AMD Athlon64 Systems

Mike A. Harris mharris at redhat.com
Sun Oct 19 19:07:06 UTC 2003


On Sat, 18 Oct 2003, Benny Amorsen wrote:

>> Your argument is entirely hypothetical however.  I've not seen 
>> any proof that the increased cache footprint makes much of a 
>> difference personally.  Speaking in terms of what exists today 
>> here and now - AMD64 64bit OS and userland apps outperform x86 
>> 32bit OS running on the same machine hands down.
>> 
>> If anyone has information to the contrary however, I'm definitely
>> open eared and eyed to have a look at their non-hypothetical test
>> results.  ;o)
>
>How do you get non-hypothetical test results, when AMD64 does /not/
>support the extra registers in 32-bit mode?

I never claimed it was possible.

>The fact is that all other CPU architecture which can run in
>both 32-bit and 64-bit mode are faster in 32-bit mode. And there
>are non-hypothetical test results showing that.

I don't discount that at all.  However, AMD64 is not one of those 
architectures.  The hardware is mass market available now, so 
feel free to do benchmarking of the CPU running both 32bit and 
64bit OS, with 32bit and 64bit applications.

I don't have numbers for you, however it seems rather clear to me 
that the registers gained in 64bit mode, more than offset any 
performance losses due to larger cache footprint.

Feel free to debate amongst the nonbelievers however if you 
wish...


-- 
Mike A. Harris     ftp://people.redhat.com/mharris
OS Systems Engineer - XFree86 maintainer - Red Hat





More information about the test mailing list