Re: speed on amd64
- From: Janwillem van Dijk <xyz.van.dijk@xxxxxxxxx>
- Date: Sat, 02 Sep 2006 17:26:14 +0200
Richard Engebretson wrote:
Fat Ass Fred wrote:
As others suspected your optimizations of your code may be the
culprit, but as well you mention not how your Linux installation was
configured.... Did you install anything optimized for the AMD 64? A
Linux installation without making any changes in some cases are
optimized not even for a Pentium but for a 80486 or 80386 CPU. Some
optimizations are optimized for Pentium Pro, etc. That's why one
particular version of Linux (can't remember the name, think its Gentu
or something like that) re-compiles ALL source code for the entire OS
(kernal and all) after downloading all updates, such that everything
on your system ins 100% optimized for your particular CPU! If your
CPU isn't optimized then every time you tell it to print, its going to
run slower, though if your sorting or generating random numbers, you
should be dealing with more number-crunching power and not so much
with output, though if you print all 1,000,000 numbers, you leave much
room for calling the OS to print, and the time could be because of the
OS..... Though I suspect your printing the time BEFORE you output all
1,000,000 floating point numbers....
I agree. Please see my earlier posts. Linux is not like win32 on a PC.
For beginning linux developers, SuSE is best. Even for an old linux
developer wanna-be, SuSE is great.
I still suspect you are looking at the wrong timer, or have other tasks
competing for the cpu. Linux is not DOS.
I am not writing a deamon for which it might be important how much time the cpu needs. I am writing Monte Carlo applications (numerical integration and that kind of stuff) and I am interested in how much time I (the human being) have to wait until results appear. This is just what t:=now; do something t:=now-t; does like using your wristwatch (in Turbo Pascal 4 already when I am not mistaken).
In this case we are not looking at a few % slower or faster but at a factor of 2 or more. If I have a look at system load (using e.g. top) before and after I run the test I have between 1% and 2% cpu load from all running processes (on Win2k also by the way). So it cannot be that the problem is in the multi-tasking.
I am very well aware of the fact that Linux is not DOS just as well as I know it's not CPM, MPE, AIX, AOS, MPOS, OS/IBM360 to name a few I worked with and that I am not behind a punched card machine anymore (grumpy old man eh?).
In my opinion either I am doing something fundamentally wrong or there is a fundamental problem with the AMD64 fpc implementation or the way Lazarus makes use of fpc 2.0.4.
- Prev by Date: Re: speed on amd64
- Next by Date: Re: speed on amd64
- Previous by thread: Re: speed on amd64
- Next by thread: Re: speed on amd64