Strange behavior on 1.7G Celeron
 

I recently purchased a P4 based Celeron MB+proc (SiS 645 chipset, DDR333 one bank) for use as a primer horse. I put a minimal system on it and had no problem getting mlucas setup for it. It's running 22.5 (most recent stable, right?) and saw something strange.

On the first run of a 16M exponent, I get a per iteration time of 89ms. After it completes and does the next similar sized exponent, I get a per iteration time of 113ms. If I stop it and reboot the machine, I get back to the 89ms time per iteration.

I've seen this behavior on my EV56 Alpha system with Ernst's code (hello, Ernst!) and it was atributed to page coloring problems--by Jason P.

I've never seen this effect on any other system, until now. Any ideas why it does this? Any suggested solutions? Is this normal?

System info: 1.7GHz Celeron, Sis645 MB DDR333 memory (256M in one DIMM). Stock RH7.3 install.