Huge temp spike with 5930K OC and H110

[PromitR]

Okay, so I am probing the OC/heat limits of my new Asus X99 build, just to get a feel for where things are at. No particular goals here. I'm running in a Phanteks Enthoo Luxe case but no sides right now so close to open chassis. H110 is intaking up top and I'm driving it with the Phanteks SP fans, which generate lots of static pressure.

At stock clocks (3.5), I'm seeing idle temps around 34 C and prime95 load temps right at 60 C (dissipation at 140W). I think this is pretty much expected, especially as my ambient temps are a little warm (75 F).

Next, overclocked to 4.5 using the Asus 5 way optimization utility. I'm seeing voltages at 1.312 or so under prime95 load, adaptive. Power dissipation according to the utility is 230W. Temperatures basically level off around 80-82 C, which is toasty but still not cause for concern in stress test, I don't think.

BUT, here's the part I'm unhappy about. After a couple minutes (like three), the temps suddenly sky rocket. It levels off at 103 C and stays there, occasionally with a slight voltage and clock drop on the CPU. The air coming off the radiator is significantly warmer at this point. This makes me extremely uncomfortable. Did I do something wrong? Bad thermal paste contact maybe? Fans not the right choice? I feel like the temperature shouldn't suddenly explode.

 

[krnmastersgt]

Temperature won't suddenly spike after 3 minutes due to an incorrect mount or anything, it will slowly rise if not mounted well or bad fan choice.

For it to be triggering so suddenly and abruptly there has to be a setting or voltage increase kicking in somewhere in the background either disabling or enabling a setting.

I'd really look into the power options on the board and in your OS to make sure some type of limiter isn't being disabled due to the load duration or that your voltage and how it will adjust to board load is set up. Currently I believe programs like Prime95 are able to completely obliterate CPUs set on the Hybrid power settings common on a lot of Haswell boards, not sure if they made it to Haswell-E. While the setting is great for normal usage, when benching it'll adapt and think you need more power and kick it into dangerous voltage ranges but more importantly generate a ton more heat for you to deal with.

 

[PromitR]

Well I ran a few more tests. First of all, this only happens in prime95 small FFTs; blend levels off in the high seventies. Under a small FFT run, the temperature is stable around 81 C for three minutes, and then rises to 103 within another minute and stabilizes. So while it's not an instantaneous rise, it is rather quick. I'm not seeing extra voltage going to the CPU, although the Asus utility says up front that stress tests can force voltage to go higher than appropriate.

The system IS stable at these settings, but I don't know how to interpret the results or if I should be concerned. A P95 small FFT run is obviously far more extreme than any real workload, and the blend stability is reassuring.

 

[Micrornd]

Just a quick check, you do have the H110 pump hooked into a constant power supply (always 100%), not a fan connector that varies voltage and speed ?

I complained to Corsair when I bought mine that this was not clear in the manual, and they said they would make it clear.

 

[krnmastersgt]

If the tasks are never severely loading on the CPU I wouldn't worry too much about it, stress testing gives an unrealistically high load and temperature. That being said I would load whatever is most CPU intensive for you and watch your temps while its running to see if it exhibits the same behavior, if something in the power settings is still changing the decreased load might keep the temperature in check but still damage the chip from over volting or something.

 

[BonzaiDuck]

Perhaps I should give more attention to the "Official Haswell-E Thread" in "CPUs & OC'ing." There might be a lot of info there, since it's already got a couple hundred posts!

But is this what we can expect from an i7-5930K with an AiO cooler? Approximately 82C, 230W power dissipation -- @ 4.4 to 4.5?

I'm trying to design "next year's build" -- but if there's a chip lottery of uncertain expectations, the decision I've (already) made to do it is . . . tentative . .

 

[Essence_of_War]

So while it's not an instantaneous rise, it is rather quick.

This sounds like you may be getting close to the maximum efficient dissipation of the AIO, and perhaps the coolant temps are getting too high?

 

[VirtualLarry]

Is the coolant boiling?

 

[BonzaiDuck]

Is the coolant boiling?

What would happen with an AiO cooler shipped with some air in the assembly? This would be the reason you bleed a custom-water configuration. . .

 

[krnmastersgt]

Air does lower the potential cooling capabilities, that being said air has a tendency to find a place and get stuck in the loop when all components are stagnant unless you have a really large amount, in which case I'd say hit up the place of purchase and get it swapped or send it to corsair.

