BARE facts on Macintosh speed FEATS

 

Small Dog is giving away a G5 Power Mac
SmallDog.com has refurbished towers, laptops, displays, etc., with 12 month warranty.

 

HOW WE TEST

Updated July 29th, 2004, by rob-ART morgan, mad scientist

 

TEST SOFTWARE FOR CPU CRUNCHING

For Adobe Photoshop tests under OS X (and Windows XP), I've created two new action files: "SP.atn" and "MP.atn" which can be run with version 6.0 as well. "MP.atn" features filters that tend to favor dual processor computers. It includes functions like Rotate, Gaussian Blur, Motion Blur, Lighting Effects, Lens Flare, Pointilize, Radial Blur, and Spherize. On the other hand, "SP.atn" uses "non-MP aware" actions like Watercolor, Accented Edges, Unsharp Mask, Convert Mode, Color Halftone, Polar Coordinates, Imagfe Size, Crosshatch, Clouds, Torn Edges, and NTSC Colors.

Both action files create a test document that is 5.71 x 3.75 inches x 700 pixels per inch or 30MB's in size. To nullify the effect of the scratch disk, I had each function UNDO itself before going on to the next. I realize that isn't real world but it's the only way to take the scratch disk speed out of the equation.

(A DEMO version of Photoshop is available from Adobe if you want to test your own machine. A Windows version of Photoshop 6 DEMO is available. )

Corel has posted a Trial version of Bryce 5. If you have downloaded it and want to render my test files (Rock_Island and Beach_Chair), email me and I'll attach them in an email. Although Corel claims it has been carbonized for OS X, but the rendering code does NOT take advantage of the G4 or multiple processors. However, it does do network rendering. A Windows version of Bryce 5 is available.

Maxon Cinema4D XL is a 3D modeling application. (See "DOWNLOADS" on their site for a DEMO copy.) For my "render logo" test, I upon the "scene0.c4d" document in the Cinebench 2000 benchmark folder and set the screen to 1024 x 768 x millions. I invoke RENDER VIEW. (Caution: leave the "scene0.c4d" file in the Cinebench 2000 folder or you will get render errors and erroneous numbers.) Cinema 4D keeps track of the rendering time to the nearest whole second. This is a cool test because when you run it on a dual processor G4, you see a visual how each cpu does part of the rendering. And if one finishes its part early, it helps the other CPU. A Windows version of Cinema4D is available. )

Maxon Cinebench 2003 was released in Feb of 2003. It's more like a real world application than a benchmark. It tests the speed at which you can render a 3D model as well as how fast you can "fly" or "walk" through or around the model. If you have multiple processors, it renders diffferent parts of the model simultaneously and visually, which makes for a great show.

There's now a G5 optimized version. A Windows version is CineBench 2003 is available.

AltiVec Fractal Carbon is a quick and dirty CPU crunch tester. It's aware of the G4's Velocity Engine as well as Multiple CPU's. It can even do network rendering. Instead of the default test with Maximum Count set to the default 4096, I recommend a 65536 setting. I "Refresh" it two or three times (Command+R) and take the fastest MegaFlop rating (and lowest time in seconds).

Apple's iMovie. Take the 6 Video Clips in the Tutorial and drop them into a New Project file. Then, without adding effects or transitions, EXPORT a QuickTime movie file in Medium Quality, 320 x 240, 15 fps. This is timed with a stopwatch.

But the test I recently devised is to take Clip 01 and 06 from the iMovie Tutorial and apply the Soft Focus effect.

iTunes-- Three Beach Boys songs ("Shutdown," "Little Deuce Coupe," "409") are converted from AIFF to MP3 "Best Quality" (192kbps). The total play time of the three songs is 332 seconds. The songs are imported from the internal hard drive instead of the CD drive. (Otherwise, you end up measuring CD drive read speed instead of CPU crunch speed.)

The FileMaker Pro test is courtesy of Paul Fabris. You can download the test files from his FMBENCH website. The test we use is 12 typical FM actions that can operate on a sample database containing from 1000 to 10,000 records. It measures the total time in whole seconds. A Windows version of FileMaker is available.

