'Mystic' Power Mac G4/500 Dual

These are the first benchmark results we've done using Mac OS X 10.5 "Leopard" - but they won't be the last. I've had Leopard on my production machine, a "Mirror Drive Door Power Mac" G4/1 GHz dual, for some time, but I still use Mac OS X 10.4 "Tiger" almost exclusively, as I use Classic Mode daily. I'll make that our next benchmark project.

This computer is a "Mystic" Power Mac G4/450 dual that's had its CPUs upclocked to 500 MHz by Operator Headgap, which usually has upclocked 400 MHz to 550 MHz single G4 modules available at very reasonable prices.

The Mystic was Apple's second generation of AGP Power Mac, and its claims to fame include gigabit ethernet and dual processors. It's a lot like the "Sawtooth" G4 - same Rage 128 Pro AGP 2x video card, same four memory sockets, same drive support, same case. With Motorola unable to provide CPUs beyond 500 MHz, adding a second CPU was the only way Apple could offer more power.

This computer has 768 MB of RAM (three 256 MB PC133-333 modules) and an 80 GB 7200 rpm Deskstar, which is one of my favorite hard drives. The drive has been repartitioned since I tested at 450 MHz - it now has Mac OS X 10.2.8, 10.3.9, 10.4.11, and 10.5.6 installed. The system was configured with a Kensington mouse, Acer keyboard, and 17" Samsung 700DF monitor at 1024 x 768 resolution and millions of colors for the 500 MHz tests. The Leopard installation was freshly clones to the internal drive this morning, and I remembered to turn Spotlight off after running most of the benchmarks - it made a big difference!

Test results compare the overclocked 500 MHz machine against results at 450 MHz. Based on the difference in CPU speed, we would expect some results to be 10-11% better.

Let 1000 Windows Bloom

The system was tested at 500 MHz on 19 February 2009.

                   450 MHz    500 MHz
Mac OS X 10.2.8   56.4 sec.  53.6 sec.   5.2% faster
Mac OS X 10.3.9   41.1 sec.  41.3 sec.   0.5% slower
Mac OS X 10.4.11  32.2 sec.  26.4 sec.  22.0% faster
Mac OS X 10.5.6              35.4 sec.

Apple had definitely been improving graphics performance on older hardware as Mac OS X has moved forward - until Leopard. This computer's antiquated Rage 128 video card just can't keep up with Leopard.

This Power Mac was tested on 6 July 2009 with 1.25 GB of RAM installed, a Radeon 9000 graphics card, and booted from an external FireWire hard drive. With OS X 10.4.11 it came in at 27.5 sec., and with 10.5.7 it finished in 36.9 sec. - both surprising slower than with the original Rage128 Pro video card.

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 7.0 seconds and scored 3,662 Mflops. Under Tiger, that dropped to 7.9 seconds and 3,232 Mflops.

                    450 MHz       500 MHz
Mac OS X 10.3.9   3,662 Mflops  3,788 Mflops   3.4% better
Mac OS X 10.4.11  3,232 Mflops  3,769 Mflops  16.6% better
Mac OS X 10.5.6                 3,742 Mflops

This is one benchmark where the results are pretty consistent today - the difference between Panther and Leopard is just 1.2%.

The system was retested on 6 July 2009 with 1.25 GB of RAM, a Radeon 9000 graphics card, and running from an external FireWire hard drive. Under OS X 10.4.11 it scored 3,411 Mflops, and under 10.5.7 it achieved 3,322 Mflops. Again results are lower than before.

Xbench

Xbench has been out for some time, and version 1.1.3 runs on all three versions of OS X installed on the Mystic, and tests a lot of things. Here are the test results using Xbench 1.3 for 10.3 and 10.4 (100 = 2.0 GHz G5):

         --- 450 MHz --  --- 500 MHz --
         10.3.9 10.4.11  10.3.9 10.4.11    
Overall    22.8   25.7     23.6   26.3     3.5/2.3% better
CPU        29.5   28.9     32.7   29.7    10.8/2.8% better
Threads    34.9   39.4     36.7   41.7     5.2/5.8% better
Memory     20.7   23.2     21.4   23.0    3.4/-0.9% better
Quartz     27.2   25.1     28.0   25.9     2.9/3.2% better
OpenGL     34.0   38.6     34.9   40.8     2.6/5.7% better
User Int.   9.1   12.0      9.3   12.7     2.2/5.8% better
Drive      51.2   49.4     50.8   43.0    0.8/13.0% worse

