How does the Blue and White G3 Perform? Benchmark & Applications Performance Tests Updated: 4/23/99

Introduction:

This will list B&W G3 performance in benchmarks and real-world applications. My results are from a Blue and White G3/350/DVD (now running at 400MHz as noted on my Yosemite Speed Settings page [voids warranty!]). Results from other models will also be listed based on reader feedback (real world comparisons are of the most interest). The latest updates are listed first. Latest Additions: OpenGL 1.0 Update: Apple (ATI does them I think) has released the first non-beta update to the OpenGL drivers. According to reports in my main site news you can use DVD with these drivers (the resource manager is now compatible I believe). Get them at http://www.apple.com/opengl/. The previous beta cost me a few FPS in games, hopefully this version corrects that. Note that it comes with QD3D v1.6. B&W G3 ROM Update: Apple has posted a G3 Firmware Update that is said to improve performance of PCI cards and more. (Quake3 Test users said it did boot framerates.) Here is the info listed on the G3 Firmware Update page "The Power Macintosh G3 Firmware Update version 1.0.2 should be installed on all blue & white Power Macintosh G3 or Macintosh Server G3 systems. Features delivered with this update include improved performance of PCI devices including built-in FireWire ports, improved support of NetBooting, and minor improvements to Open Firmware for Mac OS X Server. " Dan Knight wrote with a tip on the install:"Power Mac G3 Firmware Update 1.0.2. Instructions disagree on procedure. Read Me file indicates you should hold the programmer's button until the tone is finished (step 5). Program's internal instructions say to release the button when you first hear the tone. Waiting until the tone is done did the job for me." To those that had problems I created a ROM Update page with tips and info. 64-Bit PCI Dual-Channel U2 SCSI: Read my Exclusive First Look at the 3950U2B - Adaptec's OEM Mac 64-BIT PCI dual-channel Ultra2 SCSI card. HOT - OpenGL Beta drivers for ATI Cards! Get them at: http://developer.apple.com/opengl/. These are said to work for Rage128, RagePro and RageII (don't expect miracles on the RageII). Check the Readme before installing them (they expire on 6/1/99). More info on the OpenGL page. Please send me your feedback on using these drivers. Thanks! B&W G3 vs G3 Upgraded S900: Bus speed does seem to benefit audio apps according to this reader report: " Hi Mike, I was recently able to perform a comparison between my Umax S900 running at 450/180/56 and a BW Mac running at 450/180/100. Both used Powerlogix G3 upgrade cards, the BW using the 466 ZIF [reviewed here - running up to 550Mhz!-Mike]. I used Cubase VST playing 26 tracks of digital audio through a Sonorus StudI/O card from a U/W Cheetah. On the Umax I was able to run 4 Waves TrueVerbs (a very DSP intensive plugin) at once, while on the BW six were available, (and I thought my Umax was fast). Very impressive. [He commented on a recent Audio Column here by Thad Brown] Issue 38 was a great ad for Ensoniq (PARIS) but far from objective, and his comments about native processing were just wrong. "Those of you who don't do tons of audio work and think that a 300 Mhz G3 is fast, I suggest you try to find somebody to show you what a couple of reverbs will do to a G3. It's almost comical to watch a hot rod Mac suddenly act like an SE/30 running Photoshop 4.0." Absolute nonsense. The Trueverb is one of the nicest reverbs around and certainly uses more CPU horsepower than any plugin I am aware of, and yet the BW disposed of 6 of them without any problem and negligible interface slowdown. That's only now with existing G3 CPUs. Steinberg say they are expecting between 500 and 3000% performance improvement when the AltiVec G4 comes along compared to a same clock speed G3. The overall DSP power available from this BW exceeds that of last years