Let’s just get it out of the way: The
Macintosh G5 is most certainly not the
“world’s fastest PC,” as Apple so boldly
asserts. Its architecture looks hot on
paper, no doubt, but in many of our
tests, the G5 platform was in fact far, far,
slower than the fastest personal desktop
platforms based on AMD and Intel CPUs.
This, however, doesn’t mean the G5 is
a complete loser. Quite the contrary. In
fact, we’re happy to say—for the very
first time—that the new Apple Dual G5 is
a legitimate PC that must be taken seri-
ously. In the old days, we could count on
Apple’s benchmark results to be as over-
cooked as crazy Aunt Patty’s pot roast.
But today, while we certainly found
pockets of shameful performance, we
also see a silver lining to Apple’s cloud.
More about this later; for now let’s look
at the test results.
The Dual G5 fell down the hardest in
a couple tests we thought it would own.
In Mathematica , for example, the Mac
was 35 percent slower than the Pentium
4 Extreme Edition and 42 percent slower
than the Athlon 64 FX. The G5 dualie was
also 39 percent slower than the Athlon
64 FX and 36 percent slower then the
Pentium 4 Extreme Edition in the float-
ing-point intensive SETI@Home bench-
mark. We were pretty stunned by these
performance differences, as we expected
the G5 to equal or exceed PC perfor-
mance, what with its 1GHz system bus
and considerable floating-point features,
which look so good on paper.
So why such poor floating-point per-
formance? After all, the G5’s PowerPC
970 processor is a derivative of chips
used in supercomputing clusters made
by IBM.
We think it comes down to the soft-
ware code. The most logical explanation
is that the code in Mathematica and
SETI@Home just can’t take advantage
of all that the G5 has to offer. Apple
and the Mac faithful may shrug this off
as a minor, temporary problem, but if
history has shown us anything, it’s that
software optimizations can take longerthan expected. For example,
it’s taken almost three years to
get the code-base up to speed
for the Pentium 4’s uniquely
long processing pipeline and
special instructions.
Regardless, the new Mac
was a massive dog in these two apps,
and would likely be so in other floating-
point apps that haven’t been expressly
recoded for the G5. On the PC side of
things, the Athlon FX-51 kicked major
P4EE ass in these two math-intensive
tests.
Gaming, you ask? The G5 blew embar-
rassing chunks in the very popular
Unreal Tournament 2003 and Jedi Knight
II. Since we ran both games at low reso-
lutions and with audio disabled, we can’t
blame the abysmal numbers on the G5’s
video or sound subsystems. And we do
mean abysmal. Horrible. Ugly. Perhaps
even criminal. Yes, we’re talking just that
slow. In UT2003 , the P4EE and Athlon FX
were near identical in performance, but
some 317 percent faster then the new
G5. No, not 17 percent— three hundred
and seventeen percent. In Jedi Outcast,
the PC duo turned in scores that were
about 80 percent faster than the G5.
That’s a less egregious differential, but
it’s still a felony, not a misdemeanor.
Because the G5 scores were so low,
we checked inside the box to make sure
someone hadn’t swapped an ATI Rage
Pro for the ATI Radeon 9800 Pro. Nope,
the latter was present and accounted for.AND THEN CAME QUAKE III
Before we began testing, we weren’t so
sure the PC would retain its dominance
over the Mac, what with the impressive
system specs of the G5. But after a few
test runs, we weren’t sure if we should
even continue testing. After four tests,
the affair was shaping up to be a com-
plete rout—Apple’s version of Dunkirk,
with troops fleeing for their lives across
the English Channel. Could the PowerPC
970 chip possibly be this slow? Wecouldn’t believe it. Maybe something
was wrong with Apple’s drivers or
the OS itself. We continued our tests,
and actually hoped that the G5 would
bounce back.
Well, bounce back it did. In Quake III
Arena, we finally saw respectable perfor-
mance from Apple. Although the G5 was
still the slowest of the three machines,
its score was nothing to be ashamed of.
The Athlon 64 FX was about 10 percent
faster than the G5, and the P4EE was
about 24 percent faster. Considering that
the Athlon 64 FX roundly trounces non-
Extreme P4 and Athlon XP machines in
this test, that’s a respectable showing
for the G5. In fact, in this benchmark, the
G5 is about on par with a 3.2GHz P4 and
faster than an Athlon XP 3200+.
The QuickTime test found the tables
completely turned for the G5. Encoding
a 2GB DV file to Apple’s QuickTime MOV
format gave the G5 its first win—it was
about 5 percent faster than the Intel box,
and about 14 percent faster than the
AMD box. Sure, PC users may croak that
using an app written by Apple skews the
test, but the truth is that QuickTime does
spit out perfectly fine DV conversion, the
G5 works with it splendidly. QuickTime
isn’t the world’s most popular format
for sharing video, but, hey, that’s Apple’s
problem, not ours.
The G5 also did well in the
Photoshop 7 action script crafted by
MacAddict —it was one second faster
than the Athlon 64 FX, and four seconds
faster than the P4EE. As for the
Maximum PC script that uses all filters,
fortunes reversed and the FX-51 and
P4EE were 18 and 19 percent faster than
the G5, respectively. It’s difficult to judge
exactly which platform is the Photoshop
king, because it’s so incredibly easy to
model a benchmark script that will elicitMAXIMUMPC DECEMBER 2003
What It All Means—To
You, Your Brethren, and Your
Apple-Loving Friends
The G5 separates airflow for the processors, RAM,
power supply, videocard, and optical drives.