Is the speed difference between an i386-optimized distro...

yarjar · 2004-07-27 06:12:07

...and an i686-optimized distro noticeable on a really high-end computer (Pentium 3.2 GHz)?

Bobonov · 2004-07-27 09:25:53

Yes there are.

This is a very simplistic explanation so also very newbe can understand, there forgive me for the naive language.

Basically i386 processor compared to an i686 has a very limited set of instruction, so that mean for particular i686 instruction an i386 has to execute more instructions to obtain the same result of 1 instruction on i686.

If you compile for i386 even if the program run on i686 processor it will execute more instruction (because it contain them) than an optimized executable for i686.

Having an executable for i386 means also that it "contains more instruction" so it will use more ram to execute.

At speed execution level there are very discordant opinion.
There are people that say there are no difference, other says that there is a gain of 5-10%, few say that the performace are worst. Personally I think depends on the kind of use.
The problem is that is very difficoult to make a precise comparison.
To be sure you need 2 twins machine and use exaclty the same distribution, one compiled for i386 and the other compiled for i686 with exacly the same installation and configuration.
Also in this case depens on the kind of application you use.
For instance i686 has a subset of instruction for "multimedia" so probably if you use application that do a large use of these instruction the speed gain is more evident than if you compare vim use.
On very hig end machine this difference should become less relevant since the execution time became more little.
Example: on my 500MHZ processor an i386 program take 10 minutes to perform an operation so the equivalent i686 should take 1 minute less so 9 minute.

other machine, 10000MHz processor the same i386 program take 10 second to perform the same operation, so the i686 versione should take 1 second less that mean 9 second.
In both cases the gain is 10% but of course on shorter execution time the gain is less evident.

But thinkg to a lot of application running all togheter on the long time the sum of all the time gain make a difference.

So mostly depens on the use.
Anyway using i686 optimized code does not make think worse......

punkrockguy318 · 2004-07-27 21:01:01

Thanks bobonov! That cleared things up for me... I was also wondering that...

ravster · 2004-07-27 21:41:38

Cool. So that's what it is
Thanks for the explanation.

phrakture · 2004-09-20 19:54:07

nicely put Bobonov... and I must say your english is pretty good.

I'll simply add on my 2Mhz:
Multimedia extensions make a big difference in audio/video apps.
Simple mathematics and "Hello World" type programs won't notice any difference at all.
Larger apps should respond quicker: less instructions to load into memory

In short, why not optimize for i686 when 686 instructions include the Pentium 2 and up.... does anyone here even have an original Pentium still (that is being used!) - I know I have some 386s, 486s and Pentium/Pentium Pros in boxes somewhere, but I'm not using them, so I don't care.

kakabaratruskia · 2004-09-20 20:06:49

I know i686 has more instructions, but I thought risc was better than cisc, so in this case having less instructions would be better.

sarah31 · 2004-09-20 20:34:03

i could have told you yes without the big long explanation or even using a high-end chip example.

when i switched from Debian to Arch on my XP 1800+ (which was a mid high chip at that time) the speed difference was incredible.

having optimizations that allow your processor to be utilized in a more efficient way is always a good thing. (which is why i will wait until the next osx comes out before getting a g5 machine)

Xentac · 2004-09-21 05:49:40

kakabaratruskia wrote:

I know i686 has more instructions, but I thought risc was better than cisc, so in this case having less instructions would be better.

That's really the wrong comparison to make. We're comparing two cisc instruction sets, i386 and i686, on the same chip. The key is that one is optimized while the other isn't/can't be.

There's a significant difference between risc and cisc and risc usually does win out because of the sheer simplicity that the processors have, they're easier to optimize and make faster (run less hot, etc). But you can't trim down the i386 instruction set on an i386 compatible processor, so you've got to work with what you have.

Bobonov · 2004-09-21 06:41:27

sarah31 wrote:

i could have told you yes without the big long explanation or even using a high-end chip example.

Is there something wrong to give some explanations?
I assume that when people do a question is because they do not know the answer therefore they do not have the base/knowledge to reach the answer.
Just yes, does not give any value to the proposed question, spend 5 minute to write also why can be realy usefull for some people, as you can see from some answer.

For who wants some more about cisc/risc can read this article
http://www.osnews.com/story.php?news_id=3997

dpb · 2004-09-21 12:22:10

phrakture wrote:

In short, why not optimize for i686 when 686 instructions include the Pentium 2 and up.... does anyone here even have an original Pentium still (that is being used!) - I know I have some 386s, 486s and Pentium/Pentium Pros in boxes somewhere, but I'm not using them, so I don't care.

I got a Pentium 100mhz at my parents house that I use whenever I go there. And I know others who use them too.

Arch Linux

#1 2004-07-27 06:12:07

Is the speed difference between an i386-optimized distro...

#2 2004-07-27 09:25:53

Re: Is the speed difference between an i386-optimized distro...

#3 2004-07-27 21:01:01

Re: Is the speed difference between an i386-optimized distro...

#4 2004-07-27 21:41:38

Re: Is the speed difference between an i386-optimized distro...

#5 2004-09-20 19:54:07

Re: Is the speed difference between an i386-optimized distro...

#6 2004-09-20 20:06:49

Re: Is the speed difference between an i386-optimized distro...

#7 2004-09-20 20:34:03

Re: Is the speed difference between an i386-optimized distro...

#8 2004-09-21 05:49:40

Re: Is the speed difference between an i386-optimized distro...

#9 2004-09-21 06:41:27

Re: Is the speed difference between an i386-optimized distro...

#10 2004-09-21 12:22:10

Re: Is the speed difference between an i386-optimized distro...

Board footer