Changing the default 32-bit x86 arch for Fedora 12

[Warren Togami]

On 06/15/2009 08:31 PM, Tom Lane wrote:
> Bill Nottingham<notting at redhat.com>  writes:
>> drago01 (drago01 at gmail.com) said:
>>> Moving to i686 is fine, non i686 chips are mostly dead (but the
>>> perfomance gain from moving to i686 from i586 is questionable at
>>> best).
>
>> ... how so? It's consistently 1-2% in reasonable benchmarks (real-world
>> code, albeit cpu-specific).
>
> I don't understand how this proposal can survive even momentary
> consideration.  We're going to cut off some nontrivial fraction
> of our userbase to get 1-2% speedup for the rest?
>
> As was already mentioned, the people who need speed are probably
> on x86_64 already.  The x86 builds are for legacy hardware *now*,
> and should be understood as such.
>
> 			regards, tom lane
>

This sounds like a pretty good argument against changing x86-32.

I suppose I had liked the idea of SSE2 being the minimum to squeeze 
maybe a few % more performance, but then again I wouldn't even use the 
i686 SSE2 Fedora myself.  I have been using x86-64 for years where I 
care about performance.

No longer being able to upgrade my Athlon Thunderbird file server would 
be annoying.  I don't care about performance there.

Then perhaps a significant % of the LTSP thin client hardware would be 
impossible to upgrade beyond Fedora 11.

How serious of an effort would a second 32bit Fedora be?

Is that effort really worth a few extra % of performance for an arch 
where people don't really care?

Warren Togami
wtogami at redhat.com