AppleWorks is used (and sometimes Microsoft Word) to run a 2D scrolling speed. I use a stopwatch to see how long it takes to Scroll through a 250 page document. The results are either listed in TOTAL SECONDS or PAGES PER SECOND. A Windows version is available for Microsoft Word (Duh).

I also use the same 250 page document (made up of many copies of "The quick brown fox jumped over the lazy dogs.") to run a Find/Replace ("dogs" to "cats") and Spell Check (find one misspelled word at the end of the document). I use a stopwatch to time it. Sometimes I combine the results on one graph. (With MS Word, turn off "check as you type" option or you'll end up in an infiite loop).

 

3D GAMES

QUAKE 3 ARENA Version 1.3.2 (available from VersionTracker) in one or more of the following modes:
1. "Fastest" 32 bit depth and textures, 640x480
2. "Normal" 32 bit depth and textures, 800x600
3. "High Quality" 32 bit depth and textures, 800x600, or 1024x768 or 1280x1024
4. "Maximum" 32 bit depth and textures, 800x600, or 1024x768 or 1280x1024 or 1600x1200 or 1920x1200
(Maximum is the same as "High Quality" setting but with Geometric Detail and Texture Detail pushed to maximum)

Default config file is used. The only tweaks in console mode ("~") is to set r_smp 1 (for MP machines) or r_smp 0 (for single cpu machines).
All other settings remain at default.


TEST METHOD: When the main screen appears, I press "~" and enter "timedemo 1" (return) and "~" once more. Then I click on DEMOS and run the DEMO. Once it finishes and returns to the main screen, I press "~" once more to get the average frames per second readout.

(A Windows version of Quake3 is available also. )

Unreal Tournament 2003 (UT2003) DEMO was released on May 9th, 2003. For more info on UT2003, visit MacSoft. To download a copy of the Mac DEMO, visit MacGameFiles.

You'll also want the app that automatically runs the benchmark samples: BenchUT2003. First launch the DEMO (or commercial version) to set video quality to maximum and screen rez to 1024x768. Then quit and run BenchUT2003.

If you have the full version of Unreal Tournament 2003, an upgrade was released in October 2003 that dramatically improves the speed of actual game play.

Unreal Tournament 2004 Full Version or DEMO by MacSoft. We use a modified version of BenchUT2003 (renamed to BenchUT2004). It only works with the Bridge-Of-Fate botmatch.

Halo for Mac has its own Time Demo test which takes you through several different recorded game sequences. We've changed the settings depending on what we're testing. See our Halo article and Radeon 9800 article for examples of settings.

 

STORAGE TESTING

DUPLICATE ONE LARGE DOCUMENT
Duplicate in Finder one very large document. Compute megabytes per second with the formula size*2/time or just note the total time in seconds. I sometimes use the pak0.pak3 file from Quake3 since it is large (457MB).

PLAY BACK UNCOMPRESSED QUICKTIME MOVIE
Using QuickTime Player 6.0 and a stopwatch, time how long it takes to play a 154 frame (5 second) 720x576 uncompressed QuickTime movie (play all frames). Calculate frames per second (total frames/time) or megabytes per second (size/time). The only hard drives that can play this in real time are the
two fastest Ultra ATA drives in a dual Hardware RAID.

PHOTOSHOP ROTATE (LOW MEMORY)
Using Photoshop 7.0, I set application size to minimum or about 37MB. I launch Photoshop and set the scratch drive to the test drive in preferences. I quit Photoshop. Then I relaunch it and I open a 45MB test file. Using a stopwatch, I time how long it takes to rotate the photo 30 degrees clockwise. The low memory allocation forces Photoshop to write to the scratch disk. Graph is posted in total seconds.

SUSTAINED READ AND WRITE
I use
Intech's QuickBench for OS 9 and Intech's QuickBench X for OS X. Version 2.0 tests at up to 100MB block size, thereby defeating the buffer size advantage.

RANDOM READ AND WRITE
In some older tests, I used the 1024K block size random read/write results from QuickBench. In most recent tests, I used an average of 128K, 256K, 512K, and 1024K block size random read/write measured by
Intech's QuickBench native OS X utility.

 

 

1995 - 2004 Rob Art Morgan
"BARE facts on Macintosh speed FEATS"
Email webmaster at
rob-art@barefeats.com

(Bare Feats is hosted on a G4 Power Mac server by MacDock.com)