Macs OS X 10.3.7--How Fast Is It?
Along the way, Apple made some speed progress with a couple of 10.0.x updates and with 10.1 and 10.2. But, as we noted in our last Jaguar benchmark, all of the Jaguar versions through 10.2.8 were basically the same speedwise. Our speed test of Panther (10.3) in October of 2003 concluded that it was much faster in some very important aspects than the last version of Jaguar, 10.2.8. But through version 10.3.4, Panther has been basically the same speedwise, at least using our test computer.
Panther's latest update is version 10.3.7. It came out last Wednesday, so just how fast is it?
Test Computers. Our longterm test computer has been a completely stock 450-MHz G4 Cube, i.e. stock except for its 704 MB of RAM. However, Apple moved to another performance level with last year's release of the Power Mac G5. Our stock G4 Cube is not only getting "a little long in the tooth" but many of the improvements in Mac OS X 10.3.6 and 10.3.7 are aimed at the much faster G5s with much faster graphics cards. So we also tested Mac OS X 10.3.7 versus 10.3.6 and 10.3.5 with our dual 2.5-GHz Power Mac G5. It has been upgraded with 1.5-GB RAM and an Apple OEM Nvidia 6800 Ultra DDL graphics card that drives Apple's new 30" Cinema Display.
Speed Tests. We ran our complete suite of speed tests on Mac OS X 10.3.7 and 10.3.6 for updating our speed tables. Our speed tests consist of a number of "canned" benchmarks and several so-called "real world" tests. They are described in detail on our "How We Conduct Speed Tests" page and most are simple enough for anyone to repeat on their own Mac for comparison. We added two recent and popular 3D games, Halo and Call of Duty, to the G5 tests--they will not run on the G4 Cube's venerable ATI Rage 128 Pro graphics card. Each test was run three times with a cold boot in between and the average score or number of seconds computed.
Here are the results for the G4 Cube comparing 10.3.7 with previous versions of Panther and the last version of Jaguar, 10.2.8:
| Test | Cube 450 MHz | ||||||
|---|---|---|---|---|---|---|---|
| Mac OS X Version: |
10.2.8
|
10.3
|
10.3.2
|
10.3.3
|
10.3.4
|
10.3.6
|
10.3.7
|
| Benchmarks: | |||||||
| Altivec Fractal 1.1.3 (GFLOPS) |
1.5
|
1.5
|
1.5
|
1.5
|
1.5
|
1.5
|
1.5
|
|
Cinebench 2003 - Rendering
|
49
|
48
|
48
|
47
|
48
|
48
|
49
|
|
- OpenGL Hardware Lighting
|
92
|
90
|
93
|
93
|
93
|
91
|
91
|
| Let1kWindowsBloom 1.0 (sec) |
61
|
42
|
55
|
42
|
51
|
52
|
53
|
| Xbench 1.1.3 - CPU |
52
|
52
|
53
|
53
|
52
|
52
|
54
|
| - Threading |
34
|
42
|
42
|
42
|
42
|
42
|
42
|
| - Memory |
64
|
76
|
76
|
76
|
74
|
75
|
76
|
| - Quartz Graphics |
63
|
72
|
72
|
72
|
72
|
72
|
73
|
| - OpenGL Graphics |
84
|
81
|
81
|
81
|
81
|
81
|
81
|
| - User Interface Graphics |
60
|
104
|
103
|
105
|
106
|
105
|
104
|
| - Hard Disk |
66
|
64
|
56
|
51
|
54
|
55
|
52
|
| "Real World" Tests: | |||||||
| Startup (sec)* |
54
|
44
|
75/52
|
51
|
52
|
55
|
54
|
| Shutdown (sec) |
7
|
13
|
13
|
14
|
14
|
15
|
15
|
| Launch Classic Mode (sec) |
-
|
-
|
25
|
28
|
29
|
26
|
26
|
| Dup. 700 MB Folder 3600 Files (sec) |
131
|
134
|
147
|
140
|
138
|
146
|
153
|
| Scroll 142p Acrobat Reader File (sec) |
57
|
55
|
56
|
55
|
55
|
55
|
55
|
| Encode MP3 in iTunes 4.7 (sec) |
105
|
105
|
108
|
106
|
113
|
114
|
119
|
| Export QT Movie in iMovie 3.0.3 (sec) |
35
|
32
|
33
|
33
|
33
|
36
|
34
|
| Quake III Arena 1.32 (frames/sec) |
22
|
21
|
33
|
34
|
34
|
36
|
34
|
*First number is after upgrade and second number is after applying SpeedStart.
