Stress tests in detail: Test CPU and overall system correctly (Tutorial - Part2) | Page 7

Let us now turn to the temperatures that could be measured as a direct result of the waste heat generated. Since we have also used different areas of the CPU with the individual test runs, we have also used very different areas of the CPU, so interesting findings arise.

Let's start with the core temperatures, which are measured with the integrated digital sensors (DTS) in the computing units themselves and which react almost in real time to the actual load. Our cooling with the chiller and the cooling fluid, which is always exactly 20°C cold even under load, thus also shows a very real measure of utilization. We calculate the means from all cores and over the entire duration:

What is considered a package temperature or return value of the PecI (PecI = Platform Environment Control Interface) is approximately at the level of the mean core temperature. Here, too, we have determined the mean value over the entire running time of 30-45 minutes (until the stable final temperatures are reached), which is usually at the level of the core temperature or even in individual cases, depending on the type of test.

We are now evaluating an additional sensor that detects the temperature directly below the socket. Since the CPU cooler cools the CPU via the heatspreader and the Die only from above, this value is very sluggish compared to the DTS, especially for the stability and cooling performance. Here the CPU reaches a whopping 105°C without any overclocking, despite the chiller used!

Precisely this result is one of the reasons to use Prime95 with AVX, if at all, as a stability test over a longer measuring period, if one is also able to read all relevant temperatures safely with a suitable program.

What is the end use of a supposedly reasonably well cooled CPU (DTS) if the socket then burns unnoticed in the evening wind over time?