Noctua NH-D14 very high temprature

[paul878]

I am getting very high temperature overclocking with the nh-d14, I was expecting a lot lower with this huge cooler. It hit 80c degree in Realtemp at 4.7ghz with 1.35v running prime95. My room is 77f degree, case is open and the fans are at full blast. I have remounted this Noctua 3 time, is this typical temperature for the D14? At stock it was 55c running prime95.

asus gene-z
2600k
nh-d14

 

[Madmanden81]

That sounds a bit high... Did you try reseating it? How much paste did you use (and what method)?

 

[BonzaiDuck]

I am getting very high temperature overclocking with the nh-d14, I was expecting a lot lower with this huge cooler. It hit 80c degree in Realtemp at 4.7ghz with 1.35v running prime95. My room is 77f degree, case is open and the fans are at full blast. I have remounted this Noctua 3 time, is this typical temperature for the D14? At stock it was 55c running prime95.

asus gene-z
2600k
nh-d14

Responding equally to Madman, or inviting a three-way discussion as I'm an eavesdropping interloper.

I KNOW I installed my D14 correctly. I'm not "at 4.77" yet. Plenty of time for that. But in the 4.4+ to 4.5 range, I think my voltage was still close to what you show. My PRIME 95 for the current 4.34 setting peaks at 69C. At nearer 4.5, 71C. I'm not sure your 80C (and we're operating at near-identical room-ambient) may not be that far off.

What remains to be sorted out in this and other threads: Whether there's a difference between i7 SB TCASE and TJunction temps (and I assume your monitor is showing you 80C for the latter. Until I MYSELF know that there is a dependable spread between TCASE and the cores, I won't allow my cores to go above 75C. Maybe 80C, I haven't decided. The throttling temperature for the SB K chips is about 72.6C.

Now . . . . the stock throttling temperature may not be the spec where we'd expect to see significant degradation over time to the CPU. Then again, it may indicate the beginning of a "range" where bad stuff slowly, slowly begins to work its damage.

But the main thing to remember may be this. You can OC it to that level, then test it, and if it's stable at 80C with voltages such as you and I both see, it will seldom run in normal or gaming usage to stress the cores that much with heat.

Depending on the motherboard, you can either disable the throttling C1E feature, or you can actually "tune" it so that the throttling doesn't cut in until a higher temperature.

Getting back to your observation and the Noctua, I can only say that there is a convexity in that cooler's base, so it is not making full contact with the entire area of the Sandy's IHS. And I thought I read something the other day about some newer approach Intel had chosen that has something to do with the sensors, the throttling, the temperatures, and the temperature of the entire IHS or some level of thermal energy being absorbed by that entire base.

I'm just not so sure there's any problem with the Noctua unless for the fact that its base is designed to be convex. I suppose I'll have to see what my system does at 4.7 to offer any further help. But I'm quite sure the Noctua and Prolima-Megahalem are giving you the same range of thermal resistances and cooling performance.

 

[john3850]

Try holding any fan by your open case to see if there is a drop in tempertures
You need to list your mb or hd temps also.
Could you be runnig a viedo card that doesnt expel the hot air out of case.

Next TIM use the thinest coating of TIM on the IHS.
Then place a drop TIM on your hs base and Rub it in using a plastic bag over your finger and let it dry.
Then using a soft rag and very lightly pressure rub most of the tim off only leaving a slight haze left on base.

 

[paul878]

I have reseated it many time and I have try various method of applying the thermal compound. I also have try different thermal compound, the results are the same. I do have a fan blow at the open case and the motherboard is at 34c degree.

According to the Asus utility I am drawing 100amp at 1.35volt, that give me 135 watt. Base on all the review I have read about the Noctua, I was expecting lower temperature.

BonzaiDuck, I would love to have more data from you.

1. do I have a hot sandy?
2. nh-d14 is defected?

 

[cubeless]

something's way wrong... are you using the fan resistors that come with it? my 930 with ht on @ 4ghz 1.28v runs under 70c doing8 p95 threads... that's fans plugged directly into headers but still running quietly in an antec 300 with all the case fans on medium and the house @ 78f...

put a straightedge across the hsf base and the chip... it's possible the the 14 is nfg, but how can you ever tell if a heatpipe is broken... i did have some samurai hsf's that just got less effective over the course of a couple years... maybe the sauce in the tubes does work its way out some times...

