debian i686?

[swim]

Hi, I wonder if anyone has seriously considered making packages for i686, seems since debian supports so many this one would be a good idea. I actually do find i686 optimized systems run noticably faster, and benchmarks agree.

[gabriel]

Actually I don't understand why Debian is not i686 optimised? I mean, it would work faster, right? Besides, older architectures are becoming obsolete... imho. Correct me if I'm wrong.

[jomamma]

it would probably not be noticeably faster.Tthere are a few packages which actually benefit feature-wise; we have special packages for those, and d-i install them automatically.

[Guest]

I tried lots of distros but I think Debian is really good , but unfortunately it is i386 optimized so I recompile the kernel with i686 optimization and now my machine is faster.

My machine has a i686 optimized kernel so I was wondering:

- How faster could be my Debian if it could run i686 optimized packages?
If it could be noticeably faster then now I could recompile step by step all
packages but it will take me really much time cause I have an old laptop
and I can't typr "apt-build world" in this way I risk to "fry" my laptop.

- Suppose that I could install on the same machine a Debian with kernel and packages optimized i686 and a Gentoo compiled from scratch with medium optimization cause a strong optimization could destabilize the system.

Do you think there are differences in speed or not - Everybody has a
different opinion but I'm intersting in knowing the truth specially cause I can't install a Gentoo from scratch.

- I tried Arch linux cause it's i686 optimized and I don't have to spend days in compiling. Do you think Gentoo could be faster?

P.S. I was googling and I read that there's a russian project named "Debian i686 project"... however I think Debian is really good if only there was a release for i686 it would be sure the best than ever.

Thanks by

deadlinx

[domecq]

"Some users wonder why all this effort is put into porting to sometimes rather obscure or obsolete architectures when that effort could be put into improving the PC version of Debian. Even if all you ever use are standard Intel processor-based beige boxes, you benefit from porting. The process of making software work in environments other than the ones they were designed for reveals hidden assumptions and hard-to-find bugs that would otherwise have gone unnoticed. Currently, Debian projects officially supports these architectures:
i386 - Intel's 80386 and its descendants and clones. Except for the kernel and a few packages where such optimization would actually make a difference, Debian doesn't optimize for Pentiums, AMD Athlons, and the like. Should a user wish to so himself, he can recompile, but there isn't enough benefit to do this on a global basis."

Source: Debian GNU/Linux 3.1 Bible - Benjamin Mako Hill, David B. Harris, and Jaldhar Vyas, page 7.

[harpo_marx]

"Some users wonder why all this effort is put into porting to sometimes rather obscure or obsolete architectures when that effort could be put into improving the PC version of Debian. Even if all you ever use are standard Intel processor-based beige boxes, you benefit from porting. The process of making software work in environments other than the ones they were designed for reveals hidden assumptions and hard-to-find bugs that would otherwise have gone unnoticed. Currently, Debian projects officially supports these architectures:
i386 - Intel's 80386 and its descendants and clones. Except for the kernel and a few packages where such optimization would actually make a difference, Debian doesn't optimize for Pentiums, AMD Athlons, and the like. Should a user wish to so himself, he can recompile, but there isn't enough benefit to do this on a global basis."

Source: Debian GNU/Linux 3.1 Bible - Benjamin Mako Hill, David B. Harris, and Jaldhar Vyas, page 7.

btw I read the book and it is not good just an average(oh btw when u make a statement as they did they really should test it and give us some hard data on their theory)

[onemyndseye]

ok ...think it about it for a second.


Recompiling on the kernel level make a differnce for obvious reasons...


But how much good do you think recompiling 'ls' would do?.... none - since it doesnt have direct access to the hardware. It only sees through the kernels "eyes"..... and such is the case with the majority of packages in the repo's


Media intensive application have seen some public recompiles over the last few years... like mplayer, xmms, firefox but I really think one could still question the validity of such based on the statement made above......

For Example: The kernel knows NOTHING about decoding MPEG4. Hence recompiling Xine or mplayer with special intructs for decoding MPEG on i686+ offers some speed increases.

[sinical]

Think about this for a second

Posted: 01-10-2006

Posted: 22-06-2007

[chrismortimore]

You know, Debian supplies a 686 kernel. And Debian Multimedia supply a 686 optimised version of mencoder.

But having used Gentoo for ages with all sorts of weird and wonderful optimisations (and then reinstalling with sensible ones), I can't say it has ever really been worth it. With an optimised system, mencoder was running at 52fps instead of 51fps. Given that it took 10 hours to install the system, I was worse off in the end

I've also used Arch extensively, and I agree that it is fast. But I'm certain it is fast because it cuts out a lot of crap you don't need, rather than because they use slightly different CFLAGS.

And strangely enough, Debian boots the fastest, and runs mencoder the fastest out of all these three distributions that I've mentioned. And Gentoo was always the slowest to start (by about 20 seconds), regardless of what CFLAGS I used.

And by the way, I'm quoting mencoder as my benchmark because it tells me how many fps it is doing, it cobbles all of the CPU time, the only bottleneck it suffers from is the CPU, and I use it a lot.

[perlhacker14]

You know, I heard about the so called benefits of optimisations deriving from compiling for source all programs from source, and I went and compiled nearly all essential utilities (bash, ls, cd) and a few other things like KDE and GNOME, not to mention the Debian base system. I recieved a marginally faster response time from programs (measured through a system monitor and timer), but nothing grand enough to justify the time and effort of compiling (.9% less CPU usage and nearly the same RAM usage).

[swirling_vortex]

Unless every Linux application is written to use all of the features of the i686 architecture, you won't notice any difference other than a slightly quicker startup time. What impacts speed is:

1) the number of services you're running
2) the amt. of memory each service needs
3) your desktop manager (kinda goes back to 1 & 2)

Usually, only the kernel and maybe some heavy duty applications have to be optimized in order to notice a difference. Optimizing is more hype than anything else.

[plugwash]

Optimizing is more hype than anything else.

A little word of warning: your statement is somewhat true on the PC architecture because PC processor manufacturers put an insane ammount of effort into running code intended for older PC processors at high speed. MMX and SSE can still make a noticable difference in the right apps though.

but don't make the mistake of assuming its true in general. I was perusing the debian-arm mailing list recently and i've heared mention of 50x (yes thats times not percent) speedups of floating point heavy software on some hardware from moving to the new eabi port which uses software floating point in the libraries rather than using kernel based fpu emulation.

[llivv]

but don't make the mistake of assuming its true in general. I was perusing the debian-arm mailing list recently and i've heared mention of 50x (yes thats times not percent) speedups of floating point heavy software on some hardware from moving to the new eabi port which uses software floating point in the libraries rather than using kernel based fpu emulation.

Just curious, but would that be related in anyway to the IBM/Sony developed cell processor as well?
The one in the Playstation is said to poduce a mere 2.2 (off the top of my head) tera flops
for perspective: again off the top of my head,
The Gray T3 circa 2000 at a cost of around 10 mil produced a wapping 3.4 tera flops