FC2 based on Athlon architecture....

[gabolander]

Hi all.
I'm recompiling all FC2 SRPMS in *.athlon.rpm format. Anyone interested by that ?

Or just, does anyone know a site or a link that makes available packages compiled for Athlon architecture already ?

Anyway ... if I receive answer from more than one (or two :p ...) persons, as soon as my Barton finishes to compile and pack, I will prepare a webspace and make them available.
Make me know!

;)

Gabolander

[crackers]

Is this actually going to buy you anything (beyond the lovely experience of doing it)? It's my understanding that the 2.6 kernel automagically configures itself at runtime to use the appropriate architecture. Anything else beyond that is not likely to give you any major performance gains...

[gabolander]

Originally posted by crackers
Is this actually going to buy you anything (beyond the lovely experience of doing it)? It's my understanding that the 2.6 kernel automagically configures itself at runtime to use the appropriate architecture. Anything else beyond that is not likely to give you any major performance gains...

... so let me understand: are you telling me that recompiling with --target=athlon all stuff and reinstalling them, don't make me give any performace gains at all ?
Pleaze answer me, 'cause maybe I ignore this matter.

:confused:
Thanks.

[crackers]

You might get a little bit, but I don't think you'll actually "see" (from a normal user point of view) any difference. Where it might make some difference is in processor-intensive tasks, but I think (notice I said think) it'll be less than 10%.

What I would suggest is try a couple of packages and see if there's any difference between the "normal" packages and your re-compiled ones. I'm always quite willing to be proven wrong...

[gabolander]

Effectively I grant you that in the greatest part of cases you are right. You may see the difference in applications that peform certain kinds of operations. Further, the optimization also depends by the capability of the compilator to make well-optimized executable for an architecture too. In most cases, anyway, I was able to see more difference (note that is my impression) between a i386 and a athlon package on a athlon xp system, than between a i386 and a i686 package in a p4 system.
Anyway, if I can, I'm much happier to install a .athlon.rpm than a .i386.rpm of the same package ;)

Thank you for your reply.

Gabo

[mars_hall]

Let me know where you post the files. I wouldn't mind giving them a try :)

[crackers]

Originally posted by gabolander
I was able to see more difference (note that is my impression) between a i386 and a athlon package on a athlon xp system, than between a i386 and a i686 package in a p4 system.
Anyway, if I can, I'm much happier to install a .athlon.rpm than a .i386.rpm of the same package ;)

Thank you for your reply.


Well, then I was (apparently) wrong. Of course, I'm a lazy-ass bastard when it comes to maintenance, so I stick with the "stock" packages from apt-get repositories -- works more than good enough for me.

For NVidia, though I do make an exception. (/me taps foot impatiently...)

[helder.meneses]

hi there people...

i have little experience in linux, have tried almost all the distributions that really works, and i made some conclusions. one of them is that athlon optimized kernel is 30% faster that pentium kernel. an example is the suse professional 9.0 that have k_athlon.rpm optimized kernel and i noticed that things were much faster.

i would like to try the athlon rpm's in my laptop 1600+.

gabolander, am i the third person?

please put whe rpm in a webspace... :)

[saBrEwolf]

I'd be interested in an atlon only version of Fedora.. if your still gonna do it?

[ghaefb]

I'm interested too :)

I rather see .athlon.rpm than i386.rpm :p
Coun't me in.

[imdeemvp]

the current rpm's packages work great in my amd athlon xp 2800+

[Deathshroud]

This sounds promising. And I'm in the process of experimenting with my Fedora install. So count me in.

[FreeEagle]

i am in waiting the Pakages that you will compile, try to informe me ...... i can not wait to install them, well i use AMD Athlon XP+Barton Full Cash 2600+.
wish you all the luck in your mission.


FreeEagle

[saBrEwolf]

the current rpm's packages work great in my amd athlon xp 2800+

Yeah, they work fine on my Athlon XP 1700+ also.. but any speed increase is good, right?

[Jman]

athlon optimized kernel is 30% faster that pentium kernel. an example is the suse professional 9.0 that have k_athlon.rpm optimized kernel and i noticed that things were much faster.

Notice that this is the kernel package. From what I've heard recompiling it and maybe glibc will get you the best performance gains, but for most other packages it won't make much of a difference, like crackers said.

[imdeemvp]

Yeah, they work fine on my Athlon XP 1700+ also.. but any speed increase is good, right?

that sure will help the cpu...... :D

[Ned]

Notice that this is the kernel package. From what I've heard recompiling it and maybe glibc will get you the best performance gains, but for most other packages it won't make much of a difference, like crackers said.

I've recompiled my kernel quite a few times (different kernel versions) and optimised it for speed including for athlon and I've never noticed any real difference. That being said, I was really dissapointed when Fedora droped the athlon.rpm kernel package with FC2.

I guess I'm just a sucker for this sort of stuff, so count me in too :D

Although I'm thinking it would probably be a lot easier to just go with something like gentoo where you compile everything anyway and can set the archetecture and optimisation flags you want.

...and here's another thought - how about a "Linux from Scratch" type procedure to compile and install the base OS using the latest Fedora sources and then use rpm to install the rest that won't benefit much from athlon optimisation.

Ned

[saBrEwolf]

that sure will help the cpu...... :D

Hey! I've only just turned 17 and have a none-exsistent income (Apart from EMA) so, it was the best I could afford at the time. Leave my CPU alone!! Lol

[Sniffer]

Put me on the list.

ATHLON XP 2.6GHZ Here.

Where is the Link..... :)

[Ned]

Hi,

I recently recompiled the SETI@Home DC project Boinc program which has built in Whetsone and Dhrystone benchmarks. This makes it a particularly good model for examining the effects of compiler flags on mathematical functions. I thought I'd share my finding here as it's sort of relevent to this discussion.

Firstly, I was surprised to notice what little effect -march settings made, with absolutely no difference betwen athlon and athlon-xp. Also, -march=i686 was actually _faster_ for the Whetstone benchmark on my AthlonXP system. Further, a lot of compiler flags touted around the web as being beneficial were actually detrimental and I actually settled on quite a minimal set of compiler flags for my optimal settings (I've actually benchmarked over 20 different compilations now). Incidentally, I also saw no difference when running the benchmarks on a standard FC2 2.6.8 kernel versus a highly optimised compiled kernel for athlon processors.

How much effect this would have when compiling the complete system I don't know and it would be almost impossible to benchmark in a non-subjective manner. I think a small effect on things that use mathematical functions extensively like compile time (if you do a lot of compiling), compression software (gzip etc) may be present but I'm not convinced real world apps would really benefit noticably.

Interesting stuff though :)

Ned