The spike is still something weird to me if it can sit stable on ~80. Now if its just stopping by as the heat ramps up, and the temp curve is a bit more constantly rising than what I read from the posts, it could very well be that your set up is not capable of cooling your CPU at the settings you're running it at. I'm surprised your system is actually on at 103C as the Tjunc for most CPUs is ~90-100, which will trigger their thermal alert and just shut down to protect the silicon.

 

[Flapdrol1337]

yeah, the cpu is throttling, making the stability test kind of pointless.

If I remember correctly Prime does the stuff that causes maximum power consumption first, then a bunch of other tests. I guess the water in the cooler and the air in the case warm up, and the cpu temps rise proportionally.

I'd recommend taking your time for the overclock, and set everything manually in the bios, see what voltage you need for 4.2 or something first.

 

[BonzaiDuck]

May be that our discussions are all over the place, between the "Official 'E'" thread in CPUs & OC and here, but that's going to happen. It seemed to me from the reviews that the i7-5820K was the better OC'ing candidate, but you can't really tell from the sparse info we have. And we don't know whether it's his cooler, his CPU, his board or his settings -- or the tests themselves [EDIT].

At least, I can't tell which it is . . . I only know that someone reported a successful OC for a 5820K @ 4.6 using an NH-D14 cooler. Or -- so they said . . .

 

[PromitR]

Thanks for all your input.

yeah, the cpu is throttling, making the stability test kind of pointless.

This is actually okay with me.

I'd recommend taking your time for the overclock, and set everything manually in the bios, see what voltage you need for 4.2 or something first.

The thing is, I may just sit down and dial things in manually. I've been lazy and let the auto system do it, but I simply don't feel the need to run all 12 threads flat out at full overclock. I only want the extra speed for maybe four to six thread applications that lean heavily on single thread performance. *cough* games *cough*. So it's probably time to go in and set the load multipliers by hand.

And it does seem like the temperature of the coolant rises significantly after that first three minutes of stability.

But is this what we can expect from an i7-5930K with an AiO cooler? Approximately 82C, 230W power dissipation -- @ 4.4 to 4.5?

At least, I can't tell which it is . . . I only know that someone reported a successful OC for a 5820K @ 4.6 using an NH-D14 cooler. Or -- so they said . . .

I crash almost immediately at 4.7 and 1.35V. Don't know how much voltage is needed, don't have any interest in finding out. 4.5 and 1.312 is totally stable, heat notwithstanding. Asus' optimization sets 4.6 by default but I don't trust it, so I've been setting lower targets.

 

[BonzaiDuck]

Thanks for all your input.
This is actually okay with me.

The thing is, I may just sit down and dial things in manually. I've been lazy and let the auto system do it, but I simply don't feel the need to run all 12 threads flat out at full overclock. I only want the extra speed for maybe four to six thread applications that lean heavily on single thread performance. *cough* games *cough*. So it's probably time to go in and set the load multipliers by hand.

And it does seem like the temperature of the coolant rises significantly after that first three minutes of stability.


I crash almost immediately at 4.7 and 1.35V. Don't know how much voltage is needed, don't have any interest in finding out. 4.5 and 1.312 is totally stable, heat notwithstanding. Asus' optimization sets 4.6 by default but I don't trust it, so I've been setting lower targets.

I only used the auto-OC features to get a handle on an upper voltage level and expected clock value initially, and I found it very useful for that. I have a notebook of the last build, and I'd recorded all the numbers, read the actual values reported in BIOS, and then went back to stock as I stepped up the various settings using the record to set baselines. I actually had a whole series of tested settings with bCLK values above 100, and the new motherboards with the higher strap might be worth exploring. I won't know how it all works until a month or so before I build it. I'm guessing I don't even have an acceptable handle on how the BIOS structure and the chip features have changed.

But I'll just throw this in. I'd pretty much guess that the prudent upper-voltage limit for that chip is somewhere around 1.30V. It could even be lower. Somebody could speculate or throw in their well-tutored experience in these chip generations.

 

[PhIlLy ChEeSe]

Okay, so I am probing the OC/heat limits of my new Asus X99 build, just to get a feel for where things are at. No particular goals here. I'm running in a Phanteks Enthoo Luxe case but no sides right now so close to open chassis. H110 is intaking up top and I'm driving it with the Phanteks SP fans, which generate lots of static pressure.

