Dennis Clarke wrote:

>  You would be surprised how often an open source application gains little
> advantage by being a 64-bit build.  The same binary on sun4m and sun4u will
> operate within 3% of the clock speed limited rate.  I am sure you have seen
> this too.  The creation of and inclusion of chip specific libraries in
> dependencies and in ISAEXEC/ISALIST type of directory strucs (
> v8,v8plusa,v9,v9a etc etc ) often duplicates and triplicates the efforts
> needed to create a single package.  For what gain?
> 
>  I am not being argumentative here.  I am often wont to do I know.
> 
>  I am simply speaking from measured experience.
> 
>  The other side of the coin exists also.  There are applications in which
> the virtual memory space and other goodies ( VIS !!! ) can make all the
> difference in the world.  I still have a copy of Ultra Computing Volume 2
> here that shipped with the UltraSparc units that had Creator 3D
> framebuffers.  The performance of certain operations in UltraSparc with VIS
> is just blistering fast.
> 
> I am preaching to the choir.  Sorry.

While most 64 bit sparc apps are actually slower than the 32 bit
versions, I can assure you that is not the case for 64 bit amd
binaries... in general, we see a 15% win; sometimes much more.

I'm not arguing that we need 64bit versions of all apps - but
libraries is probably a good idea.

- Bart

-- 
Bart Smaalders                  Solaris Kernel Performance
barts at cyber.eng.sun.com              http://blogs.sun.com/barts

Reply via email to