Correct, but that's the problem IIUC. The CPU's clock speed got cut by 50% during the thermal throttling, and then increased after the CPU cooled down, however QPF kept happily returning the same value the entire time. Google on SpeedStep and QPF.
Right, ok so we have different problems here : QueryPerformanceCounter: - pros: no need to link to additional libs, lower overhead - cons: might bug with CPU's clock speed changes timeGetTime: -pros: should always have the same accuracy -cons: requires linking to winmm.lib I think I'll try to do two windows implementation for the template parameter of the timer class, one with QueryPerformanceCounter and one with timeGetTime which would try to #pragram comment(lib,etc) it. The question that comes is that the design proposed until now poses a logical problem when it comes to the implementation: If the template param is a clock class, then using QueryPerformanceCounter/timeGetTime etc is a problem as it's not really related to the current time. I'll think more about it and propose something. Philippe