64bit Gentoo not that much optimizable?

[ial]

Reading this thread:
https://forums.gentoo.org/viewtopic-t-772298.html

[loki99]

These are just my 2 cents but here I go.

There is no need to add packages to keywords to a stable amd64 install unless you want some apps that are stil keyworded. Most people just don't want to wait when they find out that a new version is available and so they add a few packages to their keywords out of curiostity which may pull in other apps that are keyworded. But this is same no matter which arch you choose, so if you want it stable you can easily have it stable.

And the speed difference that Gentoo can achive is not due to uber optimization of cflags and the like but to the possibility to only install the things you really want and need while other distros tend to pack as much features as possible into any application, trying to please everyone. For example:

[ial]

[Hupf]

Different compiler flags will, on a global scope, only increase performance by ~3% *maximum*. Much more important is e.g. the use of SIMD extensions (like SSE), preferably through -march=native. Also, switching to 64bit will allow for more processor registers (greatly improving performance by itself) and simplify memory management (no PAE needed).

Basically, setting C(XX)FLAGS="-march=native -O2 -pipe -fomit-frame-pointer" should be more than enough for performance. Anything else will be more likely to heavily break things (yes, even -O3 may lead to programs hanging etc.). In my experience, disabling unneeded USE flags is much more of a performance giver than any compiler flags.

[ial]

Let's put things into numbers... Please assess (approximately) by how much % speed will be increased because of 'natve' flag, 64bit processor registers, no PAE etc. and by how much thanks to USE flags?

And also let me ask whether performance boost from USE trimming is about CPU and executing program's code itself? Or do you rather mean other types of overhead like loading programs/libraries into memory etc.?

BTW. Are binary distros compiled with the 64bit C(XX)FLAGS (like '-march=native') and do they utilize 64bit registers?

[Hupf]

[ial]

[tgR10]

i got both x86 and x86_64 on 1 pc
switched to x86_64 like month ago or so, i used amd64 like year ago but it wasn't stable during work as it is now
i don't have any benchmarks i just don't go back to 32bit
the only thing i noticed that on 32bit system cfq scheduel works much better than on amd64 for me, on 64bit i'm using anticipatory, also i did change preamption kernel to high latency desktop and rtc to 1000hz(i had 350 or so on 32bit) and i don't have any random half sec or a sec mouse locks none of those :)
and im using ~arch both
so what i can say - you should try it, 64bit grown up a litle since last time i used it
that's what i'm using

[i92guboj]

[ial]

[kernelOfTruth]

who says ~amd64 / 64bit isn't stable ? I've made the experience that ~amd64 is much more stable and working than the so-called "stable" amd64

and who says that optimization on ~amd64 isn't as much possible as on x86 ?:



try that:

[loftwyr]

Wow a huge list of redundant flags. Most are default and most are simply overridden with -march=native.

Nice work.
_________________
My emerge --info
Have you run revdep-rebuild lately? It's in gentoolkit and it's worth a shot if things don't work well.
Celebrating 5 years of Gentoo-ing.

[krinn]

[tgR10]

[Hupf]

[kernelOfTruth]

[ial]

[stmiller]

This is an interesting unscientific poll:

https://forums.gentoo.org/viewtopic-t-774221.html

[kernelOfTruth]

[ial]

[tgR10]

[energyman76b]

try this: download vegastrike

once compile it with no cflags.

then compile it with -enable-release=3 --enable-flags=-march=amdfam10 -O3 -msse3 -msse4a -mabm -pipe --disable-debug

or something like that.

Difference in fps up to 20%. Depending on the situation.

I started with self compiling stuff when an optimized glibc,X and xine meant I was able to watch dvds on my k6-2 400 back then...

so with the right choice and the right packages some nice increases can be achieved.

BUT: no matter which flags you use, slow hardware will always be slow hardware. You might be able to make it less bad, but never good. Most distris turn on -O2 anyway, but skip march and sse flags. Depending on the app/lib that can mean no difference at all or a nice boost. Stuff like mplayer sets its own flags - partly for some very good reason. Optimizing expat will probably don't do much, or bison or vi or nano.

So it depends.

USEFLAGs don't do much for the running code - but the smaller the binary, the less linking on startup, the faster it loads and the more 'quicker' feels a system. Also, less crap, more room for your own stuff. Sure, harddisks are huge - but movies are huge too
_________________
Study finds stunning lack of racial, gender, and economic diversity among middle-class white males

I identify as a dirty penismensch.

[devsk]

[durian]

[energyman76b]