Different methodologies I suspect. Interesting to note that Garmin refer you to training peaks for an explanation - support.garmin.com/.../case.faces
But that still does not go anywhere to explain why the difference. You need to get hold of the underlying algorithms in order to see the difference. TP seem a little more open in that respect.
Except that I would have thought that Garmin would have licensed from PeaksWare not only the trademarked terms TSS, NP and IF, but also the underlying algorithms for computing those metrics. I would have thought this ... but evidently I would have been incorrect in thinking this.
For those interested - it was because I was excluding zeros from power averaging on the Garmin 910. Have changed that now and it all matches very closely.
Just for interest really, although the algorithms may be the same the information transferred might be different depending on what is passed. Just done some research with 910s and 310s and found the distances recorded by GPS were consistently higher than a measured wheel (from 60 GPS samples). We would have anticipated a normal distribution either side of the mean wheel measure. We've not analysed the csv files yet but there may be some indication in there.
As I said, interesting but not necessarily relevant
