1 |
On Fri, Feb 24, 2012 at 03:13:07AM +0200, Nikos Chantziaras wrote |
2 |
|
3 |
> The speed gains of building for specific submodels of CPUs might |
4 |
> be there, but they're minimal. Benchmarks have shown (can't find |
5 |
> the article, it was on Phoronix) that after -march=i686 you get |
6 |
> diminishing returns. |
7 |
|
8 |
In that case, the benchmarks are useless. From my personal |
9 |
experience... a fresh i686 install on a 4 and 1/2 year old Dell with |
10 |
onboard Intel GPU was not able to keep up with the slowest available |
11 |
speed on NHL Gamecenter Live. Ditto for 1080i TV from my HDHomerun |
12 |
tuner box. After rebuilding system+world+kernel with "march=native", |
13 |
it works just fine for the above tasks. I'm not the only one to see |
14 |
this. See thread... |
15 |
"Slow not in sync movie playing with mplayer2, ffmpeg, x264 with intel core i5" |
16 |
starting Sun, 12 Feb 2012 on this list. |
17 |
|
18 |
As I mentioned in that thread |
19 |
> Optimizing one library may seem very minor, but it all adds up when |
20 |
> you optimize every library on your system. |
21 |
|
22 |
To get the full benefit of optimization, you need to optimize your |
23 |
entire system. The i686 code used for the install CD has to be generic |
24 |
lowest-common-denominator i686 code, in order to run on every 6-year-old |
25 |
i686 cpu out there. The tradeoff is that you lose the benefits of |
26 |
optimisation. |
27 |
|
28 |
-- |
29 |
Walter Dnes <waltdnes@××××××××.org> |