About the Tests:
PSBench's 21 filter action script was created as a cross-platform test of Photoshop performance, long before the G4 CPU based Macs were available. Only a few of the 21 filters in the test are accelerated by Altivec (Lighting Effects benefits most, about 4x faster). The overall score is heavily weighted towards non-Altivec supported filter functions (the filters that took longest to run don't benefit from Altivec), so the total test times don't show a large difference between a G3 or G4 of the same clock speed. There is enough Altivec gain that the PB G4/400 was just a bit faster than a PB G3/500 Firewire model in this test.
I could have only shown results with Altivec enhanced filters which would have shown a dramatic difference between G3s and G4s, but used PSBench since that is the Photoshop benchmark I've used in past G4 CPU Upgrade reviews and system comparisons. It also illustrates that not all filters benefit from Altivec, so you can see which do and decide if those filters are ones you use frequently. (As an aid in deciding if you'd benefit from a G4 system or upgrade.) See the table below the graph for filter-by-filter test results.
(PB G4/500 time = 98.9 sec. See below)
Only the two PowerBook G4s were running OS 9.1, other Macs used OS 9.04, which may account for some of the differences such as why in one filter the PB G4/500 was faster than a dual G4/450. (Perhaps that filter doesn't use both CPUs and maybe the later revision G4 CPU chip in the PB helped). Of course for things like games, the desktop systems would be faster than a portable. The desktops also have drives with faster sustained rates, but these tests are primarily CPU bound since PShop is given enough RAM to not use the swap file (PShop's virtual memory).
Filter-by-Filter Comparisons:
A web page isn't wide enough to show every system tested, so I selected several for a filter-by-filter comparison including both a PB G4/400 and PB G4/500. (This also allows you to compare the PB G3/500 to the PB G4/400 or /500 performance filter-by-filter.) The table below shows the times in seconds to perform each of the 21 filter operations on a 10MB image. (Fastest filter times are in bold.)
|
PS5Bench Test
(10MB Image File)
|
HP Pavilion 1G
1GHz PIII CPU
256KB L2 at 1GHz
133MHz bus
256MB PC800 RAMBUS
Geforce 2 AGP (64MB)
Win98 SE
VM On |
Apple G4/500 AGP (rev 1)
500MHz G4 CPU
1MB L2 at 250MHz
100MHz Bus
256MB SDRAM (222)
Rage128 Pro AGP (16MB)
OS 9.04
VM OFF |
Apple G4/450 MP
Dual 450MHz G4 CPU
1MB L2 at 225MHz x2
100MHz Bus
256MB SDRAM (222/322)
Rage128 Pro AGP (16MB)
OS 9.04
VM OFF |
PowerBook G4/400
400MHz G4 CPU
1MB L2 at 200MHz
100MHz Bus
256MB SDRAM (222)
Rage128 Mobility (8MB)
OS 9.1
VM OFF |
PowerBook G4/500
500MHz G4 CPU
1MB L2 at 250MHz
100MHz Bus
256MB SDRAM (322/222)
Rage128 Mobility (8MB)
OS 9.1
VM OFF |
|
Rotate 90
|
0.2
|
0.2
|
0.2
|
0.2
|
0.2
|
|
Rotate 9
|
1.6
|
2.0
|
1.3
|
2.2
|
2.0
|
|
Rotate .9
|
1.5
|
1.9
|
1.2
|
2.0
|
1.8
|
|
Gaussian Blur 1
|
0.5
|
0.5
|
0.4
|
0.5
|
0.4
|
|
Gaussian Blur 3.7
|
1.3
|
1.4
|
0.9
|
1.5
|
1.3
|
|
Gaussian Blur 85
|
3.2
|
1.6
|
1.1
|
1.8
|
1.6
|
|
Unsharp 50/1/0
|
0.6
|
0.7
|
0.5
|
0.7
|
0.7
|
|
Unsharp 50/3/7/0
