Re: [Cooker] 64 vs 32 bit Cooker

[Giuseppe Ghibò]

Stéphane Téletchéa wrote:

>> 2)  I read somewhere that there is little advantage to 64bit Linux for
>> desktop.  I also recall some problems with OpenOffice and Firefox
>> plugins.  How up to date is the X86_64 side and what in known not to
>> work?  In other words, can someone spell out what the advantages and
>> disadvantages of using 64bit OS vs 32bit (from Mandriva perspective!)
> 
> I would presume you get some better feeling with 64 bits (i tried it on my
> athlon 64 bits, 32 or 64 bits, and the 'feeling' was the system is more

I've this feeling too.

> responsive in 64 bits mode). For some applications (OpenOffice.org) or

I had the same feeling.

> proprietary plugins (flash, real, ...) there is no 64 bits versions atm. As far
> as i know, there are well integrated to work flawlessly in the 64 bits
> environnment (in other words, the 32 bits are installed apart on the system and
> you won't notice problems).
> 
> For the 64 bits by itself, i assume you *should* get some significant
> performance increase for some applications (for instance media
> compression/decompression, video edition), but i'm not sure all codes benefit
> from the 64 bits power.
> 

On average on AMD architecture the 64bit boost is on average from 5 to 20%.
On Intel sound there is not such significative boost over 64bit. Peaks
might be also much higher on some specific field (e.g. encryption, etc.).

You have also to consider a few aspects. 64bit model in use
is the LP64, which means that datatype "long" and "pointer"
are at 64bit. So tends also to consume more memory.
Secondly if you have 4GB or more, on 64bit takes advantage
of this, and you may have a single process using it, while on 32bit the
maximum amount of memory a single process can access
(i.e. without using strange paginating techniques)
should be around 3GB.

Regarding the performance, one aspect should be also considered.
The RPM i586 flags are choosen to work on any Pentium architecture.
The SSE set for instance (unless taken into specific account
for a specific package) is not taken in consideration for 32bit distro,
because the choosen flags are for having a compatibility for
Pentium CPU which didn't had such instruction set (it was
introduced in Pentium II if I remember right).

On 64bit instead the lowest common denominator was higher,
because when first AMD 64bits were out, the SSE (and SSE2) instruction
were already available, so the internal compiler flags take
also in account this (AFAIK -msse -msse2 -mfpmath=sse is the default
for 64bit gcc [note that I speak about compiler flags, not %optflags of
RPM]).

Bye
Giuseppe.