We can see how OS X has become more efficient over the years. Running at 500 MHz, the overall score is about 2.3% higher under Tiger than Panther. The results are anything but predictable. We only see the expected 10-11% improvement in the CPU score, and then only under 10.3. It is nice to see that a faster CPU also gives us faster graphics even though there has been no change to the video card.

Next we test using Leopard at 500 MHz:

Overall    22.6  14.1% slower
CPU        30.4   2.3% faster
Threads    36.7  12.0% slower
Memory     22.7   1.3% slower
Quartz     26.4   1.9% faster
OpenGL     36.8   9.8% slower
User Int.   8.5  33.1% slower
Drive      43.2   0.5% faster

As always, Apple is tweaking different parts of the Mac OS, and Leopard is a bit faster on the CPU and Quartz tests, almost unchanged on Memory and Drive tests, but a fair bit slower on the Threads test and a lot slower - one-third slower - on the User Interface test. Overall, Xbench says Leopard is 14% slower than Tiger.

We ran fresh benchmarks on 6 July 2009 with 1.25 GB of RAM, Radeon 9000 graphics, and an external FireWire drive. Here are the results under Tiger and Leopard:

         10.4.11 10.5.7    
Overall    24.0   20.0
CPU        30.2   29.9
Threads    39.1   35.1
Memory     21.6   21.5
Quartz     25.3   24.6
OpenGL     34.7   31.4
User Int.  11.0    7.1
Drive      36.8   36.5

Here's solid evidence that running from a FireWire 400 hard drive is not faster than using an internal hard drive than one on the internal Ultra ATA66 bus (rated bandwidth for FW 400 is 50 MB/sec., while ATA66 is 66 MB/sec.). This will definitely change my working strategy - I have been using a FireWire drive because it's so easy to move to another Power Mac should my main machine have problems.

We're seeing lower scores in most tests, with the CPU score under 10.4.11 being the only exception - and that's just a 1.7% difference. Once again, we get better graphics scores with the stock Rage 128 Pro than with the far more modern Radeon 9000.

Geekbench

Geekbench only runs in Tiger and Leopard (which I ran from an external 2.5" hard drive, the 40 GB 5400 rpm drive that used to reside in my late TiBook).

   TIGER       450 MHz  500 MHz
Overall          439      459    4.6% faster
Integer          529      560    5.9% faster
Floating Point   529      547    3.4% faster
Memory           266      278    4.5% faster
Streams          165      162    1.8% slower

After the 500 MHz upgrade, three of four tests benchmark faster, giving a 4.6% performance boost overall - but bear in mind that Geekbench doesn't measure hard drive or video performance.

Here are the Leopard results at 500 MHz:

Overall          450    2.0% slower
Integer          551    1.6% slower
Floating Point   548    insignificant
Memory           244   12.2% slower
Streams          169    4.3% faster

A 2% overall difference in performance is virtually imperceptible.

The computer was benchmarked on 6 July 2009 with 1.25 GB of RAM, Radeon 9000 graphics, and an external FireWire hard drive.

               10.4.11  10.5.7
Overall          454      460
Integer          524      506
Floating Point   587      632
Memory           250      226
Streams          161      170

Geekbench only tests the processor and memory systems, so we'd expect the 10.4.11 results to be almost identical, yet the Integer and Memory scores are significantly lower in the latest test. Perhaps having more RAM to deal with is a factor.

The overall score for Mac OS X 10.5.7 is a bit higher than under 10.5.6, and that seems to be primarily due to improvements in the Floating Point test, which is 15% higher. Memory is again a bit slower, possibly due to more RAM for the computer to manage.

Conclusion

The best news is that OS X keeps getting better, even on old hardware like the January 2001 Mystic. This machine has been our testbed for hard drives, memory configurations, CPU upgrades, and AGP video cards, but now we're turning it into a low-end website server (not for LEM).