As regular readers of this blog will know, I have recently acquired a 1.67 GHz PowerBook G4, model A1138. Since I am already in possession of a G4 Cube that had been upgraded to a 1.2 GHz G4 by the seller, I thought it might be interesting to benchmark the performance of these two G4s using real world applications, vs. using a more formal but more artificial benchmarking suite like GeekBench. Both were loaded with Mac OS X 10.5.8 Leopard and so a fair fight seemed in the offing.
At first glance, this should be a case of “no contest”. The PowerBook should be the clear winner. The CPU is nearly 40% faster, the RAM is more than five times as fast (533 MHz vs. 100 MHz), AND there is more of it, and finally the disk technology is 5 years younger than its peer on the G4 Cube. As the icing on this particular cake, the graphics cards are grossly mismatched as well: a 128 MB ATI 9700 in the PowerBook, but only a 32 MB nVidia GeForce 2 in the Cube.
Here are the relevant configuration screen shots for the G4 Cube:
… and here are the equivalent screen shots for the PowerBook G4:
So how did the benchmarking work out with real life tasks? Let’s take a look:
Task PowerBook G4 G4 Cube
Cold Boot (Leopard 10.5.8) 53s 43s
iTunes 10.6 Launch 10s 8s
TenFourFox 31.5 Launch 19s 13s
Photoshop CS4 Launch 18s 16s
As you can see from the numbers above, a curious thing happened on the way to “no contest” – the results got inverted! Inexplicably, the older, slower and less powerful G4 Cube trumped the newer, faster and (at least on paper) more powerful PowerBook G4! How could this be?
I quickly realized that two salient differences had not been accounted for in the lineup of specs I presented above. Firstly, the G4 Cube’s CPU is equipped with 2 MB of L3 cache, while the PowerBook’s G4 only has 512 KB. That had the potential to make a big difference, perhaps even neutralizing the putative 40% speed gain of the PowerBook over the Cube. Further to that, I failed to account for the fact that I had upgraded the G4 Cube’s hard drive with a new and fast 320 GB ATA-100 drive. This drive was not just fast, but also had 16 MB of onboard cache vs. the PowerBook disk’s 8 MB of onboard cache. Finally of course, we must not forget that as a notebook, the PowerBook uses a slower 5400 RPM hard drive, vs. the Cube’s desktop class 7200 RPM hard drive.
Since all of the metrics above relate to loading large amounts of data (programs and OS) from the hard drive, perhaps the skewing of results in favor of the G4 Cube was strictly a function of its faster hard drive and more disk cache? What about the CPUs? Did the Cube’s more generous CPU cache also contribute to the unexpected results above? How would the two machines fare when the task was CPU bound, not disk bound?
To answer these questions, I loaded the same very large photo (4128×3126) into Photoshop CS4 on each of the machines, and applied several CPU intensive filters to them. Let’s see how each machine fared under this testing:
Filter PowerBook G4 G4 Cube
Watercolor 21s 27s
Chalk and Charcoal 12s 16s
Mosaic Tiles 21s 28s
Here we get the “expected” result. In CPU bound tests, the Cube’s slower G4 was significantly outperformed by the PowerBook’s higher spec’d counterpart. Unexpectedly however, the onboard cache difference did not seem to provide the Cube with as much of a performance boost as I had anticipated… or had it?
In comparing the raw clock speeds of the two CPUs, we see a nearly 40% faster CPU on the PowerBook G4, but in comparing the two sets of test results above, we see only about a 25% faster performance by the PowerBook. Why didn’t the 40% CPU speed difference translate directly into 40% result differences? My guess is that the missing 15% is the visible result of the extra L3 cache on the upgraded G4 Cube’s processor. Quite a nice boost for a small amount of extra cache! As I said in one of the early posts of this blog, cache is king! 🙂
Clearly the extra L3 cache improved the Cube’s performance in the above tests, but not enough to neutralize the significant clock speed advantage that the PowerBook had. Returning to the question above (“did the Cube’s extra L3 cache contribute to its superior performance in the OS and application loading tests?”) the answer is … probably. Just how much is hard to say, but loading either OS code or program code requires a fair amount of CPU, and so it is reasonable to assume that some part of the Cube’s OS/program loading performance results can be attributed to the extra L3 cache.
So, what do we conclude from all of the above? Well, the fundamental takeaway is that in real world testing, G4 clock speeds are not the best direct measure of how well a machine will perform for you. Disk speed, disk cache and CPU cache are all factors that must be taken into account.
Secondly, and particularly in subjective evaluation of boot times and program load times, disk rotational speed and disk cache depth are significantly more important than CPU speed.
There you have it then, a tale of two G4s. Something of a tie between the two, but one that delivers some important lessons on what really influences real world performance.