I recently built a Sandy bridge system for my self, details in my signature. I'm getting an unexplainable sudden tempurature rise after runnuing Prime95 for approx 17-18 mins:

If I stop Prime95 and run it again the CPU tempuratures drop down to normal:

After a while they suddenly rise again:


Above graphs are running default clock settings. When trying a modest overclock at 3,8Ghz at lowest stable vcore the tempurature rise is even larger, about 10 degrees. I cant come up with any explanation, I see no remarkable changes in speedfan beside CPU temps. Do note that the temperature rises very rapidly, rougly 30 secs. I could only suspect some component close to the CPU such as voltage regulation circuitry being the cause but I'm really stumped.

Turbo boost kicking in?

Or the program doing different computations. Or something loading the GPU. There are a number of software-related explanations.

Something physical might be happening however. If you're using heatpipes, you might be heating them more than they're deisgned to and thereby temporarily lowering their efficiency. But this shouldn't be happening unless you have a weird heatsink design.

The readings could also simply be bogus.

Now, why didn't I think of that! Being on the same die as the CPU it would certainly explain the rapid temperature rise. I've got some testing to do...

TurboBoost kicking in ? It kicks in if a certain performace/heat is right, right ?

Someone should slap you guys for not knowing how TurboBoost operates. Turboboost doesn't apply here as all cores are fully loaded.

Here are the results of my testing:

Graph starts with only Furmark running. Furmark is stopped and temps drop. Prime95 is started and temps rise. Furmark is started on top of Prime95 and temp rise further. So far everything looks normal. But, take a look at the above temps and compare that to the highest temps in my first post. You'll see it's 2-3 degress lower. Note that only Prime95 was used in the first post. Ambient temps feel a degree or two warmer if anything. I let it run for a while longer:


Something is obviously going on that shoudnt be. Both the CPU cores and iGPU should be generating max heat within minutes of starting stress testing, not suddenly rise a while later.

My understanding was that there was still a single step available even when all four cores were fully loaded. This seems to be supported by Intel's documentation http://www.intel.com/support/processors/corei5/sb/CS-032278.htm?wapkw=(cs-032278). It might be worth downloading the Intel Intel Turbo Boost Technology Monitor and seeing if that tells you anything further http://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&DwnldID=19105.

Ran another Prime95 sesson and kept an eye on the multiplier in Realtemp as well as Intels monitoring utility suggested by lodestar. Stays at x34 throughout. Baseclock stays at a stable 99.78MHz after a minute or two of Prime95. As expected, sadly (would have been a pleasant surprise if not).

Just for fun, what heatsink and CPU fan are you using?

Edit: Also, FYI for everyone, x34 should be a turbo mode for a 2500k. Base clock is 3.3GHz.

Why the same one Anandtech used to overclock a 2500k to 4.5GHz, Intel stock cooler. That said I'm avaiting delivery of a Thermalright True Rev. C.

Theres a setting in UEFI called 'power saving mode' that applies a conservative undervolt to Vcore. I've had it enabled so far. Wanna see what happens when I disable it and my tempurature bug kick in during Prime95?

the heater in the house kicked on?

I wish!

I was just trying to get an idea if the ~70*C was a reasonable temperature or not. I have a 2500k cooled with a Kozuti in a GD-05 with all the case fans on full, so plenty of airflow, and it was peaking at about 70 also under P95+FM (before today's cold snap and snowstorm, anyway). I'd suspect the stock cooler may have roughly the same cooling capacity. I've never used the stock cooler - if it has a PWM fan, could the fan speed ramping be causing this, with the spike coming after the fan is on full? If it's a PWM fan, I'd suggest throwing it onto a 3-pin header and having it run at a constant 12v and see if you still get this behavior.

As for the heater, when I had a hideously overclocked Voodoo2 back in the day, we put some insulation around my bedroom's heater pipe. Five minutes later, heat crash.

Yeah, if the fan speed was targetting a temperature I imagine something like that could happen... but would it be so slow? Over 10 minutes of the fan slowly ramping up followed by such a quick temperature rise that ends abruptly?

Yes, that's right. Under the extreme load of Prime95 Turbo Boost is applying the one bin available (3.3 ->3.4 Ghz) to all four cores virtually immediately. So Turbo Boost is working as intended. I must say the temps look OK for a CPU under that load with Turbo Boost engaged and the Intel stock cooler. The slight rise in temperatures after the initial period could be any number of reasons but it seems to me the cause is probably external to the CPU and chipset. Something like HWMonitor from CPUID http://www.cpuid.com/softwares/hwmonitor.html, as well as CPUID itself might be useful to just see what CPU voltages and fan speeds are doing, as well as system temperatures, etc. Both HWMonitor and CPUID are free downloads.

Again, I don't know, since I've never used a heatsink with pushpins. I could believe the temperature would rise that quickly once the fan speed increases got exhausted - the stock cooler doesn't have a huge amount of space to store heat, and temperature goes up quickly under that kind of severe load on a 2500k. When I tried to cool mine with a passive NT01 it spiked like that. In any case, the PWM from the stock fan is all I can think of that might do that, and plugging it into a 3-pin header would be an easy test.

What do the CPU monitoring tools say for the voltage ? is it stable ?

You know, the more I look at your graphs, the more I think the PWM settings are wonky. The Prime95 graphs show the temperature spiking, then drooping. Mine do NOT do that, and I've never seen that on any of my machines, ever. P95 always causes that curving ramp up until it hits a zone of a few degrees it bounces in, and then it just stays there indefinitely. Please try a non-PWM header.

When run with default settings, most variants of Prime95 (eg, OCCT) alternate between small and large FFTs. Small FFTs generally fit in the CPU cache and run hotter. Large FFTs thrash the cache, so the CPU cores spend more time waiting on memory fetches.

What specific Asrock board do you have? Do you use the onboard graphics in any way? I've only used the in-place FFT option as it generates the most heat. The other modes seem worthwhile though and you made me download OCCT (nice program) so thanks for the heads up!

The behavior is a result of temps during Prime95 landing close to target for the fan to spin up as set by uefi. The fan will kick in and out for a short while until stabilizing. What you would expect to see would require a lower target temperature.

As I mentioned earlier everything but temps are as expected, including voltages, fan.

I've read up a bit on a confined stability issue with a different Z68 Asrock board that could arise when the is iGPU is active. A solution proposed by an Asrock Representative was making sure to use memory supported by the manufacturer and truthfully my RAM isnt one of those. Wish I had some discrete graphics on hand to see if it makes any difference. Somehow I find myself jonesing for a nice board like MSI Z68A-GD55, VRM section looks too sweet. Awaiting replies from Asrock and my retailer about my issues.

Ah, I see, the Kozuti I use stays on and just slows down when the CPU is not loaded. Still, is there a reason you haven't tried using a non-PWM fan header and seeing if you still get your odd temperature spike? Your comment here only makes me suspect more strongly that odd PWM settings could be causing this.

It's been connected to the 3-pin header since day one, I discovered the fan could spin at ~800rpm minimum compared to ~1000rpm minimum from the 4-pin header. It stays at constant ~1500rpm IIRC after a few minutes of Prime95. Nevertheless I'll see if anything changes when my True Rev. C arrives and hook up up my S-flex to the PSU molex connector. That said I'm in a quiet enviroment and my ears confirm what the monitoring program tells me though, the fan stays constant during Prime95.

Hmm, any chance the PSU might be responsible? Like feeding the VRMs off spec voltages that makes them work harder and generate more heat?

Oh, you're running it on a 3-pin, but controlling it? Can you please try setting it to a constant speed on that header and then try the test? I'd like to see if the temperature curve still behaves oddly with a constant cooling capacity. No offense to your ears, but that'd be more convincing to me. Still, I was sure that was it, and if it's not, I'm very puzzled.

I'd think that if the VRMs were getting off-spec voltages but providing the correct voltage to the CPU (as you said it's getting stable voltage), the VRMs would heat up extra but nowhere near enough to heat up the CPU without burning themselves up. With everything stable except temperature, I'm stumped.

Nothing that you've measured this far changes, so I'd give what is suggested by Scrooge a try. The only thing you don't know is fan speed ; maybe it's the combination of all your fans that causes the temps on your CPU to fluctuate like they do... still curious to see what it turns out to be !

Asrock Z68 Extreme4. I use a graphics card. The CPU-based graphics are configured with Virtu, but I'm doubtful they ever actually get used.

Okay I connected the fan header to a PSU molex connector via an adaptor. Same results.

Of course. The fan was already at max.
Must be a change in what the CPU is doing. I would measure the temperature of the heatsink. And the power draw.

Well, it confirms that the fan wasn't slowing down, so I do think it was worth doing. Total power draw would be interesting. I'm out of ideas for a cause though, sorry.

Didn't except different result but I have one less thing to rule out with certainty now. Klusu, What could we learn by measuring HS temp and power draw? On a diffident note, could the PSU have a part in the temp rise by feeding off-voltages to the board?

EDIT:
Damn, wish I payed attention this graph while skimming through reviews before buying the Pro3:

Would have scared me off, for sure.