Re: Upcoming Debian multiarch support (amd64, sparc64, s390x, mips64) [affects sarge slightly]

[Goswin von Brederlow <brederlo@informatik.uni-tuebingen.de>]

Colin Watson <cjwatson@debian.org> writes:

> On Mon, Jan 12, 2004 at 10:47:22AM -0500, Anthony DeRobertis wrote:
> > On Jan 11, 2004, at 15:16, Scott James Remnant wrote:
> > >Or just have separately named packages?  With the "lesser ABI" package
> > >depending on the "major ABI" one as I outlined in my other e-mail.
> > 
> > AMD Opteron processors apparently run faster in 64-bit mode, due to 
> > having more registers available. Having one of those systems, I'm sure 
> > that 64-bit is the "major ABI" --- I don't really see much reason to 
> > keep 32-bit around if I don't have to.
> 
> It's not necessarily that simple; many applications are disk-bound
> anyway, and furthermore in 64-bit mode you spend more time shuffling
> pointers around. When I was doing benchmarking work on the Opteron at
> AMD early last year (albeit on web servers, so perhaps not a good
> analogy for desktops), it was becoming clear to me that having the
> flexibility of 32-bit mode for some applications and 64-bit for others
> was useful, and the results of some of those benchmarks were slightly in
> favour of 32-bit mode.
> 
> I haven't touched an Opteron for nearly a year now, though ...

I think in the end a pure 64 bit system would win that last bit of
speed because less code needs to be cached. Running a 32bit biary
might be faster than the 64bit flavour but it might slow down all the
other 64bit apps.

As with all benchmarks the truth lies in the eye of the testcase
creator. I'm sure you can make a perfectly reproducible and meaning
full test either way.


But thats just amd64. All the others are realy slower as 64 bit and
only the need to >4GB ram drives one to it.

MfG
        Goswin