The Source for Mac Performance News and Reviews

Review: Powerlogix's G3/800MHz ZIF CPU Upgrade Tests in B&W G3 vs. various G3 and G4 CPU Upgrades By Mike B. Published: 3/25/2003 (Updated 4/3/2003 for OWC G4/550 Tests)

Real World Performance Tests

Intro | Benchmarks  | Apps/Game Tests | Software Controls | Installation | Specs/Design

Applications and Game Performance

This page lists test results in common Mac applications like iMovie2 (OS 9), iMovie3 (OS X), iTunes (OS X and OS 9), Photoshop 5.5, Word 98 Scrolling, Stuffit 7.0.1 compress/expand tests and popular 3D games like Quake3 (Classic and OS X) and Unreal Tournament. Although my later AGP upgrade reviews use Photoshop 7.0 (and a 50MB test image), this system had only PS 5.5 installed which had been used for tests of past ZIFs in this system and therefore their scores could be used for comparison. (Updated results to include tests with OWC G4/550 ZIF at up to 600MHz.) Logic Audio Users Note: Although I do not own/use Logic Audio, several readers that do have reported Logic Audio issues (pops/noise) with G4 upgraded PCI Macs in the past with pre-OS X versions of Logic. (reports in that article and in CPU upgrade database reports here - although the first OS X Logic report didn't note the problem). A G3/800 probably is a better choice for PCI macs running Logic Audio under OS 9 for instance.

iMovie DV to Quicktime Movie Export Tests

I used the same test I have since iMovie was released - stacking the 6 tutorial file clips end-to-end (no transitions) and timed how long it took to export the movie using the standard "CDROM Medium" settings. (By using the tutorial with no variables like transitions, it's something everyone can easily test with their own systems.) iMovie 2 (v2.1.1) for OS 9 and iMovie3.02 for OS X 10.1.5 were used. (QT 5.02 for OS 9, QT 6.1 for OS X was used.) Note: The graph also includes a test with the G3/800 set to 900MHz via Powerlogix's Software Control just as a FYI. 900MHz did not prove to be stable in extended use with this sample and stock heatsink however, apps would quit when launched after about 1/2 hour of use at 900MHz. Just to see how a 750FX G3 w/512KB on-chip cache (full CPU speed) compared to the G3/500 w/slower 1MB external L2 cache at the same CPU speed, I set the G3/800 to 500MHz using PL's control. The 750FX @500MHz completed this test in 224 seconds (about the same as the older generation G3/500 w/1MB external L2). I also ran the same test under OS 9.2 with iMovie 2 (graph includes results from previous OS 9.2/iMovie2 tests with more upgrades). Although the G3/800 showed a large gain in performance over the G3/500, a G4 CPU is clearly a benefit for this Digital Video to Quicktime conversion.

iTunes MP3 Conversion

Time to convert an Audio CD (6 min, 30 second) song to MP3 (192Kbs rate quality setting) using the original OEM 6x DVD ROM drive/FW 48x read speed drive. (Outermost song track used to try and minimize the effect of the slow drive, although a test with a faster Firewire 48x CDRW drive cut times dramatically with the G3/800 in OS 9 tests as shown below.) OS X iTunes3 tests showed no benefit in this test from the FW vs OEM DVD ROM drive, although iTunes2/OS 9 tests did. (Note: The graph also includes a test with the G3/800 set to 900MHz via Powerlogix's Software Control, although 900MHz wasn't stable for more than 1/2 hour or so of use.) As with the DV-QT movie export, the G4 CPUs do well against the G3/800 even with a much slower speed (but has twice the L2 cache size of the G3/800, although the G3/800's cache is running at full cpu speed). I also ran tests in OS 9.2.2 with iTunes 2.04 (last version for OS 9) with the same song, but unlike OS X, in OS 9 performance was better with a fast Firewire CDRW (40x12x48x rated) than the internal OEM 6x DVD ROM drive: Even in OS 9/iTunes 2 MP3 encoding, the G4 does well at much lower clock speeds than the G3/800, but then Powerlogix admits on their G3/800 product page that a G4 is better for MP3 encoding. The original 6x DVD ROM drive is a clear bottleneck in OS 9/iTunes tests, even when using the outer track.

Stuffit 7 Compress/Expand Tests

A reader with clients that use OS 9 exclusively asked that I test with Stuffit 7.0.1 to time compressing/expanding a 64.6MB file. This B&W G3 OS 9 disk is an older Maxtor 6800 series (very slow compared to today's drives) and fragmented, so to eliminate the effect of the hard drive, I created a RAM disk and ran the Stuffit tests from the RAM disk. (Shorter bars are faster.)