 

[BonzaiDuck]

I have reseated it many time and I have try various method of applying the thermal compound. I also have try different thermal compound, the results are the same. I do have a fan blow at the open case and the motherboard is at 34c degree.

According to the Asus utility I am drawing 100amp at 1.35volt, that give me 135 watt. Base on all the review I have read about the Noctua, I was expecting lower temperature.

BonzaiDuck, I would love to have more data from you.

1. do I have a hot sandy?
2. nh-d14 is defected?

More revelations are falling on us all, so the data sharing here is valuable.

You say "ASUS Gen[e] Z" -- I know there is a Maximus board that also uses the Z68, but I'm guessing yours is the P8Z68[whatever -- all mostly the same]?

DO THIS: Instead of RealTemp, download the very latest HWMonitor off the CPUID web-site. I THINK the version is 1.8. Raise THAT software before you start your stress-tester.

RealTemp gave me a max-"highest-core" the other day of around 69C with IBT, High-Stress. Today I'm running Maximum Stress, same room-ambient, and instead of RealTemp -- HWMonitor (which reports voltage info consistent with ASUS' monitor). Suddenly, my highest temperature of the cores is 62C. That's with HT "OFF." Also, sad to say, I'm using VISTA-64 and you can check another thread for similar conversations on this topic (I just posted) -- Win 7 has an instruction-set enhancement for which IBT responds with higher thermals.

All this information is good. Before you fret over the temperatures at the higher clock-speeds, you'd want to know if the sw-monitoring can be trusted. Or whether your OS permits IBT to do its scorchiest.

Put this back for near-term or future reference, though. At this point, I rather doubt your Noctua is defective, and I even doubt (though have no way of knowing for sure at this point) that you "ate a hot one" with your 2600K purchase.

Again -- I can only say what I do know (as opposed to faith and belief):

The coolers from TR and Noctua --maybe Prolimatech but can't remember -- all use a convex-base design. They're also nickel-plated. Some here have said that grinding the base flat may "reduce effectiveness," while TR says the convexity "serves efficiency." But we'd been dealing with these coolers since the TRUE phenomenon in 2007 July. IT had a convex base. I've only seen improvement from two absolutely flat surfaces -- that's observation #1.
Observation #2: The nickel-plating on both the cooler and the processor has a higher thermal resistance than the copper it supposedly protects. See some other posts I've made on this, the reason they do the plating, etc. But (the observation . . ) grinding the nickel off each surface is worth a 3 to 5C improvement (per surface!). That is -- such is what I'm sure I observed two and three years ago.

Observation #3: Here and elsewhere, nobody seems convinced about choices of TIM when I or others say "Get the nano-diamond paste, or make your own." JOe Citarella at Overclockers.net has been researching and experimenting with this stuff per years. Did testing back in 07 with a $22,000 calorimeter setup, posted his "disclaimer" that he was going into business -- an outfit called "Innovation Cooling." Well, find out who makes "IC Diamond" thermal paste. In his case, his self-interest has nuthin' to do with the honesty of his promotion -- which is all based on thorough empirics, testing and fact. The tests prove it. I ran tests, with bar-graphs and circles and arrows on the backs of each one, and I proved it. It may only be worth 2C -- maybe 3C improvement, but it's a big grain of rice if all these things are additive.

HeatsinkFactory.com sells the IC-Diamond 7-carat tube for about $7, as does SidewinderComputers.com out of Chicago. Maybe that "liquid-metal" stuff is also good, but I think it's also conductive. Nothing about those diamond particles or their oil base will conduct anything. And diamond has about 10-times the thermal conductivity of Silver, and more over comparison with copper.

Also -- back to the ASUS board. Check again the setting that can add amperage to the processor for better OC'ing. If "Auto," try 10% instead, then either OC-Tuner or however -- get the "TPU" to reset the OC setting, but also -- only after some other stuff:

Get the spec voltage on your RAM, go to BIOS and set it to that voltage, leave the timings to auto or fix them at the RAM spec. Also, there's some good advice out there about how to sit the VCCIO or CPU_VTT, PLL and other voltages to safe values. Fix those. And try the ASUS setting for "mild power saving" on the DIGI+VRM settings.

Then see what it gives you for a speed close to or under the 4.7 you're attempting here. See what the VCORE (left on "Auto") is. You can tweak all this stuff and fine-tune later, but this board (Z68 from ASUS) seems really good for finding you a stable setting from which to start with that.

Just try it, the HWMonitor, etc and see what kind of thermals you get.

