> It should also be noted that K6's are rumored to be less stable under
> Linux than Intel processors (there was a website on this very subject,
I have zero evidence of this, and several fine counter-examples.
there were problems with initial revisions of the original K6,
which were fixed, and haven't been made for some time. there has
apparently been a recent revision to the cache-policy code on
higher-clocked K6-2's, but noone seems to know the "skinny" on it.
> As for K6-2,3 single processor and K7 SMP support, the x86-compatability
> in the chips should keep things in line with current kernel development.
sure, except for K7 SMP, which cannot be Intel-compatible, unless
AMD solves some licensing problems.
> Many of the patches have been released for enhancements and optimizations
> to specific registers in the K6-2 and K6 as well as to make proc/cpuinfo
what optimizations would those be?
> and binutils behave (at least that's all I've been able to notice with my
> experience tinkering around with them).
yes, HJ Lu released binututils with 3dnow opcodes recently,
and they work nicely. I was disappointed to find that the prefetch
instruction made only a minor difference to Stream performance
(large memory bandwidth).
regards, mark hahn.
