BenchmarksTo me, the iMac running Mac OS X felt very responsive; much more so than my G3 running Mac OS 9, even though OS 9 is much less sophisticated than OS X. But the feel of a system is hardly an accurate measure of its actual speed, so I decided to run a few benchmark applications. They were all Classic applications, rather than being OS X native, so I tried each one in three ways: running on the iMac in OS X via the Classic environment, running on the same machine booted into native OS 9.2.2, and on my 350MHz G3 running OS 9.1. For the native tests, the 'Mac OS 9 All' set of extensions was enabled.
The benchmark utilities I used were RAVE Bench, which tests 3D graphics performance using RAVE, ThroughPut, which tests the data rate to the graphics card that a Mac can achieve (both of which are significant for games), and TechTool Pro, which executed a number of tests to determine the performance of the overall system.
The results are shown below, along with some notes about each test.
TechTool Pro 3.0.5I used TechTool Pro to measure a couple of video performance tests and then a range of benchmarks to check the speed of the processor. The video tests just generate colour and monochrome noise very quickly with different pixel sizes. The Whetstone and Dhrystone tests check maths and string processing speeds, whilst the Cycle test shows the number of event loops processed per second. Sieve calculates prime numbers and Sort sorts a complex array. Larger numbers are better for all tests except for Sort, where the time taken to complete the test is shown.
The Cycle test is particularly revealing, as it shows that the Classic environment is receiving fewer processor cycles per second than on even the G3. This is undoubtedly to do with the multitasking nature of Mac OS X and, given the reduction involved, it's impressive that the Classic environment works as well as it does.
ThroughPut 1.5The ThroughPut utility tests the amount of data, in megabytes per second, the Mac can send to its graphics card via the PCI or AGP slots. These figures are of course significant in terms of games performance.
The iMac easily outpaces the G3 system even in Classic mode. Note that there's no AltiVec result for the G3 because this was one of the enhancements introduced with the G4 processor.
RAVE Bench 1.0Finally, RAVE Bench also tests graphics performance by timing a number of graphical operations such as the speed of plotting a number of texture-mapped and interacting objects.
I performed two separate RAVE Bench tests: one with software-only rendering, to show the relative data-processing performance of the two machines, and the other using the appropriate RAVE engines associated with each OS version and graphics card to show the actual level of graphical performance that can be expected from each machine. Both tests used just the basic 640x480 setting.
In the first software-rendering test, both G3 and G4 machines are being compared with a 195MHz PPC 604e machine (a set of test figures supplied with the RAVE Bench utility).
The second test shows the performance of the iMac relative to my own G3 (so the G3 itself is obviously not present on the graph: it equates to 100% for each test).
As can be seen in the second test, running through Classic can have a significant impact on performance, and in the Intersections routine, the iMac was actually fractionally slower than my G3. This was a one-off, though: in all other cases it was faster, and certainly the machine feels faster even in Classic mode. However, it's clear that running natively under Mac OS 9 can give a massive performance boost, with most tests running two-and-a-half to four times as fast as my G3. In other words, the graphical improvement is significantly greater than the improvement in base processor speed may suggest.
Of course, real-world performance will depend on individual applications, and, as noted below, Mac OS X-native games seemed to give the same performance under both OS X and OS 9; even games running in the Classic environment seemed noticeably smoother than on my G3. The tests above should not, of course, be used to judge OS X in its own right, as all the OS X tests were performed in Classic mode. Native OS X games and other applications will not suffer the overheads imposed by the Classic environment.
Going purely on clock speed alone, I would have expected the 800 MHz G4 iMac to be rather more than twice as fast as my 350MHz G3 Power Mac, and indeed this is the case. The 800 MHz G4 is perhaps not quite as fast for pure number-crunching as I would have anticipated in comparison with my G3, given that the G4 has an improved architecture, but the results are reasonably in line with my expectations for native OS 9 performance. A performance penalty for running in Classic is unavoidable and to be expected, but in reality the iMac feels like a very fast computer regardless of which of the operating systems I run on it!