[Beowulf] [craig.hunter at nasa.gov: Re: Intel?]

Brian R Smith brian at cypher.acomp.usf.edu
Wed Jun 8 11:09:24 PDT 2005


> If Apple announced that they were going to run on PPC and Intel hardware, it
> would have made a lot of sense to me -- good for competition, pick the best
> CPU for each particular application, etc..., and it wouldn't have made
> current and near-future PPC systems seem like lame ducks.  Instead, they
> announced a very dramatic shift which sent out a huge ripple.  It just
> doesn't make sense to me, so I think there must be a lot more going on
> behind the scenes that we don't know about (Apple/IBM politics).  I just
> can't think of a rational reason why they would make such a dramatic
> announcement in this fashion without any short-term PPC roadmap (at the very
> least, a detailed phase-out plan would have made sense).

Actually, Apple plans on supporting BOTH PPC and x86.  The developer
white-papers released on 06/06 provide a nice primer on the migration.
It discusses the use of "universal" binaries which support instructions
on both architectures (not unlike "Fat Binaries").

Also, there has been some discussion on what this will mean with regards
to AltiVec and the time and resources put in to optimizing these codes.
The white-papers provide a very nice table on converting existing
AltiVec calls to SSE2.  Although the SSE2 calls don't appear as elegant
as their AltiVec equivalents, Apple engineers have stated that the
conversion to SSE2 might provide additional performance as the routines
they have produced for such things as matrix multiply, vector adds and
subtracts, appear to be more efficient on SSE2 than on AltiVec (though
they provide no benchmarks to substantiate this nor can I recall the
source of this statement, so I could be wrong).



Brian R. Smith
Systems Engineer
Research Computing Core Facility, USF
Phone: 1(813)974-1467   Cell: 1(813)230-3441
Address: 4202 E Fowler Ave LIB 613
         Tampa, FL 33620
Web: http://rccf.acomp.usf.edu

More information about the Beowulf mailing list