In comparison with the last Jaguar version on our stock G4 Cube, Panther is much faster in 2D graphics (Xbench Quartz Graphics) and in finder operations (Xbench User Interface). This speed advantage continues in 10.3.7 but no faster than previous Panther versions. Startup in our G4 was much faster with Panther 10.3 than with Jaguar 10.2.8 but the 10.3.2 through 10.3.7 updates lost the startup speed advantage of 10.3. At least SpeedStart 0.1 is no longer needed since 10.3.3. Disk performance continues to be be slightly poorer with the 10.3.x updates than with 10.3 or 10.2.8. The new 10.3.7 maintains the improvement in 3D graphics performance (Quake III Arena frame rate) that began in 10.3.2. But overall, 10.3.7 is no different speedwise from 10.3.2 on the 450-MHz G4 Cube.
Here are the results for the dual 2.5-GHz Power Mac G5 comparing 10.3.7 with the 10.3.6 and 10.3.5 versions of Panther:
| Test | Dual 2.5 GHz Power Mac G5 | ||
|---|---|---|---|
| Mac OS X Version: |
10.3.5
|
10.3.6
|
10.3.7
|
| Benchmarks: | |||
| Altivec Fractal 1.1.3 (GFLOPS) |
12.7
|
12.7
|
12.7
|
|
Cinebench 2003 - Rendering
|
647
|
646
|
644
|
|
- OpenGL Hardware Lighting
|
1569
|
1587
|
1579
|
| Let1kWindowsBloom 1.0 (sec) |
11
|
11
|
11
|
| Xbench 1.1.3 - CPU |
243
|
242
|
243
|
| - Threading |
241
|
242
|
260
|
| - Memory |
340
|
357
|
335
|
| - Quartz Graphics |
300
|
300
|
342
|
| - OpenGL Graphics |
209
|
195
|
178
|
| - User Interface Graphics |
374
|
367
|
410
|
| - Hard Disk |
110
|
109
|
108
|
| "Real World" Tests: | |||
| Startup (sec) |
45
|
38
|
38
|
| Shutdown (sec) |
9
|
9
|
9
|
| Launch Classic Mode (sec) |
9
|
9
|
9
|
| Dup. 700 MB Folder 3600 Files (sec) |
62
|
58
|
56
|
| Scroll 142p Acrobat Reader File (sec) |
13
|
13
|
12
|
| Encode MP3 in iTunes 4.7 (sec) |
54
|
53
|
54
|
| Export QT Movie in iMovie 4.0.1 (sec) |
20
|
19
|
19
|
| Quake III Arena 1.32 (frames/sec) |
417
|
409
|
407
|
|
Halo 1.0.5.3/1.5 (frames/sec)*
|
57
|
58/57(63)
|
59(63)
|
|
Call of Duty 1.4 (frames/sec)
|
na
|
84
|
84
|
*Halo 1.0.5.3 was used with ATI Pixel and Vertex Shaders for 10.3.5 and 10.3.6 (first number). Halo 1.5 was used for 10.3.6 and 10.3.7 with ATI Pixel and Vertex Shaders and Advanced Shaders (number in parentheses).
The dual 2.5-GHz Power Mac G5 speed tests with 10.3.5 illustrate the superb performace of the current top of the line Mac. There's virtually no comparison in the benchmarks between the Cube and the G5 and in most of the so-called "real world" tests. Xbench scores are faster for 10.3.7 Quartz Graphics and User Interface Graphics when compared with previous versions but the OpenGL Graphics score is slower. The latter result seems unusual to us. Xbench's scores are much more variable among runs for the G5 than for the G4 so we are leery of any resulting differences on the order of 10 per cent or less. Among the "real world" tests we note that the new Advanced Pixel Shaders in Halo 1.5 appear to be consistently faster than the ATI Pixel and Vertex Shaders in the current and previous versions of Halo but not between Mac OS X 10.3.6 and 10.3.7, at least not at the resolution we use in our tests,1024x768.
In summary, the speed of the Mac OS X 10.3.7 update on our 450-MHz G4 Cube in all tests is virtually identical to that of versions back to 10.3.2. On our dual 2.5-GHz Power Mac G5 with an upgraded Nvidia 6800 Ultra DDL graphics card, the graphics seem to be faster with 10.3.7 than with 10.3.6 and 10.3.5. Reports of a slowdown in startup time with 10.3.7 that we read on the web are not borne out by our tests nor subjectively on our PowerBook G4.
We've had a few intermittant problems since applying the 10.3.7 update. Occasionally, if our dual 2.5-GHz Power Mac G5 went into sleep mode automatically, its graphics did not work on awakening and we needed to force a reboot. We have not had the problem if we put it to sleep purposefully so we have set it to never sleep automatically. This has also happened a couple of times when awakening on our PowerBook G4 from sleep mode regardless of whether it fell asleep or we put it to sleep. Others have reported this problem for previous versions of Mac OS X but we have not experienced it until installing 10.3.7. In addition, our Power Mac G5 actually dropped into sleep mode during graphics tests several times after updating to 10.3.7. It was impossible to get the G5 completely out of sleep mode without forcing a shutdown with the power button. We have not had any such occurences during normal use. [Bill Fox]
Saw something? Send a tip
The archive ran on reader tips. What did you see, where, and do you want the credit?