CPU stupidly hot, suggestions....

[roachcoach]

Morning,

I was mucking about last night as I was debating thinking about overclocking my CPU but I thought, lets check the temp profiles properly.

Up until last night I'd been using asus PC Probe that came with the mobo and it was reporting CPU temps of 29-31 idle and mid-high 40's during gaming. I figured that was fine and dandy.


Anyway, being new to this overclocking business I started to do some research and picked up the various tools of the trade (realtemp, prime97, linX etc) and had a very nasty fright.

Realtemp has the idle temps anything from 10-15 degrees higher than PC probe did and after about 8-10 minutes of prime95 hammering it, it's clocking in the 99-100 degree region (pc probe was claiming 86 at this point). Rather freaked at this point, but I'm not sure what's right, even at the peak, the air exiting the case was pretty cool - I do have a pretty decent case cooling, but I'd think a CPU at boiling point would get the exit air at least room temp, it was still blowing out cool. Big case though, maybe I didnt let it run long enough to blast heat out.

For reference a run of 3Dmark11 right through the benchmarks saw peaks of 70-75ish in realtemp (can't recall exactly).


So, my question then is....what's the likely cause? Badly mounted is about all I can think of. I am assuming something is wrong to be hitting those temps.

It's really annoying because I've had pc probe up religiously and it's never even wiffed of an issue but it seems the damned thing reads super low Also there doesnt seem to be much that hits the machine as hard as Prime95 did (I guess we can thank consoles for something), even gaming the CPU isn't topped out so it won't typically hit those temps in day to day use.

Spec is still per profile here, case has about 6-7 120mm fans push/pulling, and the big jumbo one up top.

[kalniel]

You would expect the air to be warm if temps are high. If the air is cold then it's not carrying the heat away from the processor (though if you've got a huge volume of air it could be just that the increase is too small for you to notice).

As for differences between probes.. that's kind of the nature of software monitoring, it's never that accurate. I wouldn't worry about absolute values, but consider the delta between idle and load (especially when the computer is started from cold), and how stable the temperature is. The most important thing is that the temperature stabilises on the most torturous test you can throw at it. When it does so, check if there's any throttling going on (frequency, voltage etc.). If not then you can forget about temperatures - your cooling system is capable of coping with the worse case and completely unrealistic scenario, so it'll be more than fine in real use.

RealTemp has some kind of sensor calibration, but I've never played with it. Try some other programs (aida?) and check the temperatures reported in BIOS so see if you can get a consensus on which software might be more accurate. If you're still concerned then check the seating of the HSF (not too much thermal paste etc.) and that your push/pull configuration is not too disruptive to airflow.

[roachcoach]

It is a BIG case and I didn't run it for that long.

I did look at the bios but the trouble is, within seconds of ending the stressing, the temps drop like a stone again.

Is there any easy way to detect throttling? Presumably the Mhz output of the chip would drop?


The temp at idle are stable and they were relatively stable at stress. But I didn't let it sit at those heats for very long, from a great many forum posts this chip should be aiming for the 80's under load rather than 100


Edit: I know its a different beast but I'm currently beasting this laptop (i5-520) and it's peak temps are 83....oh 84 now. Different machine I know, but I'm flattening this CPU and with all the might of laptop cooling it's showing 20 under my desktop.

[shaithis]

Presumably the Mhz output of the chip would drop?

Yup.

Just run something like CPU-Z while running the stress test. While stressed, the CPU should never drop down to a slower speed unless it is throttled.

Also be aware that there is more then 1 sensor to read "CPU Temp" from, different software may read a different sensor.

It's one reason why I like speedfan, the sensors make a lot more sense when you see where the sensor is.....

[peterb]

Why are you believing the third party software In preference to the software that was designed by the mobo mfr for their mobo?

You need to read the CPU data sheets to find out when throttling occurs (and when/if the cpu shuts down at over temp) but if those reported temps were correct I'd certainly expect the CPU to be throttling or shutting down. Some BIOSs also sound an alarm under overtemp conditions.

And if the CPU was running at that temperature, the CPU heatsink would hot too (assuming it is installed correctly)

[roachcoach]

I'll give speedfan a try, if it its a random freakout. I do have a horrible suspicion that realtemp was right though.

Temp is still the same on this laptop (between both programs), if this laptop has better cooling capability than my bloody desktop I'm going to get very upset indeed :/


@PeterB: It was the software of choice in nearly every overclockers forum, those guys do know what they're about so it seemed fair to trust it. Even the mobo stuff was showing 86, which is higher than I'd expect.

I didn't open the case to check the heatsink, it was all a bit "eeep this is bad must look...oh ****, is that the time?" last night.



