> Each cpu is scaled individually based on only its total tics. But I believe that's not correct. If all those 20 ticks are 'user', then the result would be 100% user and 0% idle, whilst in fact the CPU is 80% idle ... > Assuming nothing strange happens when a Nehalem core is revived, I > expect the % distributions to be spread among user, system, idle, > etc. We really cannot rely on that ... the examples from the Fil's email were just ilustrative and nobody can guess what values can or cannot really appear on systems like that. In his case there were too low total ticks per CPU, but that doesn't mean that higher values can't appear. You mentioned, that the 1st line is not affected by C-states at all. Does it mean, that it could be used for calculation of the 'maximum increment' ? In such case we should scale the values against this 'maximum increment' and that should be sufficient and correct. > > J. J.