Thanks to a donation from a reader, we have our first 'Digital Audio' (DA) Power Mac
G4, which we received bare bones - no RAM, no hard drive. It's the
533 MHz model with an ATI Rage 128 Pro AGP video card, the same card
found in earlier G4 Power Macs. I transplanted the PC133 RAM and 80 GB
7200 rpm Deskstar hard drive from our 'Mystic' Power Mac, which
has separate partitions for Mac OS X 10.2.8, 10.3.9, 10.4.11, and
10.5.6. Everything just worked.
The DA was Apple's third generation of AGP Power Mac, and its big
improvement over earlier generations was moving from a 100 MHz system
bus to a 133 MHz bus. On the down side, there's no microphone
input and there are only 3 RAM sockets, giving it a maximum of 1.5
GB.
I have a Kensington mouse and Acer keyboard attached and used a 17"
Samsung 700DF monitor at 1024 x 768 resolution and millions of colors
for all tests.
Speedometer 4.02
The system was tested on 10 March 2009 under Mac OS 9.2.2 in my
standard configuration. Display run at 1024 x 768 at millions of
colors. Results are relative to a Quadra 605, which rates 1.0. All
tests were run with the cache at the 8 MB default setting. Numbers
rounded off to one or two decimal places.
CPU graphics disk math
30.61 n/a 3.35 1124.6
The CPU and math settings are virtually identical. Both the external
Western Digital drive/FireWire combo and IBM Travelstar drive handily
outperformed the stock Toshiba drive. In the case of the IBM drive, the
5400 rpm speed and larger cache managed to boost overall drive
performance by 35%.
MacBench 5
The system was tested on 10 March 2009 under Mac OS 9.2.2 in my
standard configuration. Display run at default 1024 x 768 at millions
of colors. The disk cache was set to 8 MB. Results are relative to
a Power Mac G3/300, which rates 1000.
CPU math disk graphics
1692 2008 2985 n/a
Let 1000 Windows Bloom
The system was tested on 10 March 2009. Running Jaguar (OS X
10.2.8), the Digital Audio displayed 1000 windows in 49.7 seconds. The same
test took only 40.1 seconds under Panther (10.3.9), and just 22.44 sec.
in Tiger (10.4.11). Apple definitely improved graphics performance
through Tiger, but it took a real hit with Leopard (10.5.6) because
this old video card doesn't support Core Image, which Leopard uses
extensively. Running 10.5.6, this benchmark took 44.4 seconds.
Power Fractal
This program creates a fractal pattern, which can really bog down a
CPU. It does not run under Jaguar. Under Panther, it took 6.4 seconds
and scored 2,096 Mflops. Under Tiger, performance decreased slightly to
6.5 seconds and 2,079 Mflops. There was another small decrease when
running Leopard, which completed the test in 6.6 seconds and scored
2,043 Mflops. The difference in scores between these three versions of
OS X is insignificant.
Xbench
Xbench has been out for some time. Version 1.3 runs on all versions
of OS X since 10.3 and tests a lot of things. Here are the test results
for 10.3, 10.4, and 10.5 (100 = 2.0 GHz G5):
10.3.9 10.4.11 10.5.6
Overall 24.7 29.8 16.1
CPU 34.2 33.4 30.3
Threads 25.3 25.2 22.5
Memory 25.4 27.1 26.6
Quartz 29.7 29.9 20.0
OpenGL 53.0 58.4 13.7
User Int. 9.7 17.3 5.8
Drive 52.6 49.7 44.3
Again we can see how OS X became more efficient through Tiger. The
overall score is 2.6% higher under Tiger than Panther, and improvements
in 10.4 are especially evident in the User Interface test. While CPU,
threads, memory, and drive performance was down a bit in Leopard, it
really took a hit on graphics tasks - Quartz was down by 33%, OpenGL by
76%, and User Interface by 67%.
Geekbench
Geekbench only runs in Tiger and Leopard. A 1.6 GHz Power Mac
G5 would have a score of 1000.
10.4.11 10.5.6
Overall 362 321 11.3% slower
Integer 406 354 12.8% slower
Floating Point 400 361 9.8% slower
Memory 314 265 15.6% slower
Streams 177 187 5.6% faster
As we've seen with other benchmarks, Leopard is generally slower
than Tiger, although the Streams test does run faster under 10.5.6.
Conclusion
For older G4 Macs, Mac OS X 10.4 'Tiger' seems to be the best choice
overall. Leopard would have done better with a more modern graphics card
and might have done better with more RAM, but we can only test with what
we have at present.