ProTools (dedicated DSP cards) systems, with a complete system including monitor, audio card and HD considerably less than the cost of just the 'Tools cards. Of course more impressive still was the ZIF card at 550MHz, totally stable and fast enough with VPC to really impress my Windoze colleagues. Ah well, enough of this.....take care Mike, Greg " Apple has a page citing dramatic performance improvements with the new Blue and White G3 vs the 9600/350 in Photoshop and Quark Express. The image below is being loaded from that page: B&W G3 RAID Tests: Adaptec 2940UW firmware 3.0 vs 4.1 - A reader sends test data that indicates that sometimes the latest is not the best, at least for RAID in a B&W G3. CPU/Cache Speed Macbench 4.0 Results: Note - you can't compare results of v4.0 to the current Macbench 5.0 version. [Update: Phil later wrote back to say some disk tests were run after several MegaBytes of data were copied to the disk, which most likely affected the disk scores. Threefore the comment on Cache speed affecting disk scores is not valid, and the scores were removed.] " HI Mike Some results (MacBench 4.0) MacBench 4.0 Comparison CPU Speed Cache Ratio CPU Score FPU Score 400 MHz 2/1 1379 981 450 MHz 2/1 1551 1102 450 MHz 5/2 1445 1097 450 MHz 3/1 1369 1092 As expected, the Processor score reflected the different settings. Interesting that 450 with 3-1 cache was slightly worse than 400 2/1. [I suspect real world performance would be better at 450Mhz however, as tests with an XLR8 G3 CPU upgrade card showed 324/214 was slightly faster overall than 300/300 speeds.-Mike] [Phil late wrote that he saw a few errors at 450MHz and reverted back to the stock 400MHz speed]. Cheers...Phil 3D P.S. 31'C at 450mhz 27'C at 400mhz " As shown in the CPU Card reviews and G3 Apps Tests (see below for G3/400 results), FPU and general apps performance is most affected by CPU speed, not cache speed. Lightwave 3D 5.6 OpenGL: Below is a small sample of the layout window of LW 5.6 with the software only OpenGL patch applied. It was fluid and had good image quality when the layout options were set to preview. The scene below has a spotlight that is passing over the guitar which is illuminated in real time as you move the light source. My sources say a hardware accelerated OpenGL patch is in the works that could offer up to 2X improvements in some functions. Watch www.newtek.com for its release. Cache Mode Tests: Michiro Isobe reported that the backside cache of the Yosemite is set to Write-Through, vs. the normally faster (based on past history) Copy-Back mode. He sent a small utility to toggle the cache mode between CB and WT modes and I ran some tests to see what the differences were. Below are the results: As you can see there was little improvement with Copy-Back mode in MacBench 5's CPU/FPU tests. I then ran memory bandwith tests using the Stream benchmark and obtained the following results: Stream Test Results Function Write-Thru Rate (MB/s) Copy Back Rate (MB/s) Assignment 139.9491 135.2618 Scaling 153.8198 151.9595 Summing 156.4111 152.2090 SAXPYing 145.7323 152.2201 I was surprised to see Write-Through mode deliver generally higher scores in this test. I'll have more real world application test results soon to see if they mirror these results. Bottom line is the cache mode does not seem to be anything to be concerned about. Please don't ask me to send the utility as it was beta software and in one case crashed the Mac after switching back and forth several times. If Michiro Isobe agrees to allow it to be mirrored here I will, but I won't do that without his permission. Electric Image Tests: " Hi Mike Thought you might be interested in another real world application being run on a Yosemite 400MHZ. I ran an Electric Image rendering on both my 9600/350 and the Yos 400 (EI lets you render on multiple platforms w/o another dongle). I assigned the same amount of memory to each Camera (the rendering app). 