I agree their is no easy extrinsic method for validating DTS accuracy, but you can simply do a paired t test on his data and show sensor error is statistically improbable to scientific standards of 95% (p<0.05) or even 99% if more runs (but not necessary). I had same question in past, and ran 10 tests of same cured tim at different ambients, and no question i7 DTS sensors are capable of .1 to .2C relative accuracy within above tested range and are linear within a few degrees C. Even on his 3 runs here of each, T paired assuming null hypothesis, comparing 3 feser one cpu temps to 3 distilled water cpu temps their is a 98% probability that a difference is significant and not do to random chance, ie same 98% chance not due to random sensor error. Doesnt rule out systematic error (other controls for that), but that alone to a scientific standard eliminates DTS error, and you can show 95% confidence that error DTS is less than 0.2 and i dont feel like redoing skinnees math, but his cpu error is likely around .125.
I guess you were looking at the discrepancy between GPU results and cpu, for example minnesota tap. Some of the gpu sensor data had SD of .4 vs cpu all were near .1, so perhaps GPU data/sensors have higher error causing slight discrepancy. But always some screwed data points in any testing.




Reply With Quote

Bookmarks