dear darrel

thanks a lot for your message. alone the commitment "it must work" helps me a lot. i tested now a bit and it works fine! unfortunalety i don't know why... it seems that i had somewhere a "magic number" which i've deleted once ....

also thanks for the instruction execution time link, i know them ;-)

probaply i was not really clear : my problem is not "how many cycles i need" because i'm knowing that, my problem is "how long is acycle in realtime"?
menas the instruction time on a 20mhz is 200ns. if i repeat them 5 million times, i need a second. but when the instruction time is only 199ns (only a shift of 0.5%) i get not 1.000s.

so when i make a long time measurement for my ten minute timer, i should see if there is a small shift. that's what i need to check. so the idea was to send out a serial command at start sequence and send anotehr at end sequence. in theorie, this are xyz cycles equal to 10 minutes at the microcontroller. but my tests are around 50ms too short. ok, not the world. but when i want to build a timer for videoframes (25 frames a second) it's one frame away...

so in fact that not ever get the same result, i first of all search for a exact method to measure before i want to calibrate. does anyone know how this is done at the olympic timer example?

thanks a lot