assumed a newer, larger U14 would perform similarly on the 4790k. Once I run XTU on the U14 system I'll have better idea of what the temp gap is.

 

[Essence_of_War]

If you're at stock clocks, 3.5 GHz is expected for a 4+ core load on the 4770k. I'm somewhat surprised that XTU ran at 3.7!

If you were running a recent version of P95:

28.5 - May 30, 2014 - AVX2 and FMA3 support for Haswell-based CPUs (Core i3/i5/i7-4xxx models), additional memory caching optimizations for Sandy Bridge and Ivy Bridge-based CPUs.

Then I think The +10C can be chalked up to AVX.

 

[aigomorla]

can I ask what fans you are running?

also can you tell me at what rpm the fans are running when the system is at load?

it maybe possible that all you need to do is replace the fans with better quality ones that provide more static pressure.

Not all fans are made equal.... and noctuna stock fans are only good for low speed / quiet operation.

 

[Essence_of_War]

it maybe possible that all you need to do is replace the fans with better quality ones that provide more static pressure.

Not all fans are made equal.... and noctuna stock fans are only good for low speed / quiet operation.

Based on reviews of the NH-U14S that seems unlikely to be the issue, assuming the fans are actually working correctly.

 

[SithSolo1]

If you're at stock clocks, 3.5 GHz is expected for a 4+ core load on the 4770k. I'm somewhat surprised that XTU ran at 3.7!

That was my thinking, do I need to configure it to run 8 threads? I hadn't used it before so I just flipped to the Benchmark section and clicked Start on the 5min run.

can I ask what fans you are running?

also can you tell me at what rpm the fans are running when the system is at load?

The U14s rig is running the single stock NF-A15. If I have to switch out the fan I'd like to keep it on the quiet side.

 

[Racan]

If you're at stock clocks, 3.5 GHz is expected for a 4+ core load on the 4770k. I'm somewhat surprised that XTU ran at 3.7!

As long as temps stay under 100 C, 3.7 Ghz for a 4+ core load is the normal turbo boost frequency.

http://www.intel.com/support/processors/corei7/sb/CS-032279.htm

 

[Essence_of_War]

As long as temps stay under 100 C, 3.7 Ghz for a 4+ core load is the normal turbo boost frequency.

You're absolutely right, I mis-read the specs!

That's actually useful too. AVX2 loads on non-devil's canyon haswell with the lower quality TIM caused some thermal throttling even at stock clocks, and it looks like SithSolo1 is seeing that on their home rig.

 

[SithSolo1]

You're absolutely right, I mis-read the specs!

That's actually useful too. AVX2 loads on non-devil's canyon haswell with the lower quality TIM caused some thermal throttling even at stock clocks, and it looks like SithSolo1 is seeing that on their home rig.

vCore according to CPUZ also dropped on my home rig between the two. It was 1.063 in UTX and 1.053 in P95.

 

[BonzaiDuck]

The heatsink should at least be adequate. It has to be the AVX2 instruction-set extension used by a new Prime95 version. A fan could make a few degrees difference, but there's no other explanation.

I don't know how it can be turned off within the program, so that would make other stress-tests desired substitutions if that feature can be turned off.

Supposedly the TIM is "improved" and made of some polymer. And they were supposed to have fixed any gap between the IHS and die.

 

[Jareb]

As far as your Noctua is set right your cooling setup is more than enough - it's totally 4790k issue and nothing you can do about it's temperatures except delidding the CPU. I'm in the same boat as you with almost the same configuration in r5 case.

 

[xthetenth]

Delidding a 4790k is a lot of risk for very little reward.

 

[SithSolo1]

I went and fiddled with it at lunch today. Lowered the vCore some more and changed the fan profile so that it ramps up a lot quicker. I think in the default mode it wasn't hitting 100% even at 100C, not it hits 100% speed at 66C. Now it sits around 61C under load in Intel XTU. Still seems a little warm but so far its been completely stable so I'm not going to mess with it more.