At stock clocks (3.5), I'm seeing idle temps around 34 C and prime95 load temps right at 60 C (dissipation at 140W). I think this is pretty much expected, especially as my ambient temps are a little warm (75 F).

Next, overclocked to 4.5 using the Asus 5 way optimization utility. I'm seeing voltages at 1.312 or so under prime95 load, adaptive. Power dissipation according to the utility is 230W. Temperatures basically level off around 80-82 C, which is toasty but still not cause for concern in stress test, I don't think.

BUT, here's the part I'm unhappy about. After a couple minutes (like three), the temps suddenly sky rocket. It levels off at 103 C and stays there, occasionally with a slight voltage and clock drop on the CPU. The air coming off the radiator is significantly warmer at this point. This makes me extremely uncomfortable. Did I do something wrong? Bad thermal paste contact maybe? Fans not the right choice? I feel like the temperature shouldn't suddenly explode.

No offense, to you or anybody. Do you really think an H-100 will cool a 12 core CPU? seriously! It can not remove the heat effectively, I don't care who says its made for a socket 2011. The hose is what 3/8 ID at most? Then again when do you ever run prime for more then 10 minutes? Try some extended BF4 then check yer temps using realtemp.

Also the auto overclock will give a lot of volts to the CPU, you can manualy turn it down when you log into windows with the AI tuner.

 

[BonzaiDuck]

Well, my 2600K has -- what? -- 95W TDP at spec speed and voltage. I've just confirmed that my optimally-installed D14 gets approximately the same load temperatures -- very similar hardware, same VCORE under load -- as an H110. [I think it was an H110].

We get a 4-core-average load temperature in the mi-60s for the Prime95 sFFTs; we didn't make a special effort to control for ambient, but this is also confirmed by a Frosty-Tech article.

So I'm looking at these temperatures for coolers just a notch below as reported here and in other threads or forums. You can do your over-clocking on the lesser coolers, but your stress-tests are going to push temperatures something higher than that.

I don't really like those temperatures, so unless somebody comes out with a "Nepton 360 L" or similar, I'm looking at a "custom" parts assembly.

Aren't the Haswell-E processors made with a 22nm lithography?

 

[guskline]

Following up with PhillyCheese's comments, I have a 3930k (6 core/12 thread) and yes when cranked running certain benchmarks it is going to throw out heat.

The AIOs are convenient but run out of rad capacity for the really big chips when running the most extreme tests.

AdamK47, a great poster, uses AIOs but chose the new Thermaltake Extreme 3.0 with the 360 rad on it for his 5960x. Smart move.

 

[BonzaiDuck]

Following up with PhillyCheese's comments, I have a 3930k (6 core/12 thread) and yes when cranked running certain benchmarks it is going to throw out heat.

The AIOs are convenient but run out of rad capacity for the really big chips when running the most extreme tests.

AdamK47, a great poster, uses AIOs but chose the new Thermaltake Extreme 3.0 with the 360 rad on it for his 5960x. Smart move.

That's good. Even so, a mere 3.5C degrees improvement in OC test scores over top-end heatpipes (D15), as per a Tweaktown review linked through Frosty Tech. That's still only an increment more in cooling. But I think -- again -- I could close that gap, given the way these test-beds are prepared or the way they deploy the cooler.

But if you want a bigger radiator, you need a bigger case or you need to give up other hardware installation options. Either way, the size of the radiator means a lot, and it drives your case-selection. If the case-makers don't keep up with a niche trend in cooling preferences, you're going to find yourself tinkering, sacrificing, modding or dealing with more expensive external heat-exchangers. Or all of the above . . .

 

[BonzaiDuck]

That's good. Even so, a mere 3.5C degrees improvement in OC test scores over top-end heatpipes (D15), as per a Tweaktown review linked through Frosty Tech. That's still only an increment more in cooling. But I think -- again -- I could close that gap, given the way these test-beds are prepared or the way they deploy the cooler.

But if you want a bigger radiator, you need a bigger case or you need to give up other hardware installation options. Either way, the size of the radiator means a lot, and it drives your case-selection. If the case-makers don't keep up with a niche trend in cooling preferences, you're going to find yourself tinkering, sacrificing, modding or dealing with more expensive external heat-exchangers. Or all of the above . . .

I went back and double-checked the reviews pertaining to "Thermaltake Extreme 3.0." I'm puzzled. the unit in those reviews sports a 240mm radiator or dual 120mm fans. What am I missing? Is there a version with a 360 radiator?