PhotoShop 5.5 Tests

Note: Although my reviews since August 2002 have used Photoshop 7.0 and a 50MB test image, to compare the G3/800 ZIF's performance to the fall 2001 ZIF upgrades tests, I used the the same copy of Photoshop 5.5 and the PSBench action script from that older review (10MB image size). (You can't compare results with PS 7.0/50MB image tests in other reviews to these results - much larger image file and PS 7 has better G4 support than 5.5 had, even with the Altivec plugin and lighting effects filter updates for PS 5.5.) 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 with Photoshop 5.5 are accelerated by Altivec (Lighting Effects benefits most, about 4x faster). The overall score is therefore 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. However since users don't always use Altivec aware filters all the time, the results are still a valid comparison of system performance for the filter functions listed. The graph below shows a comparison of the B&W G3 with OWC's G4/400, G4/500, G3/500 and Powerlogix's G3/800 upgrades in Bold, with some other systems for comparison, based on results from past reviews here that used Pshop 5.5/10MB image tests. (Arrow in graph shows PL G3/800 results which was appx 20% faster than the G4/500 ZIF upgrade in this test/system - see table below for more details.) Filter-by-Filter Performance Comparisons: Rather than just a simple graph showning total times for the filter series, I've listed each filter, its description and the time each upgrade took to complete it. (Each filter is run 3 times, the avg. is displayed in the table below. PhotoShop's timing function is used, not a stopwatch.) This also allows you to compare performance filter-by-filter to see the difference by function, not just total time. The table below shows the times in seconds to perform each of the 21 filter operations on a 10MB image. Of all the ZIFs tested in the B&W, the PL G3/800 was the fastest overall in this test (but again it uses OS 9/Photoshop 5.5, not the latest Photoshop 7.0 under OS X. The G4/550 was reliable in this system at 600MHz, but results are not included here since most PShop users probably would not overclock the CPU and it saved some page width and time.) Photoshop 5.5 Filter Test Details PS5Bench Test (10MB Image File) B&W G3 350MHz G3 CPU 1MB L2 at 175MHz 100MHz bus 256MB PC100 (222) Radeon PCI OS 9.2.1 VM Off B&W G3 500 MHz G3 CPU 1MB L2 at 250MHz 100MHz bus 256MB PC100 (222) Radeon PCI OS 9.2.1 VM Off B&W G3 400MHz G4 7400 CPU 1MB L2 at 250MHz 100MHz bus 256MB PC100 (222) Radeon PCI OS 9.2.1 VM Off  B&W G3 500MHz G4 7400 CPU 1MB L2 at 250MHz 100MHz bus 256MB PC100 (222) Radeon PCI OS 9.2.1 VM Off B&W G3 800MHz G3 750FX CPU 512KB L2 at 800MHz 100MHz Bus 256MB SDRAM (222) Radeon PCI OS 9.2.1 VM OFF Rotate 90 0.9 0.7 0.3 0.2 0.4 Rotate 9 3.4 2.5 2.5 2.0 2.0 Rotate .9 3.1 2.3 2.3 1.9 1.8 Gaussian Blur 1 1.5 1.1 0.7 0.6 1.1 Gaussian Blur 3.7 3.7 2.8 2.3 1.7 2.2 Gaussian Blur 85 4.7 3.6 2.4 2.1 3.1 Unsharp 50/1/0 1.9 1.5 1.1 1 1.2 Unsharp 50/3/7/0 4.1 3.1 2.3 2.0 2.5 Unsharp 50/10/5 4.1 3.1 2.4 2.0 2.5 Despeckle 2.8 2.0 1.1 1.0 1.5 RGB-CMYK 5.8 4.1 5.0 4.0 2.5 Reduce Size 60% 1.4 1.0 0.5 0.4 0.8 Lens Flare 4.9 3.5 4.2 3.4 2.4 Color Halftone 4.4 3.2 3.9 3.2 2.5 NTSC Colors 4.8 3.4 4.3 3.4 2.3 Accented Edges 12.6 8.9 10.9 8.8 5.7 Pointillize 18.0 12.5 16.0 12.7 8.1 Water Color 26.6 18.6 23.1 18.5 12.0 Polar Coordinates 6.1 4.4 3.7 3.1 3.4 Radial Blur 40.9 29.1 35.8 29.1 20.5 Lighting Effects 10.4 7.3 2.1 1.8 4.4 PS5Bench Index (time to complete) 166.1 Seconds 118.7 Seconds 126.7 Seconds 102.9 Seconds 82.9 Seconds System B&W G3 350MHz G3 CPU 1MB L2 at 175MHz 100MHz bus 256MB PC100 (222) Radeon PCI OS 9.2.1 VM Off B&W G3 500 MHz G3 CPU 1MB L2 at 250MHz 100MHz bus 256MB PC100 (222) Radeon PCI OS 9.2.1 VM Off B&W G3 400MHz G4 CPU 1MB L2 at 250MHz 100MHz bus 256MB PC100 (222) Radeon PCI OS 9.2.1 VM Off B&W G3 500MHz G4 CPU 1MB L2 at 250MHz 100MHz bus 256MB PC100 (222) Radeon PCI OS 9.2.1 VM Off B&W G3 800MHz G3 750FX CPU 512KB L2 at 800MHz 100MHz Bus 256MB SDRAM (222) Radeon PCI OS 9.2.1 VM OFF You can download the latest PSBench action script at the PSBench home page.

