This "Mirrored Drive Doors"
Power Mac G4 has been my workhorse for a couple years now, and it's
currently outfitted with its original Radeon 9000 video card, dual 1
GHz PowerPC 7455 CPUs, a pair of 400 GB Deskstar hard drives (one for
work, one for backup), and 2 GB of RAM.
The Mirrored Drive Doors was the last Power Mac to boot into the
Classic Mac OS, and it requires Mac OS 9.2 or later (10.2 or later when
running OS X). Benchmarks were run from an external FireWire drive
(7200 rpm, probably a Deskstar as well), since I didn't want to mess
with my normal setup on the internal hard drive.
This is the second system we've benchmarked using Mac OS X 10.5
"Leopard". The first was a 500 MHz dual CPU Power Mac
G4.
Let 1000 Windows Bloom
The system was tested on 29 January 2008. Running Panther (OS X
10.3.9), the Power Mac displayed 1000 windows in 28.5 seconds. The same
test took only 13.4 sec. in Tiger (10.4.11). Apple definitely improved
graphics performance as Mac OS X has moved forward - until Leopard,
which takes 16.4 sec. to complete this benchmark.
These benchmarks run in half the time they do on our dual 500 MHz
Power Mac G4.
Power Fractal
This program creates a fractal pattern, which can really tax a CPU.
Under Panther, it took 2.9 seconds and scored 8,710 Mflops. Under
Tiger, it ran in 3.0 seconds at 8,565 Mflops. And under Leopard, it
score 8,470 Mflops. As OS X has developed, more processes are taking
more CPU time away from the app running in the foreground - that's less
than a 3% performance difference from 10.3 to 10.5.
Power Fractal benchmarks are 2.3x what they are on our dual 500 MHz
machine.
Xbench
Xbench 1.3 runs on both versions of OS X we're using for our
testing. A score of 100 is the performance of a 2.0 GHz Power Mac G5.
OS X 10.4 benchmarks were also run on March 6, 2009:
10.3.9 10.4.11 3/6/09 10.5.6
Overall 42.4 50.4 54.0 47.6
CPU 66.0 54.3 54.3 54.1
Threads 61.7 71.3 71.8 53.9
Memory 38.7 38.1 38.4 43.8
Quartz 58.0 55.2 52.4 61.5
OpenGL 55.6 60.7 60.9 54.5
User Int. 26.3 57.9 53.2 27.9
Drive 31.9 35.3 59.3 59.8
We can see how OS X has become more efficient over the years.
Although the CPU scores lower in Tiger, the increased performance in
the Threads, OpenGL, Drive, and especially User Interface tests more
than offsets it.
Leopard performance is far slower on User Interface, significantly
slower on Threads, a bit slower on Open GL, virtually identical on CPU
and Drive, and somewhat better on Memory and Quartz. Overall.
In comparison to our 500 MHz dual G4, the overall scores are roughly
twice as high, although most of the individual tests fall behind
doubling performance. The biggest difference is in the User Interface
test, where the Mirrored Drive Doors's Radeon 9000 Pro graphics card
really shines with nearly 3x the score under Panther, over 4x with
Tiger, and over 3x in Leopard.
Next we compare performance with the same 400 GB Deskstar drive on
the Power Mac's internal hard drive bus and in an external enclosure,
the NewerTech miniStack V2, connected with FireWire.
Internal Drive FireWire Drive
10.4.11 10.5.6 10.4.11 10.5.6
Overall 54.0 47.6 53.6 41.0
CPU 54.3 54.1 54.3 53.6
Threads 71.8 53.9 71.5 58.5
Memory 38.4 43.8 39.6 38.1
Quartz 52.4 61.5 52.9 53.8
OpenGL 60.9 54.5 61.2 50.5
User Int. 53.2 27.9 53.6 20.7
Drive 59.3 59.8 52.9 45.8
In theory, FireWire could be faster than UltraATA as it is a more
intelligent protocol and places less demands on the computer's CPU. If
this is true, scores should generally increase, but Xbench results don't
give us any evidence of that - drive performance is a bit slower and the
other results are almost identical under Tiger. And except for the Threads
test, performance with the external drive under Leopard lags well behind
the internal drive tests.
The only FireWire enclosures I have that support drives over 128 GB
are NewerTech miniStacks, so I can't say whether the slower drive results
are due to FireWire itself or the FireWire bridge used in the miniStack.
At least in this case, the internal bus is superior.
Geekbench
Geekbench 2.1.2 only runs in Tiger and Leopard and only tests the
CPU and memory systems. It pays no attention to graphics or hard drive
performance. A 1.6 GHz Power Mac G5 would have a score of 1000.
10.4.11 10.5.6
Overall 894 939
Integer 1159 1131
Float. Pt. 1015 1179
Memory 495 470
Stream 351 342
On average, Geekbench sees a 5% performance improvement with Leopard
- very impressive. The relatively low memory and stream scores are due
to the Power Mac G4 having a much slower memory bus than the baseline
1.6 GHz Power Mac G5 - 167 MHz vs. 800 MHz.
As we've often seen in our tests, benchmark scores are typically
twice that of our dual 500 MHz Power Mac G4.
Now we'll compare results when running from an external FireWire hard
drive, the same 400 GB Deskstar that was previously used inside this
Power Mac.
Internal Drive FireWire Drive
10.4.11 10.5.6 10.4.11 10.5.6
Overall 894 939 900 943
Integer 1159 1131 1169 1117
Float. Pt. 1015 1179 1026 1185
Memory 495 470 495 485
Stream 351 342 349 406
Geekbench doesn't test the hard drive, but if FireWire is less demanding
of computing resources, we would expect to see an improvement in test
scores - and we do. Granted, the difference is minimal, too small to notice
in real world use, and thus insignificant, but it does show that as far as
raw computing power goes, there's no penalty from using an external FireWire
drive.
Conclusion
Mac OS X keeps getting better, even on old hardware like this August
2002 Power Mac. It handles Leopard quite well, although it would
benefit from a better graphics card. Although we have Leopard, this
machine will continue as our primary production machine running Tiger
for the forseeable future, as we are hopelessly wed to Claris Home Page
and thus need a Mac and Mac OS that can run Classic Mac OS apps.
