XXI century and still using 386 packages...

[goofrider]

Alberto Núñez Wrote: 
> I suppose I'm not the first that comment this... 
> We are in 2005, with new technologies and amazingly fast systems...
> and we are using prehistoric system aimed packages. How many people
> does use a 386 desktop computer? How many people use a 486 desktop
> computer? Even Pentiums, nobody surf internet today with a pentium
> (well, someone...)


The reality is that the instruction set of the 686 family is virtually
identical to good old 386, with minor additions like a few extra
instructions (like -cmov-), power management, and SIME/MMX
instructions/registers, etc. For normal applications and system
services that doesn't use SIME/MMX, there's very little they can be
optimize for 686, with the exception of things like instruction
pairing, compile-time branch/loop optimizations and word alignments. I
believe that most of these optimizations are already applied by GCC by
default as long as they're backward-compatible with 386. The only types
of 686 optimizations you're missing are primarily new instructions like
-cmov-. (Disclaimer: IANAP, so I might be talking out of my ass.)

The fact that some distributions are labelled 686-optimized doesn't
really mean they're any faster in terms of real world performance. Most
of those packages are probably virtually identical to their 386
counterparts at the binary level.

Also bear in mind that there are modern x86 CPUs that might NOT be
686-compatible, like Via C3/EPIA, AMD Geode and Transmeta Crusoe. Add
the various embedded x86 router platforms, and all the people who run
Linux in an emulator like Virtual PC, QEmu, Bochs or the upcoming
Xen.... I guess these should be enough reasons to keep 386 as the least
common denominator.


-- 
goofrider