 

[krnmastersgt]

I went back and double-checked the reviews pertaining to "Thermaltake Extreme 3.0." I'm puzzled. the unit in those reviews sports a 240mm radiator or dual 120mm fans. What am I missing? Is there a version with a 360 radiator?

There's a 360 version known as the 360 ultimate, some sites called it the 3.0 extreme still but there is a 360mm version on the market, albeit not that common.

 

[BonzaiDuck]

There's a 360 version known as the 360 ultimate, some sites called it the 3.0 extreme still but there is a 360mm version on the market, albeit not that common.

Then my remarks about comparison testing was off the mark. That leaves the question as to how much the 360 reduces load temperatures over the 240/280 model, whichever it is. [Submodels . . . Geeesh!!]

Do you -- does anyone -- have a link to a comprehensive comparison review that would include, perhaps, the Nepton 280L, Hydro-110i/110, D15, D14 or U14S -- or just a subset that might have two of these? I apparently snagged the dual-fan model, which wouldn't be the 360, would it?

 

[guskline]

BonzaiDuck, I stand corrected. I think the Extreme 3.0 might have been a thick 240 Rad. AdamK47 has the Thermaltake 3.0 Ultimate which in fact has a 360 rad. For an AIO I think it is a wise move for purchasers of a 5960X. Otherwise, use a custom watercooler and a big rad!

You can see from this link to the Thermaltake website the difference:
http://www.thermaltakeusa.com/cooler.aspx

 

[BonzaiDuck]

Folks may disagree with my following observation, but I'm going to make it

Looking at the Intel spec sheet for these E processors, I'd give the following advice to anyone unsure of their "overclocking progress."

If the load and turbo voltages for your overclock leave you feeling comfortable, but your six/eight-core temperature average or average-of-maximums exceeds ~ 81C, your cooling system is inadequate for the clock speed you are trying to achieve.

Of course, someone could argue that the processor will never be under that kind of load in normal operation . . . .

 

[Essence_of_War]

That's a pretty specific claim.

How exactly are you getting to those numbers, are you running the TDP of the CPU vs. the TDP for the radiator @ +10C liquid temp rise?

 

[BonzaiDuck]

That's a pretty specific claim.

How exactly are you getting to those numbers, are you running the TDP of the CPU vs. the TDP for the radiator @ +10C liquid temp rise?

I remember over the years various forum members would disparage the relevance of a TCASE thermal spec when there's no such sensor in the CPU package. All we have are the core sensors.

Consider this. Intel no longer publishes a "safe range" voltage spec, but -- if you look at the spec sheet for ANY of the Haswell-E processors, they are still publishing the TCASE spec. My SB-K [aka SNB, 2600K, etc.] has a spec of ~72.5C or 73C. The Haswell-E's are spec'd at around ~66+C.

Going back to Conroe/Kentsfield days, when we were firmly into "multi-core territory," Tom's Hardware (I'm pretty sure) posted an article explaining the difference. A special effort was made to measure TCASE, determining that it could be as much as 15C below the core-sensor average and nominally 10C. However, these numbers would also depend on the TDP of a particular processor, or more specifically -- the thermal wattage of someone's overclock choices.

But the common factor in those guess-timations was the unvarying nature of integrated heat spreaders -- same copper material, same thickness.

That leaves us to further speculate (speculate! speculate!) about the meaning of the spec. Does it express Intel's concerns about the effect of heat on silicon? It IS supposedly a spec directed at case-design for the purpose of cooling. It would not so easily summarize any relationship between voltage and temperature -- but we know that significantly lower temperatures result in a slightly-reduced voltage requirement for any given overclock setting.

Ultimately, you or any others can argue that I'm way out on a limb with my pronouncement.

Put another way -- if it were MY processor, I'd stop pushing the clock speed higher when the temperatures begin to overlap that range -- or the low 80's C. If someone wanted to go higher, AND -- they deemed their voltage settings "reasonable," "safe" -- whatever -- I'd think one might wish to increase cooling capacity.

Of course, you can go ahead and tune the system so that the most severe load-tests are producing temperatures in the low 90's. Chances are -- we all agree -- you won't ever see those temperatures again. So you would not be contributing to any thermal degradation of the processor, either.

On the other hand. Would you be able to reach the same clock setting at slightly lower voltage by deploying more effective cooling?