Also, as much as everyone is seeking that 5.0Ghz pie-in-the-sky, some of the hot-dawgs in attempting to get there are volting their CPU's to 1.45V just to START! For me, I'll recycle the "In Living Color" quote from Homey da Clown: "Homey don' play dat! Homey don' play!"

 

[Ben90]

DP

 

[Ben90]

DO THIS: Instead of RealTemp, download the very latest HWMonitor off the CPUID web-site.

RealTemp is hands down the best for Intel processors. You can't get more accurate than looking at the Distance to TJmax directly.

 

[BonzaiDuck]

RealTemp is hands down the best for Intel processors. You can't get more accurate than looking at the Distance to TJmax directly.

I would've thought so. But when I see monitoring software reporting even one item incorrectly, I have to wonder if they don't need another revision.

RealTemp CoreTemp et al had last revisions released months, sometimes many months, before the Z68 chipset. You could argue that P67 preceded Z68, and that "P67 begat Z68."

But I've seen this before -- lags in the software as new hardware is released. In fact, I'd seen it with RealTemp and CoreTemp as pertains to the actual TJunction limit of a mainstream Intel processor, and we were getting incorrect temperatures reported as a result until Intel posted the spec as much as a year after the processor was released.

So all my RealTemp and CoreTemp temperatures reported before the software was revised were over-stated and incorrect. there are a few things I would've done differently had I known the true temperatures when I built the rig.

Look at it this way ASUS or whomever decides to release a new motherboard. All this time, they're working on a software suite of their own, specifically for the motherboard. They should KNOW if it reports temperatures and voltages correctly.

Now you compare idle and load VCORE readings of the mobo software to what's reported by CoreTemp or RealTemp or SpeedFan -- whatever, and you see that the latter -- EVEN for being consistent within themselves -- differ from the mobo software. Ghost Busters! Who ya gonna call? Who ya gonna believe? CoreTemp and REalTemp and SpeedFan can be revised -- but they have a quadra-zillion motherboards and chipsets to keep up with. ASUS or Gigabyte or ASRock or Intel? Each with a handful of their own mobos. It's just that they may not choose to report core-for-core temperatures, choosing to show either the "highest" or something called a "package" temperature which looks like / insinuates something reported from the TCASE sensor. Except that it seems higher by a few degrees than the cores when you actually get them recorded correctly, or when you actually THINK they're reported correctly.

Suddenly, CPUID's HW Monitor is released in version 1.8. When CPU-Z (also theirs . . but earlier revision) shows one voltage that differs from the monitored value of the mobo-maker's software, and Core Temp and RealTemp show the same voltage as CPU-Z which seems to behave "aberrant," you start doing other comparisons and suspending belief in CT, RT or CPU-Z. And then you run that 1.8 revision fo HW Monitor, which shows voltages identical to the mobo-maker's software and it shows four core values and a higher "package" value with the package temperature identical to the CPU reading in the mobo-maker's software. Even so, it may be that all four core readings and the "package" reading seem lower than the CT or RT software revisions that are three to six or eight months older than the Z68 chipset.

Who you gonna believe until the RT, CT and CPU-Z software gets a revision -- which in-freakin'-dubitably will reflect a motherboard chipset that seems to be growing in popularity?

So that leaves us with another recent news-item or forum-rumor that looks like it comes from someone who "knows something." Intel changed something about the relationship between the core sensors and temperatures, the TCASE (or "package"?) sensor and temperature, and the IHS heat capacity. Something involving a delay or lag that wasn't there in the older chip families.

This could mean that this "package" temperature I spoke of would actually be higher than the core values, even if equivalent to the TCASE temperature of yore. Whatever it is, it may be the basis for the throttling spec temperature -- for the K Sandys -- 72.6C. If watching the core temperatures in our OC'ing efforts was standard geek practice one, two -- three years ago because we could assume those temperatures were higher than the TCASE, then maybe we should be watching this "package" temperature now and instead.

 

[ICD7]

The coolers from TR and Noctua --maybe Prolimatech but can't remember -- all use a convex-base design. They're also nickel-plated. Some here have said that grinding the base flat may "reduce effectiveness," while TR says the convexity "serves efficiency." But we'd been dealing with these coolers since the TRUE phenomenon in 2007 July. IT had a convex base. I've only seen improvement from two absolutely flat surfaces -- that's observation #1.
Observation #2: The nickel-plating on both the cooler and the processor has a higher thermal resistance than the copper it supposedly protects. See some other posts I've made on this, the reason they do the plating, etc. But (the observation . . ) grinding the nickel off each surface is worth a 3 to 5C improvement (per surface!). That is -- such is what I'm sure I observed two and three years ago.

