CPU Only: Measuring Temperatures With Different Stress Tests
Next up: temperatures. These are directly related to the waste heat dissipated during our stress tests. Various workloads push different parts of the CPU in unique ways, yielding more interesting findings than we were expecting.
Core temperatures are measured by the CPU’s integrated digital thermal sensor (DTS), which reacts to actual load levels on an almost real-time basis. Between this and our chiller-based cooling solution, which maintains its fluid at a near-constant 20°C, we get very accurate load measurements. Then, we calculate the average temperatures across all cores throughout our test run.
Package temperature, or, more precisely, the reading provided by the platform environment control interface (PECI), is typically similar to the average core temperature.
Once again, we calculate the average across the entire 30 to 45 minutes it takes for the temperature to stabilize. While the results usually come close to the respective core temperatures, they do exceed them in certain circumstances, depending on the type of test.
Next, we examine data from an additional sensor located directly underneath the socket. Since the CPU cooler does its job from the top, through a heat spreader, this second sensor's more sluggish result provides important information about stability and cooling performance. Using Prime95, the CPU hits a whopping 105°C without any overclocking and in spite of us using our chiller!
This result demonstrates why you want to use Prime95 with AVX as a stability test for an extended period of time only if you're using monitoring software able to track all relevant temperatures.
In the end, knowing that the CPU (DTS) is cooled well doesn't really matter if the package underneath is burning up.
The memory load varies quite a bit between tests, as well. Here are all of the measurement results in one bar graph for direct comparison:
MORE: Best CPUs
MORE: All CPUs Content