- Jun 30, 2004
- 16,725
- 2,087
- 126
AigoMorla, in some recent thread, seemed to speak with authority about the placement of the four cores in a C2-Quad processor. I think he said that the cores followed this pattern:
0 1
2 3
Suppose CoreTemp reports temperatures for Cores 0 through 3 in this order:
57, 51, 55, 52
Sampling these temperatures over a one or two-hour double-ORTHOS "SMall FFTs," we find that there is little variation for each core. In other words, there are pairs of adjacent spikes for each core in the frequency distribution of temperatures, and the low-frequency values between the temperatures for the spikes are far and few between -- almost "outliers" for any given core. That is, looking at the log-file of sampled temperature readings, one would almost think that 57 was a rock-solid and unvarying temperature for core 0, etc.
The deviations between the cores seem significant and suggestive of a pattern: 57 varies from 51 by 6C-degrees; 55 varies from 52 by 3C degrees.
In speculating about a possible cause, I remembered how difficult it was to spread IC Diamond thermal paste; how I used it so sparingly that you could almost see the heat-spreader cap through the paste in certain places. So I imagine a possibility that the thermal paste layer is insufficient or unevenly spread, or that there are air-pocket irregularities in the thermal-paste layer.
Another possibility is "processor affinity" of various services running under the OS, or rather that some processes are more processor-intensive than others, and that the software design simply causes those processes to grab certain cores during system boot to Windows.
Yet another possibility is a variation in temperature sensor calibration or accuracy among the cores. Keep in mind that some of my speculations may not be fully informed, but I imagine that they are possibilities.
Are there any useful insights into this phenomenon? I may soon take a convenient opportunity to remove the heatsink, apply new TIM and reinstall the heatsink. Convenience is a matter of an absolute need to do something else.
0 1
2 3
Suppose CoreTemp reports temperatures for Cores 0 through 3 in this order:
57, 51, 55, 52
Sampling these temperatures over a one or two-hour double-ORTHOS "SMall FFTs," we find that there is little variation for each core. In other words, there are pairs of adjacent spikes for each core in the frequency distribution of temperatures, and the low-frequency values between the temperatures for the spikes are far and few between -- almost "outliers" for any given core. That is, looking at the log-file of sampled temperature readings, one would almost think that 57 was a rock-solid and unvarying temperature for core 0, etc.
The deviations between the cores seem significant and suggestive of a pattern: 57 varies from 51 by 6C-degrees; 55 varies from 52 by 3C degrees.
In speculating about a possible cause, I remembered how difficult it was to spread IC Diamond thermal paste; how I used it so sparingly that you could almost see the heat-spreader cap through the paste in certain places. So I imagine a possibility that the thermal paste layer is insufficient or unevenly spread, or that there are air-pocket irregularities in the thermal-paste layer.
Another possibility is "processor affinity" of various services running under the OS, or rather that some processes are more processor-intensive than others, and that the software design simply causes those processes to grab certain cores during system boot to Windows.
Yet another possibility is a variation in temperature sensor calibration or accuracy among the cores. Keep in mind that some of my speculations may not be fully informed, but I imagine that they are possibilities.
Are there any useful insights into this phenomenon? I may soon take a convenient opportunity to remove the heatsink, apply new TIM and reinstall the heatsink. Convenience is a matter of an absolute need to do something else.
Facebook
Twitter