[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Dupal Opteron on Sarge

On Sat, Sep 17, 2005 at 06:49:55PM -0700, lordSauron wrote:
> oh... as for performance increases...
> pentium 4s use a 21 stage pipeline or something like that... so they
> take approximately 21 clock cycles to get anything done.  AMD uses
> about 7 stages (or something in that neighbourhood) so if you divide
> 2.8 by 21 and 2.0 (my Athlon64) by 7, you get a really interesting
> breakdown.  You'll certainly find a HUGE increase in performance,
> ESPECIALLY if you go dual-core.  As for dual-core and 64bit kernels
> and the like... there was a few.  you'd be looking at the
> amd64-generic-SMP kernels (I think)  the amd64 is for 64bit (x86_64)
> and the generic is b/c I don't know if opterons are considered a k8
> thing (I'd have to go check) and the SMP is to give you dual-core
> action, so you can use the withering power now avaliable to you.  It's
> soooooooooooooooooo cool... someday I'll get one... until then I just
> annoy everyone with wordy descriptions of 'em.
> and for those of you who're mentally screaming at my crude way of
> benchmarking the two cpus, it's only for a ballpark comparison.  by no
> means is that ENTIRELY accurate, and should NOT be treated as such.

It's more like 31 on the current P4s and about 13 or 14 on the opteron.
Most sane designs for x86 seem to use about 12 to 14 stage pipeline
(athlon, opteron, pentium-m, etc).  Good balance for clock speed vs cost
of a branch miss.  I believe the current high end powerpcs also run
about that length of pipeline.

The real problem with too long a pipeline is the cost in branch
prediction misses and if there isn't enough independant instructions to
keep the pipeline busy working.

Len Sorensen

Reply to: