| System Settings |
We use the following settings for all of our test runs to assure consistent results:
- Memory - Virtual memory is turned off and the disk cache is set to 2048K (2 MB)
- Networking - AppleTalk and file sharing are both turned off.
- Extension Set - We use the "OS 9.0.4 All" set plus any needed by the upgrade card or product being tested.
- Monitor Settings - We set the color depth to "millions" with the resolution to 1024x768 unless noted otherwise. Our tests indicate that color depth has a far greater impact on performance than resolution.
- Font Smoothing - Is on for all tests.
|"Real World" Application ||Test Description and Settings Used |
Kindly provided by Apple Computer, Inc.
We use AppleWorks 6.0.4 for our Scroll test as well as our Search & Replace test. The test file is a 522 page word processing document that contains a mix of text and graphics. The document window is first maximized and then set to 5" height. Window size has a direct impact on the scroll test score.
Scroll Test - We clocked the time to scroll through the document from top to bottom using the down arrow on the scroll bar.
Search & Replace Test - Starting at the top of the document we issued a "replace all" command replacing all occurrences of the word "the" with the word" macbench." 12,900 items in all.
Analog is a popular shareware program for analyzing a web server's log file. Log files provide web masters with valuable information about traffic on their site, which pages are popular, where visitors are coming from etc. The log file we use is an older one from our own site containing over 2 million lines and roughly 18 MB in size.
Log Analysis - We use Analog's default settings, clocking the time to analyze the file and generate a report.
We run two distinct test suites for our Photoshop benchmarks. Tests are run on either a 10 or 20 MB file depending on the amount of RAM available to allocate to Photoshop. Unless noted otherwise, we allocate 200 MB of RAM to Photoshop to assure that the image is at 100% efficiency and the hard drive is not being accessed during the test run. We clocked the time to complete all filter operations for each suite, quitting and restarting Photoshop between the two tests.
Launch Time - Time to launch the program by opening a 20 MB test file. We stop the clock when the picture appears.
AltiVec Filter Suite - Is comprised of the following filter opperations: Lighting Effects, Spherize, Twirl, and Gaussian Blur
Standard Filter Suite - Is comprised of the following operations: Colored Pencil, Bas Relief, Film Grain, Craquelure, and Convert Mode from RGB to CMYK.
Quake III Arena
While we don't care for the "first person shooter" genre of games, they do push the graphics subsystem and therefore make for a good benchmark. We use the following sequence of commands to get the "frames per second" score.
From the main screen press ~ (the tilde key) to access the text entry area.Then enter "timedemo 1" followed by the return key. Then ~ once more to hide the text window. We then run Demo1 from the "demos" area. Once it finishes and returns to the main screen, press ~ again to get the frames per second rate.
"GL Driver" at default
"GL Extensions" are on
"Video Mode" is 512x384
"Color Depth" is 16 bit
"Fullscreen" is on
"Lighting" is Vertex
"Geometric Detail" is Low
"Texture Detail" slider is set to 1/3
"Texture Quality" is 16 bit
"Texture Filter" is Bilinear.
High Quality settings
"GL Driver" at default
"GL Extensions" are on
"Video Mode" is 800x600
"Color Depth" is 32 bit
"Fullscreen" is on
"Lighting is Lightmap"
"Geometric Detail" is Medium
"Texture Detail" slider is in the middle
"Texture Quality" is 32 bit
"Texture Filter" is Trilinnear.
Frames Rates - Are significantly impacted by the settings mentioned above. We chose the highest and lowest quality settings to give you an idea of the range of frame rates you can expect.
Cinema 4D XL
We used the Cinema 4D XL program for our rendering test. Like Photoshop, Cinema 4D XL is both multiprocessor and AltiVec aware. The level of AltiVec acceleration depends on the type of scene being rendered with acceleration ranging antwhere from 0 - 50%. The test files mentioned below do not take particular advantage of the AltiVec instruction set.
Render Test - For the render test we did a "batch render" job on three files included with the demo, "inv-vol.c4d", "MotionSequencing.c4d" and "Target.c4d". Two are animations and one is a still. We clocked the time to render all three from start to finish.
SoundJam MP 2.5.2
SoundJam, the ubiquitous, MP3 player and encoder, is yet another application that is both multiprocessor and AltiVec aware.
MP3 Encode - We encode in "joint stereo" mode at 128kbps with "sample rate" set to auto and "performance" set to best quality. We recorded a track from CD that was 4 minutes 26 seconds long.
Stuffit is the undisputed standard for decompressing files downloaded off the internet or other sources. We use the freely available shareware versions of Stuffit 6.
Decompression Test - For this test we launched Stuffit Expander and selected a 4.6 (compressed size) file for expansion.
|QuickTime Encode - For this test we used a 4.7 MB movie trailer and encoded it using the Sorenson 2X CD setting. |
Finder Copy (file) - For this test we use the 128 MB Photoshop file from our Photoshop tests. The file is selected and the "duplicate" command issued from the "file" menu. We immediately delete the duplicated file and empty the trash to reclaim the space consumed.
Finder Copy (folder) - For this test we use a 60 MB folder containing 1,107 files. The folder is selected and the "duplicate" command issued from the "file" menu. We immediately delete the duplicated folder and empty the trash to reclaim the space consumed.
|Test Machine ||Configuration |
OS - 9.0.4 unless otherwise noted. Only OS extensions are running during our tests.
RAM - 320 MB installed with virtual memory turned off for all tests.
Disk cache - Set to 2MB (2048K) except for the MacBench tests where it was set to 512K to bring it in line with the MacBench base machine.
Hard Drive - 4GB. we defragment the drive before testing each upgrade card as each test run generates a number of files that are later removed from the drive.
AppleTalk/File Sharing - Are both turned off during all tests naturally.