|
1.4
|
1.7
|
1.0
|
1.8
|
1.6
|
|
Unsharp 50/10/5
|
1.6
|
1.7
|
1.1
|
1.8
|
1.6
|
|
Despeckle
|
1.7
|
0.8
|
0.5
|
0.8
|
0.7
|
|
RGB-CMYK
|
2.6
|
4.0
|
4.4
|
4.9
|
4.0
|
|
Reduce Size 60%
|
0.6
|
0.4
|
0.3
|
0.4
|
0.3
|
|
Lens Flare
|
2.3
|
3.3
|
2.2
|
3.9
|
3.5
|
|
Color Halftone
|
3.5
|
3.0
|
3.1
|
3.5
|
2.8
|
|
NTSC Colors
|
3.2
|
3.4
|
3.7
|
4.2
|
3.4
|
|
Accented Edges
|
7.4
|
8.9
|
9.6
|
10.7
|
8.8
|
|
Pointillize
|
8.7
|
12.5
|
7.9
|
15.5
|
12.2
|
|
Water Color
|
15.8
|
18.7
|
20.2
|
22.6
|
18.5
|
|
Polar Coordinates
|
5.7
|
2.9
|
2.3
|
3.4
|
2.9
|
|
Radial Blur
|
21.2
|
29.9
|
16.7
|
35.4
|
28.8
|
|
Lighting Effects
|
1.2
|
1.7
|
1.2
|
2.0
|
1.8
|
|
PS5Bench Index
(time to complete)
|
85.8
Seconds
|
101.2
Seconds
|
79.8
Seconds
|
119.8
Seconds
|
98.9
Seconds
|
|
System
|
HP Pavilion 1G
1GHz PIII CPU
256KB L2 at 1GHz
133MHz bus
256MB PC800 RAMBUS
Geforce 2GTS AGP (64MB)
Win98 SE
VM On |
Apple G4/500 AGP (rev 1)
500MHz G4 CPU
1MB L2 at 250MHz
100MHz Bus
256MB SDRAM (222)
Rage128 Pro AGP (16MB)
OS 9.04
VM OFF |
Apple G4/450 MP
Dual 450MHz G4 CPU
1MB L2 at 225MHz x2
100MHz Bus
256MB SDRAM (222/322)
Rage128 Pro AGP (16MB)
OS 9.04
VM OFF |
PowerBook G4/400
400MHz G4 CPU
1MB L2 at 200MHz
100MHz Bus
256MB SDRAM (222)
Rage128 Mobility (8MB)
OS 9.1
VM OFF |
PowerBook G4/500
500MHz G4 CPU
1MB L2 at 250MHz
100MHz Bus
256MB SDRAM (322/222)
Rage128 Mobility (8MB)
OS 9.1
VM OFF |
I can only guess the later revision of the G4 (a 7410) and/or the OS version was why the PB G4/500 was slightly faster than my G4/500 (rev 1) system.
PB G4/400 - Mixed 322/222 RAM vs All 222 RAM:
The PowerBook G4 400Mhz results were from using a single 256MB 222 SODimm module. The total time for the 21 filters with the original 128MB 322 and a 128MB 222 SODIMM was 122.4. Running all 222 RAM resulted in just over 2% total performance increase overall in this test. The biggest gain seen in any filter was the Radial Blur - which took 36.3 seconds with the mixed 322/222 SODIMMs, and 35.4 seconds with 222 only RAM. Most other filters showed a 0.1 second reduction in times from using all 222 timing RAM.
Test Settings:
Photoshop/System settings are 1024x768, millions colors, VM off, Interpolation set to bicubic (better) and Photoshop should be allocated enough RAM to avoid any swap file activity from the 10MB test image filter actions. (128MB was allocated to Photoshop 5.5.) Photoshop 5.5's History' settings were changed from the default 20 to 1 and I unchecked the 'automatically create snapshot' option which helped make the 3 runs for each filter more uniform in time taken.
All systems were running Photoshop 5.5 (G4s used the Adobe Altivec Core Extension and Lighting Effects Plugin). Windows background tasks like McAfee's anti-virus and the task scheduler were disabled. The Windows swap file was set to the same min/max size (384MB) - a common tip to improve performance by preventing Windows from resizing the swap file during apps use).
For more info or to download the PSBench action file, see PSBench Home Page. (There may be a later version of the action script there now, I've used the same one for about a year now to make sure all tests used the exact same script.) | 
