> This is because the 3-D object modeling scheme used in Quake makes HEAVY
> use of the FPU. When you're doing large memcpy()s and such, it starts
> competing for access to the FPU. This is where your problem is at. If you
> can live with the extra bit of choppiness, go for it. I'm still on 2.0.27,
> and last I played Quake (Pentium 90, 16 MB RAM, S3 868 vid card w/2 MB
> RAM), it was not noticable to me, and I do use the Pentium memcpy patch on
> my box.
I suspect that the problem is more specific to quake than other
FPU-intensive applications. I think it's only supposed to do the special
memcpy if it would win in all cases, even with the FPU save/restore step. I
suspect that these calculation don't take into account the fact that memory
on the ISA bus will go no faster using the FPU (if I'm not mistaken. I
don't know as much as I'd like to about mb design).
Greg Alexander
http://www.cia-g.com/~sietch/