In article <201dvt$jq8@klaava.Helsinki.FI> torvalds@klaava.Helsinki.FI (Linus Torvalds) writes:

quoted text
>   In a short term, the pentium is probably the way to go: intel reportedly
>   got reasonable fp performance out of it, and "porting" linux shouldn't
>   be a problem.  As to supporting new processors (the 68k and possibly any
>   sparc chips etc), I'm happy if someone can port it, but I don't have
>   access to the hardware, so...  I don't think DEC is going to give me an
>   Alpha machine just to let me play around with :-p

Well, nearly all Sparc--systems and Alpha--systems are sold with and OS, ...
Porting Linux is not as necessary as it was to PC--architecture.

quoted text
>   How many linux users are actually doing the kind of calculations where a
>   486 isn't fast enough, but a HP-PA or an Alpha would be?

FE--Problems, 2 2/2D-- and 3D--CAD benefit most from pipelined, vector like
matrix operations. For large problems of this kind a cache offers no value
at best, a severe penalty at worst because of thrashing.

As you have to shell out about 150.000 DM to get approbriate systems from
HP, you will discover, that you have bought indeed memory and disk drives
at a multiple of the street price just to make the needed CPU run .....

So you may add many engineers and scientist to those for who a 486 is
NOT sufficient, and would like a cheaper, but capable system.

quoted text
>I think the
>   discussions about fast video chips are due to the fact that more people
>   care about window updating speed than the speed of transcendental
>   functions..

Transcendentals are not the problem, vectors and matrices are.

quoted text
>That said, I've tried to make linux use the coprocessor as
>   efficiently as possible (lazy state save/restore etc), but I doubt it
>   would have mattered for most people. 

Right, but one should admit, that there are problems a 486 or Pentium is NOT
APPROBRIATE.

best,adams