Re: [MLUG - DISCUSSION] Ouch: Apple spanked by benchmarks.

[Jonathan King <EMAIL:PROTECTED>]

On 6/3/05, Michael <EMAIL:PROTECTED> wrote:
> http://www.anandtech.com/mac/showdoc.aspx?i=2436
> 
> Looks like Apple's benchmarks really suck especially considering that
> the PC hardware it was compared with is probably cheaper.

This was a pretty weird benchmark, though.  At the outset, you've got
x86 + Linux + GCC 3.3 vs. PowerPC + Darwin + GCC 3.3.

That should raise eyebrows for several reasons.  For starters, they
could also have tested LInuxPPC on the Mac (what part of the
performance is due to Mac OS X)?  I read a suggestion at slashdot that
part of their results depend crucially on the fact that Mac OS X (and
BSDs in general, I hear) do suffer from slower thread creation than
Linux or Windows.  For that matter, the shipping compiler with Mac OS
X 10.4.1 is GCC 4.0, which is the compiler that Apple spent the last
year tweaking.  I can understand (sort of) an argument that you want
to keep as many things controlled as possible, except for the fact
that a cross-platform compiler is heavily dependent on how much effort
has been made to optimize a given target.  A lot more effort goes into
x86 for obviuos reasons.

Anyway:

> And that you
> can stack PC hardware a lot more powerful than a dual-G5. Quad-configs
> aren't unusual in servers and I've seen PC's with up to 32 CPU's in a
> single box. I imagine denser configs exist.

XServes are 2 CPUs in a 1u rack; density is not a big factor.
 
> The G5 had mixed scores. A good compiler could do a lot with it but it's
> integer crunching abilities and general memory latency really suck.

That wasn't necessarily the G5; it could have been Apple's design (for
the latency issues) and/or the compiler.

> I wonder if the new CPUs in the XBox 360 and PS3 (both PowerPC based) will
> suffer these same issues or if they fixed these issues?

See above.  Also note that the game machines will be made to live
their entire lives executing AltiVec (or equivalent) instructions and
talking to the video system.

> the PS3's cell processor at least takes care of most of these issues as
> that is part of the reasoning behind all those SPE's. If so Apple may
> have a good future using those in it's hardware.. if not then they need
> some serious help.

Interesting, Slashdot is now echoing a C|Net report that Apple will
officially announce a switch to Intel at their developer's conference
next week.  I still think that is pretty unlikely for a variety of
reasons, but who knows?
 
> OS X (on G5) totally sucks in performance when compared to Linux (on
> x86)

...if you run Apache 1.3 and MySQL and don't care about anything else.
 Performance is now quite good on the GUI side, good in the kinds of
FP apps they didn't test (I'm thinking R and Octave) and scales well
in supercomputing set-ups.  Anyway, just one more comment:

> I'd like to see Apple respond to this with any tips they have. Obviously
> out-of-the-box they get spanked but maybe there is a way to tune the
> system, or better compile the software, to run faster?

Uh...use the right compiler?  Seriously, if you have a better
optimized gcc 4.0 and gcc 3.3 to choose from, and choose the latter,
hey, whatever floats your boat.  But don't tell me that you made an
effort to make things fast that way.
 
> Speed is definately not the only measure of an OS.. but since I already
> hate OS X's interface  and apps on OS X crash frequently 

Uh...that's not what I see.  Seriously, Michael; what OS  X apps crash
frequently for *you*?  I will note that I'm not (yet) happy with the
stability of Safari on Tiger; that did seem to go backwards. 
Similarly, Preview crashed twice on me with 10.4.0 (up from never on
10.3) but that's gone away now with 10.4.1.  The only other apps I've
crashed since 10.4 hit were mplayer (on a malformed avi file mentioned
on a previous thread), and osiriX 1.7.0 (fixed in 1.7.1).

I also never did figure out what you hate about OS X's interface that
you couldn't change easily if you wanted to.

To sum up, I wouldn't doubt at all that there are things that Linux on
Intel isn't better at, but this article seemed pretty surreal to me.

jking

_______________________________________________
discussion mailing list
EMAIL:PROTECTED
http://mlug.missouri.edu/mailman/listinfo/discussion