Edit: Stopped stressing the laptop, interestingly although the peak temps were lower than the desktop, it's not shedding heat anywhere near as fast as the desktop did. If that means much, I don't know

[kalniel]

It is a BIG case and I didn't run it for that long.

I did look at the bios but the trouble is, within seconds of ending the stressing, the temps drop like a stone again.

That's no problem - just compare idle temperatures. If the software's off for load temperatures it'll be off for idle too.

Is there any easy way to detect throttling? Presumably the Mhz output of the chip would drop?

CPU-Z shows frequency, stepping and voltage. Windows 7 resource monitor also graphs frequency.

re: the laptop - it'll have a different chipset and different sensors, so while in most cases realtemp corresponds fine, it could be that for the particular sensor in your desktop it doesn't.

[roachcoach]

I'll check the bios when I fire it up when I get home (should've worked from home tbh :/)

The case could do with a cleanout, although its not that long since it was last done, it sure can't hurt to rule it out.


That's cool, realtemp has that as well as some bits & bobs which came with the board.

[sam3]

I would say the asus probe temperature is the more reliable (it should be given you the same readings as in the bios and this is the temp reading used for the thermal shutdown). The realtemp readings are probably not calibrated correctly (are the readings core readings or cpu reading as there is usually quite a difference)?

My Phenom II idles around 30-35C and gets into the 40's gaming and intels generally seem to be cooler.

[kalniel]

My Phenom II idles around 30-35C and gets into the 40's gaming and intels generally seem to be cooler.

Well it's a skt 1366 chip and they're a bit warmer than intels mainstream chips. I could give my prime95 load figure in realtemp of 70C but it doesn't mean anything because I have a different motherboard and cooling setup.

[sam3]

Well it's a skt 1366 chip and they're a bit warmer than intels mainstream chips. I could give my prime95 load figure in realtemp of 70C but it doesn't mean anything because I have a different motherboard and cooling setup.

Im using an overclocked 1090t which isnt exactly a cool running processor

just a quick google search for "i7 930 realtemp" brings up loads of pages of people asking the same question. The most common realtemp reading for this proc appears to be in the high 80's (prime95) so your reading does appear to be quite high.

So check fans are the correct orientation on your cpu cooler and make sure its mounted correctly (and try replacing the thermal paste if you have some spare).

[roachcoach]

I had similar google results, though iirc most of those hitting the 80 range were also overclocked, stock seems to be hard to spot (funnily enough)

[kalniel]

What motherboards and BIOS versions are these google results?

[roachcoach]

Couldn't say, I've followed a silly amount of links in the last few hours. I know what you're alluding to though, but still, temps approaching 99-100 I've never seen a single post suggesting that's cricket.

Anything in the mid-high 80 or more has resulted in 'might wanna check x,y,z//something aint right' responses.

[peterb]

<---snip


@PeterB: It was the software of choice in nearly every overclockers forum, those guys do know what they're about so it seemed fair to trust it. Even the mobo stuff was showing 86, which is higher than I'd expect.

Hmm, without looking at the forums in question, difficult to form an opinion, but I suspect that a lot of overclockers just follow what is considered to be the best, and just because one app works with one mobo/CPU combination, it doesn't mean it will work for all. I use a linux temp application, and that needs careful setting up, simple because of the way different motherboards and chipsets monitor and interpret sensor data. I don't know if realtemp allows for different chipsets, maybe automatically.

I didn't open the case to check the heatsink, it was all a bit "eeep this is bad must look...oh ****, is that the time?" last night.

Well physically measuring the heatsink temp (even if subjectively by touching it) is worth doing, although remember that the heatsink temp will be lower than the die temp anyway.

Edit: Stopped stressing the laptop, interestingly although the peak temps were lower than the desktop, it's not shedding heat anywhere near as fast as the desktop did. If that means much, I don't know

Not a lot - depends on thelaptop mobo and chipset, but if the same software is giving you results that are different from what you expect, it is either under-reading on the laptop, or over-reading on the desktop - or there is something wrong with the cooling arrangements on the desktop. But I'd trust the mobo mfr's software over any third party software, unless I know that the tp software has been specifically tested with my mobo.

[kalniel]

Couldn't say, I've followed a silly amount of links in the last few hours. I know what you're alluding to though, but still, temps approaching 99-100 I've never seen a single post suggesting that's cricket.

Anything in the mid-high 80 or more has resulted in 'might wanna check x,y,z//something aint right' responses.

Sure, and if it's not right then it should be throttling so you can check that, as it's a hardware response, without worrying about software. Prime95 small FFTs does it for me, or you could try OCCT.

If it concerns you though, just take it off, clean up and then remount with a correct amount of grease (that sentence could be so misconstrued..), just for ease of mind, and it might clear the problem.