9600/350 (180MB ram) rendered in 17m 40s Yos/400 (180 MB ram) rendered in 10m 20s. I knew it would be faster, but not by that much! Regards, Dan Baldwin " My Application Tests: The following common/popular applications are primarily CPU/FPU speed bound but were run since I have a large series of comparison scores from the G3 Apps Tests page. These results show the total time to complete renderings of sample scenes, etc.. Note that the older Mac CPU Upgrade cards with much lower bus speeds do very well, indicating bus speed is not a real factor in these tests. CPU speed is the primary performance factor, followed by Cache size and speed to a much lower extent. In some cases a PowerForce 400MHz ZIF upgrade in a beige Apple G3 or even a XLR8 400Mhz CPU card in my Genesis [9500 based] machine (nicknamed Elvis) finished the tests faster than the G3/400 Yosemite. See the CPU Card Reviews page for more info on the other cards listed. Bryce 2: Test scenes are from the Sample Scenes folder, Kai subfolder (I used these std scenes to allow other Bryce2 owners to compare their results with mine). Default 16980k memory allocated to Bryce2. Time to render the "Magical Night I" scene Yosemite G3/400 @400/200/100: 2:49 Apple G3 w/PF 400 ZIF @433/288/66 (chipped): 2:32 Apple G3 w/PF 400 ZIF @400/266/66: 2:44 XLR8 400/200/50 1MB (in Elvis/9500): 2:47 Apple G3 333/222/66 (chipped): 3:23 (no LibMoto) 9600/400 (chipped) Mach 5: 3:30 (no LibMoto) XLR8 266/177 @322/214: 3:29 MacTell 266 @300/300: 3:34 Railgun 250 @312/208: 3:37 XLR8 266/177 @300/300: 3:40 8600/350 (chipped) Mach 5: 3:41 (no LibMoto) PwrCenter w/PF G3 65/292mhz no L3 = 3:45 (65mhz bus speed) PowerForce G3 292mhz/no L3 = 3:46 (8500/45mhz bus speed) Railgun 250 @292/194: 3:46 PForce 220/110 @291/116: 3:52 PowerForce G3 285mhz w/1meg L3 = 3:52 MaxPowr 275 45/275/275 w/no L3 = 3:58 (60ns memory setting) MacTell 266 @266/266: 4:04 MaxPowr Pro+ 45/266/266 w/no L3 = 4:06 (60ns memory setting) XLR8 266/177 @266/177: 4:12 Railgun 250 @250/166: 4:24 MaxPowr Pro 45/250/166 w/no L3 = 4:28 (60ns memory setting) PForce 220/110 @240/120: 4:37 PowerLogix Pro 233 (604E@263mhz) = 4:46 Vimage G3/233/117 = 4:51 Stock PTP180: 6:37 Stock 8500/120 w/1meg L2 = 11:36 Time to render the "Alexandria II" scene Yosemite G3/400 @400/200/100: 3:51 Apple G3 w/PF 400 ZIF @433/288/66 (chipped): 3:28 Apple G3 w/PF 400 ZIF @400/266/66: 3:43 XLR8 400/200/50 1MB (in Elvis/9500): 3:45 Apple G3 333/222/66 (chipped): 4:35 (no LibMoto) 9600/400 (chipped) Mach 5: 4:39 (no LibMoto) Railgun 250 @312/208: 4:41 XLR8 266/177 @322/214: 4:46 MacTell 266 @300/300: 4:52 8600/350 (chipped) Mach 5: 5:00 (no LibMoto) XLR8 266/177 @300/300: 5:03 Railgun 250 @292/194: 5:08 PowerForce G3 292mhz/no L3 = 5:10 (8500/45mhz bus speed) PwrCenter w/PF G3 65/292mhz no L3 = 5:12 (65mhz bus speed) PForce 220/110 @291/116: 5:14 PowerForce G3 285mhz w/1meg L3 = 5:19 MaxPowr 275 45/275/275 w/no L3 = 5:29 (60ns memory setting) MacTell 266 @266/266: 5:32 MaxPowr Pro+ 45/266/266 w/no L3 = 5:39 (60ns memory setting) XLR8 266/177 @266/177: 5:47 Railgun 250 @250/166: 5:59 MaxPowr Pro 45/250/166 w/no L3 = 6:08 (60ns memory setting) PForce 220/110 @240/120: 6:16 PowerLogix Pro 233 (604E@263mhz) = 6:33 Vimage G3/233/117 = 6:34 Stock PTP180: 9:04 Stock 8500/120 w/1meg L2 = 15:32 Infini-D 4.0 Rendering Test: Time to render the "Chapter 7 Finished" scene. Options were set to: RayTracing, medium anti-aliasing, low patch detail, shadows on.