 

[BonzaiDuck]

I went and fiddled with it at lunch today. Lowered the vCore some more and changed the fan profile so that it ramps up a lot quicker. I think in the default mode it wasn't hitting 100% even at 100C, not it hits 100% speed at 66C. Now it sits around 61C under load in Intel XTU. Still seems a little warm but so far its been completely stable so I'm not going to mess with it more.

With me old SB-K's, I'm always playin' catch-up with these new developments. So I'm going to download the XTU program and see what it does with my Gen-2's.

I would say "a little warm" with load and 61C is a bit harsh. I'd think we've now got a better handle on these Haswells and Devils-Canyon chips. For that cooler and a load I will evaluate shortly (XTU), that actually might give a little headroom in case you want to overclock. But 61C for such a processor under a reasonable load test isn't bad at all.

 

[buklau]

Your temperature sounds about right to me. My temperature were even higher than yours at 4.5ghz 1.25v before delid on a hyper 212+ evo lapped MX2. After I delid my chip and used CLU, now my temperature is in the following at 4.7ghz 1.337v.

Idle = 25-27c
Gaming = 55-65 depending on the game
Prime95 28.5 Large FFT = 65-70c
Prime95 28.5 Small FFT = 84-90c

Before delid with MX2 1.25v at 4.5ghz

Idle = 27-30c
Gaming = 63-72 depending on the game
Prime95 28.5 Large FFT = 80-87c
Prime95 28.5 Small FFT = 95-100+c and then instant shut down at 105c.

 

[BonzaiDuck]

Your temperature sounds about right to me. My temperature were even higher than yours at 4.5ghz 1.25v before delid on a hyper 212+ evo lapped MX2. After I delid my chip and used CLU, now my temperature is in the following at 4.7ghz 1.337v.

Idle = 25-27c
Gaming = 55-65 depending on the game
Prime95 28.5 Large FFT = 65-70c
Prime95 28.5 Small FFT = 84-90c

Before delid with MX2 1.25v at 4.5ghz

Idle = 27-30c
Gaming = 63-72 depending on the game
Prime95 28.5 Large FFT = 80-87c
Prime95 28.5 Small FFT = 95-100+c and then instant shut down at 105c.

Well . . . it serves to prove the point about the AVX2 instruction-set in late versions of stress-testers. That leaves open the question as to whether one could get near-same or similar results without de-lidding. If you could show a 10C improvement in cooling on "air" -- that might also do the trick.

UPDATE in response to SithSolo1:

I went and fiddled with it at lunch today. Lowered the vCore some more and changed the fan profile so that it ramps up a lot quicker. I think in the default mode it wasn't hitting 100% even at 100C, not it hits 100% speed at 66C. Now it sits around 61C under load in Intel XTU. Still seems a little warm but so far its been completely stable so I'm not going to mess with it more.

So I tried XTU. You'd think that if anyone could design an appropriate bench-stress-monitor program, it would be the chip-maker. I think it's pretty mild on temperatures: mine would cycle between 63 and 67C. Keep in mind this is with ~140W "package power" or best bet about OC'd thermal power -- stock TDP is 95W if I recall. On the 4790K, it is 88W. But I don't think the DC's discussed here are churning out anywhere near 140W.

Given my cooling mods and other "grain-of-rice" choices, these temperatures at 4.7Ghz under ~1.35V room_F = 77F are milder -- lower -- than IBT, Prime95_28.5, OCCT:Linpack, or affinitized LinX. Just guesstimating, I'd say XTU temperatures on my machine match those for OCCT:CPU.

Every stress-test program I've seen appears to have its own pattern or a "cycle" of voltage, temperatures -- even CPU-usage -- from iteration to iteration. Intel's program doesn't seem to show "iterations" or "runs" appearing as successive lines of "reports" shown in other programs like LinX and IBT. A 5-minute pass at XTU seems to have 2-second cycles without reporting anything. This shows up as temperature fluctuations and changes in fan speed -- of what it is possible to hear.