Word 98 Scroll Tests

(Again past upgrade test results from 2001 used Word98, so it was also used to test/compare the G3/800's performance.) I measured the time it took to scroll the a 6.9MB document file (2,187 pages, 45K paragraphs, 144,799 lines) with multiple font sizes, etc. from the first page to the last. Word was allocated 40MB (many times the document size). Display was set to 1024x768, millions colors. (Apple LCD 15" display limit.) The results are shown in the graph below. All times are in seconds rounded to one decimal place. Remember these tests were timed with a stopwatch, so there differences of 1/2 a second or so can result from variations in starting and stopping the stopwatch for each run. The G3/800 clearly shows its higher clock speed advantage.

Time to Boot OS 9.2.1/OS X 10.1

Although it's not a major issue for me personally, I measured the time it took to boot from a cold start. (Time from the startup bong to having the OS ready to use.) Memory test was not disabled and no other PCI cards other than the Radeon were installed. All times are in seconds rounded to one decimal place. Remember these tests were timed with a stopwatch, so there differences of 1/2 a second or so can result from variations in starting and stopping the stopwatch for each run. I'm not sure why the G4s were faster at OS X booting than G3s (MHz/MHz) but there seems to be a lot more variation in booting times to OS X than OS 9 in general (for example varying times in network initialization during X loading). (Since I had installed other software on the drive, boot time tests were not run with the G4/550 upgrade.) OS 9.2.x takes longer to boot on all my Macs than previous OS versions (the first extension causes a long delay). Both these drives are not as fast as the current generation of IDE drives. (They're 1999 models, OS X on the original 5400 OEM Maxtor 6GB drive and OS 9.2.1 on a 6800 series Maxtor 27GB.)

Game Performance Tests:

Tests were done with Quake 3 v1.30 (OS 9 and OS X - Q3 1.32 for OS X was actually about 10% slower than 1.30 in this PCI based single CPU system). Unreal Tournament version used was the last update that I know of (v436). With the same game settings (detail settings/config file) the CPU load is the same across resolutions, so I'm showing 1024x768 mode tests just as a FYI on the PCI Radeon performance with the various CPUs tested in the B&W G3. (The Apple 15" LCD used with the system has a max resolution of 1024x768.) Tests were run with the "Fastest", "Normal" and "High Quality" Quake3 options with resolution adjusted as needed. (Using custom config files can increase these rates at the expense of image quality/geometric detail. I used the builtin standard quality options (only varying resolution) so that you can easily compare your results using those standard in-game options.) (Note: G4/550-600 tests could not be run with the OS X Q3 1.30 version used for past tests as I had inadverently overwritten v1.30 with v1.32, which runs about 10% slower than 1.30 on this system.) Modern/faster AGP based macs (with faster/dual CPUs, much faster graphics cards, etc.) deliver much higher framerates in Quake3 of course than the PCI based Macs. (See examples in other G4 reviews.) Unreal Tournament: Using v436 in OS 9.2.1. Medium Quality settings, low-audio quality, music volume set to 0, decals and dynamic lighting enabled. "Min Desired FPS" set to 0. Rave Mode used since it runs better than OpenGL mode on the ATI cards in my tests. PCI Radeon in 66MHz PCI slot used for all tests. This is the original PCI Radeon, not the later/current Radeon 7000 model. (Note: Due to other review/tests on my plate, I didn't retest UT with the G4/550-600MHz yet.) I haven't used th Cityintro benchmark for years since it's just a flyby, but did in the 2001 review so I'm including the G3/800 results here as a FYI. The UTBench demo below is much more demanding (multiple players, heavy firing, etc.) UTBench always reports low AVG rates for Macs in my experience (even the fastest Macs I have usually average under 35FPS on this test. As an example, a GeForce 3 AGP with 1GHz PL CPU upgrade tested in my Cube had an average UTBench FPS rate of 29.7. (typical game framerates are much higher, depending on how much action/firing, etc. is happening.)

The next page covers Powerlogix's software control for the G3/800, which allows CPU speed changes on-the-fly (a unique feature of the G3 750FX). Or you may use the links below to jump to a specific page.

Index of PL G3/800 Review Pages Intro | Benchmarks  | Apps/Game Tests | Software Controls | Installation | Specs/Design - or - Back to WWW.XLR8YOURMAC.COM

Copyright © 2003 All rights reserved All brand or product names mentioned here are properties of their respective companies. Users of the web site must read and are bound by the terms and conditions of use.