On the last episode (Fri, 29 Jul 2005 14:04:38 +0200 (CEST)), Wojciech
Puchar <wojtek@tensor.3miasto.net> said:

> > for private reasons (nvtv support in fact ;) I'm running NetBSD/i386 on
> > my Shuttle SN85G4v3. I've just tested WinXP-64 on it and have issues
> > (which we don't care here ;) too.
> 
> tested too - actually slower than WinXP/i386 ;)
> 

oh ? how is it possible ??? M$ homepage states it's a brand new thing
enabling users to gain significant perfomance enhancement, ... :p

> >
> > I read that even Solaris runs apps in 32bit-mode (firefox) on sparc64.
> > This makes me believe, firefox is not written 64bits in mind.
> 
> no idea why, i'm running firefox on my amd64 as 64-bit app. works fine and 
> fast.
> 
> > So, even if I compile/run firefox (from pkgsrc) on NetBSD/amd64, I
> > guess it's not using all the benefits of amd64 specifics arch, true ?
> 
> actually in NetBSD/amd64 there are some problems running 32-bit 
> non-native apps, like 64-bit only linux emulation.
> 

have you tried running NetBSD/i386 apps on NetBSD/amd64 too ?
as I have play with NetBSD/i386 on this machine and build packages, I
thought I may be able to run some apps (namely nvtv ;) in
/emul/netbsd32.

> all native apps are 64-bit.

ok

> 
> >
> > My question is, when you have to run NetBSD/i386 on an AMD64 machine
> > (for driver or app support), to what extend do you get the benefits
> > from the AMD64 architecture ?
> >
> you will have exactly the same driver support on NetBSD/i386 and 
> NetBSD/amd64 because NetBSD is really designed to be cross-platform.

yep, I knew that (my Adaptec U160 sits in my Ultra5 because of this :) -
I was thinking of performance issue/benefits. The thing I'm wondering
if I am still (really) faster (in i386 mode) with my Athlon64 3200+ than
if I had an AthlonXP 3200+ (or P4)...

there :
http://www.pcstats.com/articleview.cfm?articleid=1469&page=13
didn't seem to show real advantages in having amd64 when running 32bit
apps.

eventhough it's windows benchmarks, I would have expected better amd64
marks.

> in sources - the only real difference is small part of kernel code.