On Wednesday 19 March 2003 12:17 pm, Tom Brinkman wrote:
> On Wednesday March 19 2003 10:23 am, Per Øyvind Karlsen wrote:
> > why are everyone so obsessed about "optimizing" their packages? in
> > most cases it only gives marginal perfomance increase and often
> > introduces weird bugs and performance penalties..
> >
> > here's a little example:
> > http://www.mandrakeforum.org/article.php?sid=2472〈=en
>
>      That's been my experience.  The only 'speed' benefit compiling
> for athlon is pyshcological ;)   Building for i686 Intel proccessors
> can actually be a step backwards.

Well, I have run tests on the same athlon hardware, and I can report a 
significant improvement. In fact in many cases it is a 2x to 4x or more 
improvement, especially when the ENTIRE system is recompiled. If you cannot 
see that level of difference, OK., I can accept that. If you do not get that 
level of improvement, OK. I can see that too. To each his own. I have 
measured the improvement here, even on athlon 850s.

The nice thing is that this is a one time change, and using -mcpu i686 instead 
of -march will NOT affect users of lower/earlier "grade" processors. How can 
anyone complain about that. (Although I fully expect to see some complaints. 
- grin-)

Bob Finch


Reply via email to