Unexplainable tempurature rise from 2500k

[HFat]

If you've ruled out changes in fan speed and case temperature, these two measurements would tell if the temperature increase is due to an increase in power consumption, is due to a decrease in heatpipe efficency or is entierly spurious.

[Klusu]

You can not explain the rise. You should find out if there is a rise... PSU? No.

[markanini]

I could see the concern for temp readings accuracy for a previous generation processor, but for Sandy Bridge? Or could the board be messing those up? I don't have suitable equipment to measure the heatsink temps.

[quest_for_silence]

I could see the concern for temp readings accuracy for a previous generation processor, but for Sandy Bridge? Or could the board be messing those up? I don't have suitable equipment to measure the heatsink temps.

I think that you have just to do what HFat already suggested (providing that I have well understood him): disable any fan control and run the fans at 100%, disable any power saving feature in BIOS, then run your tests.
Use the specific OCCT PSU test in order to check voltage stability (see the resulting charts looking for any weakness, if any).
Then, if all it's all right, run 4 x Prime95 in "Large FFT" mode only, and log with SpeedFan (or what you may rather) the temperature changing.
Eventually report here these your findings. Have a good luck!

[ntavlas]

Measuring power consumption could indeed help determine if the temperature reading is bogus. Still, I somehow doubt this is the case. I think that even with the large FFT option checked, there could be some variation in cpu utilization. Perhaps some unnecessary data in the cache prevents the cpu from performing optimally at the beginning of the test? Don`t forget that the cache is used by the gpu as well.

Does the rise in temperature happen after a random amount of time? If it does, it could be because of a background process preventing the cpu from being fully utilized (see above scenario). It might be worth looking at the process tab in task manager just in case.

I still think that a variation in cpu utilization is the most likely scenario. You could try using an different stress test program like cpu burn and see what the results are like. Taking the intergrated gpu out of the equation like others recommended could also help, even an old pci card would do.

[markanini]

I think that you have just to do what HFat already suggested (providing that I have well understood him): disable any fan control and run the fans at 100%, disable any power saving feature in BIOS, then run your tests.
Use the specific OCCT PSU test in order to check voltage stability (see the resulting charts looking for any weakness, if any).
Then, if all it's all right, run 4 x Prime95 in "Large FFT" mode only, and log with SpeedFan (or what you may rather) the temperature changing.
Eventually report here these your findings. Have a good luck!

Ran it for an hour and saw nothing truly remarkable except that the temp rise did not kick in. Means my PSU is ok?

Measuring power consumption could indeed help determine if the temperature reading is bogus. Still, I somehow doubt this is the case. I think that even with the large FFT option checked, there could be some variation in cpu utilization. Perhaps some unnecessary data in the cache prevents the cpu from performing optimally at the beginning of the test? Don`t forget that the cache is used by the gpu as well.

Does the rise in temperature happen after a random amount of time? If it does, it could be because of a background process preventing the cpu from being fully utilized (see above scenario). It might be worth looking at the process tab in task manager just in case.

I still think that a variation in cpu utilization is the most likely scenario. You could try using an different stress test program like cpu burn and see what the results are like. Taking the intergrated gpu out of the equation like others recommended could also help, even an old pci card would do.

A bug in the Sandy Brigde processors cache? The temp rise seems periodic, switching in an out, as it were, every 15-20 mins. I'll try CPUburn and Small FFTs and see what that does next. Linpack apparently doesnt cause the temp bug to happen.

EDIT: Small FFT setting causes the temperature rise as well.
EDIT2: Futher Linpack testing in underway.

[HFat]

If you can't measure power consumption or temperatures, don't waste too much effort on aimless experiments. Your rig works even if you don't have the tools to analyze everything it does.

[quest_for_silence]

Ran it for an hour and saw nothing truly remarkable except that the temp rise did not kick in. Means my PSU is ok?

It would look like PSU, CPU and cooling system are all right.

A bug in the Sandy Brigde processors cache? The temp rise seems periodic, switching in an out, as it were, every 15-20 mins. I'll try CPUburn and Small FFTs and see what that does next. Linpack apparently doesnt cause the temp bug to happen.

EDIT: Small FFT setting causes the temperature rise as well.
EDIT2: Futher Linpack testing in underway.

But how have those tests been performed? With all fans running at 100%?
If it were not so, I think that everything go as it has to go and it's just how CPU/SpeedFan/tests interact with each other.

[markanini]

Sorry guys turns out it's normal for Prime95 in-place mode to vary the temperature, the same applies for Blend mode. IntelBurnTest(same as LinX basicaly) really heats up the processor, reaching 78C max in standard mode with the stock cooler at (constant)1200rpm and a Nexus 120mm at 700rpm for side intake(not my favorite setup but keeps max temps 3C lower). All at stock clocks and a modest undervolt, ambient 21C. Can't wait for the True Rev.c to arrive, should be here any day now. Already jonesing for Noctua C14 though. Aye Aye aye!

[lodestar]

Already jonesing for Noctua C14 though.

Noctua C14, well maybe. The hot ticket right now seems to be the Thermalright HR-02 Macho, which my local dealer has for about two-thirds of the price of the C14.

[markanini]

Already jonesing for Noctua C14 though.

Noctua C14, well maybe. The hot ticket right now seems to be the Thermalright HR-02 Macho, which my local dealer has for about two-thirds of the price of the C14.

Yeah, it's not value champ. I used a TR SI-120 for my s478 and s775 systems so I might be a bit nostalgic when I see the C14...