(Note: version 4.01 seems much slower than 4.0 was, so results between the two versions can't be compared.) Yosemite G3/400 @400/200/100: 39:02 [v4.01] Apple G3 w/PF 400 ZIF @433/288/66 (chipped): 35:54 [v4.01] Apple G3 w/PF 400 ZIF @400/266/66: 38:37 [v4.01] XLR8 400/200/50 1MB (in Elvis/9500): 39:35 [v4.01] XLR8 266/177 @322/214: 47:23 [v4.01] Apple G3 333/222/66 (chipped): 47:23 [v4.01] MacTell 266 @300/300: 47:46 [v4.01] XLR8 266/177 @300/300: 47:54 [v4.01] Railgun 250 @312/208: 48:50 [v4.01] Railgun 250 @292/194: 51:57 [v4.01] MacTell 266 @266/266: 54:13 [v4.01] XLR8 266/177 @266/177: 57:14 [v4.01] PForce 220/110 @291/116: 58:11 [v4.01] Railgun 250 @250/166: 1:00:32 [v4.01] 8600/350 (chipped) Mach 5: 1:04:56 [v4.0] PForce 220/110 @240/120: 1:07:18 [v4.01] Vimage G3/233/117/512K: 1:10:02 [v4.0] 8600/350 (chipped) Mach 5: 1:15:55 (v4.01 was 11 minutes slower than 4.0] 8600/250 Mach 5: 1:30:23 [v4.01] Stock PTP180: 1:54:07 [v4.01] After Effects 3.1 Rendering: Small movie rendering with an explosion special effect: Yosemite G3/400 @400/200/100: 3:33 Apple G3 w/PF 400 ZIF @433/288/66 (chipped): 3:18 Apple G3 w/PF 400 ZIF @400/266/66: 3:33 XLR8 400/200/50 1MB (in Elvis/9500): 3:34 9600/400 (chipped) Mach 5: 3:38 (no LibMoto) Apple G3 333/222/66 (chipped): 4:14 (no LibMoto) PowerForce G3 292mhz/no L3 = 4:49 (8500/45mhz bus speed) PwrCenter w/PF G3 65/292mhz no L3 = 4:51 (65mhz bus speed) PowerForce 285mhz w/1meg L3 = 4:58 MaxPowr 275 275/275/45 w/no L3 = 5:05 (60ns memory setting) MaxPowr Pro+ 266/266/45 w/no L3 = 5:14 (60ns memory setting) MaxPowr Pro 250/166/45 w/no L3 = 5:37 (60ns memory setting) PowerLogix Pro 233 w/1meg L2 (604E@263mhz) = 5:34 MaxPowr Pro 250/166/45 w/no L3 = 5:38 (70ns memory setting) 8600/250 (Mach 5): 5:54 Vimage G3/233/117: 6:08 Stock 8500/120 w/1meg L2 = 12:18 MacBench 5.0 Scores: Below is a graph of the standard MacBench 5.0 tests with a 350 and 400MHz Yosemite with 6GB IDE drive. I invite owners of other Yosemite systems (different speeds/SCSI models) to send me their MacBench 5.0 results file (make sure you set the video to 1152x870, millions colors for the graphics tests as that is the settings shown here and with the baseline G3/300 system) The 400Mhz scores were run with the latest ATI beta drivers for the Rage128 (changes were to 3D related peformance, so 2D video scores were basically unchanged from the shipping version) I've also received a limited test (CPU/FPU) on a 400MHz Yosemite overclocked to 450MHz: I'll be adding other test results on scrolling speed in Photoshop and Word (much improved with the new B&W G3) and other reader comments in the future. For owner reports on overclocking - check the searchable database (select BW G3 as the Mac model) Related Links: Exclusive First Look at the 3950U2B - 64-Bit PCI, dual-channel U2 SCSI G3/400 Photoshop Performance:I've updated the Photoshop 5 Performance page to include results with the Yosemite overclocked to 400MHz. Yosemite Game Performance: This page is dedicated to game performance and tips on getting the most from the new Mac in games like Unreal, Quake and others. Now includes results of tests with the latest Beta Rage128 drivers and Unreal beta version. Also contains scores from a 8MB Voodoo2 card in the Yosemite. Yosemite Tips/Specs/Compat. Notes: My original page of Yosemite specific info. CPU Speed Settings Explained: I've posted a page of Lab Notes: Yosemite Speed Settings with details on the jumper block settings and results of tests with a 350 and 400MHz system at 400 and 450MHz respectively. (This is a warranty-voiding procedure and I do not recommend it. It may cause loss of data or hardware failures.) Rage128 Driver Update Feedback Compatibility Reports on the new B&W G3 (software, hardare with notes on firmware updates, etc.) Internal Modem Issues/Tips IDE Slave Drive Upgrade [Ref. Only - may void warranty]

This page will be updated with additional information in the future. If you have comments or input on the new Blue and White G3's please contact me.

Back to G3-ZONE for more Apple G3 Info or Back to XLR8YOURMAC.COM (If you're trapped in a frames site - click here to clear frames)

Copyright © 1999 No part of this sites content is to be reproduced in any form without written permission. All brand or product names mentioned here are properties of their respective companies. Users of this web site must read and are bound by the terms and conditions of use.