>   It could also be your cache... the cache keeps both instructions and
> data.  to test this, you might be able to try nd hit the same memory
> repeatedly with reads and writes (the same reads adn writes of course,
> changing them every so often, and you'll need to do it in a fairly
> sizeable chunk).

Thank you for the idea. I tried it, and my cache is also O.K.

Tonight I will dissassemble and debug primes.core.

I talked with my father, who is a remote controlled fly enthusiast,
and he told me about a similar problem, which was caused by oxidizing
of old connectors. I will check that too.

Kim0