I'm fairly confident while no less humble that I've had a lot of experience with cooling the 2600K and 2700K chips. These would not have a significant variation in a cooling profile to the 2500K. If we launch into another discussion of heatpipe coolers, TIMs, lapping or any substantive detail about those matters, I won't initially promote it here.
Those processors all have an Intel TCASE spec of ~ 73C. TCASE is only a guideline for case design and overall cooling solutions for OEMs. There is no sensor corresponding to TCASE; the temperature that corresponds to the spec could only be measured at the exterior top center-point of the processor cap or IHS.
It is not a "bad" strategy to apply that spec to what the core sensors report, or the "package" temperature seeming to correspond with the highest core temperature. But such a strategy could also be described as either a bit too cautious, or a tad "sissy." Nevertheless, I loosely follow that strategy.
I have to speculate about what the spec -- TCASE -- actually means. I guess that it would be a temperature threshold such that one of these processors, running at a temperature above the spec 24/7/365, would experience slow, gradual thermal degradation. That is, OEMs would not want to design cases and build computers that slowly cooked a processor, even if at speeds near idle.
Now you are free to disagree with me about that last paragraph. But my point here is this. Stress-testing your computer at 80C, 85C or even higher for several hours or even a day shouldn't have any significant effect on its overall lifespan. You are simply not going to run your processor at those speeds 24/7, month after month.
That being said, I've been experimenting with three well-favored heatpipe coolers. I can clock the Sandy-Bridgers to stable multipliers with maximum detectible voltages no greater than 1.38V. This corresponds to severe load voltages drooped to about 1.35V. Since they are 32nm processors, this also corresponds to the last published "safe limit" for a 32nm processor just preceding introduction of Sandy Bridge.
The second-best cooler will limit temperatures for these loaded settings to between 76C and 78C. For the best of the three, I can expect between 72 and 74C.[see * below.] To be more specific, those numbers correspond to either LinX or Intel Burn Test. You would see maximums more like 68C with OCCT:CPU or Intel Extreme Tuning Utility (aka XTU).
Ordinarily, if I saw 95C, I might assume the stock cooler was installed, but the OP seems to be using a heatpipe tower in which two fans were working against each other.
Also, to address stockwiz's note, Most of my stress-tests occur with room-ambients between 77C and 82C.
As per the clocks (4.7 Ghz across the board), the "unloaded" maximum voltage trapped by HWMonitor and temperatures, and heatpipe coolers -- this seems to be the limit for me. That is, I could raise the voltage for a higher clock; the temperatures will increase accordingly. Even if I were satisfied with the higher temperatures, I would not be satisfied with the voltage.
====
* The coolers in question will not perform that well with their bundled fans and stock installation. I suspect my mods and enhancements net me at least 5C to 8C improvement. To be more precise, I have measured the improved results.
Facebook
Twitter