Have been looking extensively at the bowed bases recently. below a link to a review where the reviewer lapped the base and combined with paste had an overall improvement of 5C - also have another guy on a giveaway @ ocuk up to 7 C improvement lapping bowed base and IHS he also does a great analysis on the contact problem.


http://forums.overclockers.co.uk/showthread.php?t=18256447

This is my take on it, from the endusers contact analysis my response

Enduser overlay of his bowed sink contact pattern on his IHS

Key thing to remember - increased contact does not increase pressure, but generally increased pressure will increase contact. Pressure is the more dominant of the two.

Let's flip the problem, In engineering a common device used to limit heat flow to sensitive components in an assembly is a “Heat Dam” shown here in figure #3 Where the heat flow is “thermally choked” by reducing the contact area to material limitations. Restricted heat flows with misaligned contact, too little contact, too little pressure, too little compound anything that can restrict heat flow creates a heat dam.

Heat flows on a differential from hot to cold but if you are thermally choked at a point in the thermal cascade downstream improvements in thermal conductivity offers little difference in performance and shows up as reduced deltas of only a degree or so between compounds or what I sometimes refer to as homogenized test results.

Now back to your situation, your analysis of having contact in the right area is correct rather than forcing the heat flow across the IHS to the sink which is a thermally choked condition pushing to the material limit of the copper.

As others have suggested bowed base probably result of clamping and heating and is now awkward/expensive to machine final assembly

 

[MrTransistorm]

Suddenly, CPUID's HW Monitor is released in version 1.8.

Well, it only took them nearly seven months!

I'll give it a shot, but their previous release was all kinds of messed up.

 

[BonzaiDuck]

Well, it only took them nearly seven months!

I'll give it a shot, but their previous release was all kinds of messed up.

I can't say what else to do. I saw one project where someone ground a channel just deep and wide enough in the IHS to insert one of the flat thermal sensors we've all seen or fiddled with that come in fan-controller-kits featuring thermal-control features. That gives you your own TCASE which you can verify.

But I see people coming here for advice who might even be strongly adverse to the lapping option. Machining or filing a trough in the IHS is not something I'd want to do unless I had $1,000 to throw away on additional 2600K processors in the event of damage or failure. It's just . . . too . . . Xtreme!

Since the sensors are on the processor, we need to evaluate the time processors were released against revision-dates for the software. And as I said, general shareware developers may be dedicated, their software may be highly reputable, but they also have to research carefully the processor specifications and likely . . . the chipset specifications. They serve a diverse array of chipsets and processors in doing this. There are bound to be delays -- even errors like the one I experienced pertaining to the TJunction spec of my B3 Q6600.

The motherboard makers have a less daunting task in designing their own monitoring software. They have more reason to assure that it works and reports accurately. The currency of their software revisions would more likely arise with the release of the motherboards themselves.

As others have suggested bowed base probably result of clamping and heating and is now awkward/expensive to machine final assembly
.

Yes, they solder or weld those assemblies, and if this is just an outcome for their manufacturing process, then lapping the base of the heatsink would add cost to the assembly.

For the life of me, I can't figure out -- given the material you presented -- why they do this other than cost. TR for instance justifies it in their promotion as "more efficient." Ah! More efficient for TR!! More economically efficient in reducing cost and maintaining a "margin!!" As much faith as I place in their products, as much as I've had friendly and productive contact with their tech-support staff, this is quite a slippery way of dismissing the issue.

A reseller outfit up in Silicon Valley -- near San Jose or thereabouts and named "Silicon Valley Compu-Cycle" or SVC -- had marketed the TR Ultra Extreme 120 we all know and love in two differently priced flavors: "as is;" and "custom-lapped." I bought the latter. I think the price differential was between $10 and $20. In e-mail contact I had with SVC, their tech-support noted that they had built or acquired a machine and/or jig that allowed them to do it fairly quickly without damaging the welds of the heatpipe assembly.

That being said, and for the benefit of anyone who didn't see the last thread posted here entitled "Lapping your i7-2600K" or some-such title, it is important to emphasize that the lapping is not difficult, but you must grasp the assembly by the base -- not the fins -- or you run that risk.