Accelerate Your Mac! - PowerForce G4 350MHz Review

Click for iPod Batteries and Accessories!

A Click shows your site support to my Sponsors

The Source for Mac Performance News and Reviews

PowerForce G4 card
Review: PowerForce G4 350MHz
G4 CPU Upgrade for PCI Macs
By Mike
Published: 1/23/2000

MacBench 5.0, G4TimeDemo and Memory Benchmark Test Results

Intro | Benchmarks | Appl. Tests | Software Controls | Installation | Specs/Design | Summary

Benchmark Tests

Benchmark tests were run with MacBench 5.0 as it is the accepted Mac standard, however the current version does not use the G4's Altivec extensions and therefore will not show the performance improvement they provide for software that uses them. I've also included results with two memory bandwidth benchmarks and with G4Timedemo, one of the only benchmarks I know that utilizes the Altivec ('Velocity Engine') of the G4.

Remember the most important results are on the next page of the review which covers real world applications performance. This is where the rubber meets the road so to speak. Benchmarks have their place, but actual applications performance is far more important that pure benchmark scores.

As you'll see in the G4timedemo test results (and the Photoshop 5.5/Soundjam MP results on the Applications tests page), until benchmarks like MacBench take advantage of Altivec, they are not an indicator of what the G4 is potentially capable of. However, MacBench CPU scores do seem to be a pretty accurate gauge of G4 performance in non-Altivec aware applications.

Be aware that Macbench Graphics scores will vary depending on graphics mode and the installed video card. In Macbench 5.0 a 1000 score is the baseline based on performance with an Apple Beige G3/300 running millions colors, 1152x870, so consider this when evaluating any scores at lower resolutions and color depths. Likewise the disk scores will vary depending on drive and interface used (onboard or PCI SCSI, IDE, etc.), and even the free space and fragmentation present on the test disk. (The fastest areas of a disk drive are the outer tracks, so as a disk fills up it becomes slower, and data often becomes fragmented, further slowing down performance.) MacBench disk scores are highly affected by disk cache size set in the Memory Control Panel as well (improvements generally seen up to about 5-6MB cache size, where diminishing returns occur). In OS 8.5 and later, the disk cache defaults to a percentage of installed RAM. Earlier versions defaulted to very low settings such as 96K (increasing disk cache size was one of the tips in my Performance Basics article from the OS 7.6 era.)

MacBench 5.0 Tests: The graph below compares the follow configurations and systems

The base 9600/350 system (350MHz 604ex CPU w/1MB on-card L2 cache)
The 9600 with the PF G4/350 upgrade running at rated (350MHz) CPU speed and at 400MHz
(Overclocked - not recommended. Other samples may not be reliable at beyond the rated speeds.)
The 9600 with past results with an XLR8 G4/350 upgrade installed.

All tests used the same base system, OS 9, 320MB of RAM (interleaved), VM Off, ATI 4.2 drivers, OpenGL 1.1.2. All G4 upgrade tests had the 4 OS 9 Altivec extensions active.

MacBench 5.0 Results

Notes: Some explanation of the MacBench graph and test components.

CPU/FPU Scores: MacBench CPU scores will disappoint most people that have heard so many marketing comments on the 'supercomputer' G4 chip. The G4's FPU scores are really no higher than the 604 at the same clock speed, as it has basically the same FPU design. (The G3 however does not have as powerful a FPU design as the 604 or G4, it's based on the 603e CPU, with a backside cache)

Software: Since G4 control software often modifies the supervisor registers, it can have an effect on performance so later versions of control software may perform differently than the versions I used. The results above used PowerLogix G3/G4 Cache Profiler v1.3 and their Altivec extension v1.0. The XLR8 G4 tests used XLR8 Control Panel and Extension v1.4.3b0 (a later version exists now and may show different results).

Speculative Processing was disabled to better reflect what the typical older Mac results would be (the base 9600/350 is one of the few Mac models with a ROM version that allows Speculative Processing to be enabled without any negative effects).

Disk Scores: The 9600/350 test system was running OS 9 from a Ultra2 ATTO PCI SCSI card driving a Seagate Cheetah U2 4.5 GB hard drive that was approximately 80% full and not optimized. If the disk had more free space or was defragmented (optimized) scores would have been higher. Due to installing control software for different cards, the disk scores vary a bit from past G4 upgrade tests shown in the graph as the disk cannot be in the exact same state (free space, etc.) as prior tests.

Graphics Scores: The 9600/350 used a OEM (rev 1) Rage128 graphics card taken from a early B&W G3. This card has a slower clock speed (about 10%) than the retail Rage128 cards or later rev 2 B&W G3 systems, so the graphics scores are a bit lower than they would be with a retail Rage128 card. ATI's 4.2 universal installer drivers along with the final release of OpenGL 1.1.2 were used. See my video cards page for reviews of faster alternative cards for these systems.

G4Timedemo Benchmark:

G4timedemo is a benchmark for showing the benefit of Altivec support with G4 CPUs. I ran tests with each of the system/upgrade combos and graphed the results. As you can see, a G4 CPU w/Altivec support does make a huge difference here (60% gain with a G4 CPU of the same speed). All tests were run with the desktop video mode of 1024x768, thousands colors and G4timedemo set to 'Amazing' (max) quality.

G4timedemo results seem to vary a few FPS from run to run but you can clearly see the dramatic difference a G4 w/Altivec makes in performance. I repeated the tests several times (each after a clean reboot) and noted the best score of 3 runs. Since OpenGL 1.1.2 contains some Altivec optimizations, games will usually some some benefit of a G4 CPU as well (if the video card is not already saturated and a bottleneck of course) .

Memory Bandwidth Tests:

I want to caution readers to not be too concerned with the results below. Although the 9600's Kansas motherboard shows rates of about 1/3 that of a B&W G3 or later Mac, it's often very close to those systems in real world applications performance if running the same speed and type of CPU.

The efficiency of the L1 and L2 caches in CPUs are the reason why a 50MHz system bus speed older Mac with the same type and speed of CPU upgrade generally keeps pace with modern Macs that have 100MHz bus speeds in most applications tests. (Applications that move a lot of data over the bus are exceptions - Photoshop, 3D Games, etc.).

Due to workload and the fact this review was overdue, I don't show the results of Stream tests that I have used in previous G4 reviews. Copy-Back (often called Write-Back) cache mode was used for this test, but the slower Write-Through mode may be required for reliability in some Mac models (in 9500's for instance, based on some troubleshooting reports). Write-Through cache mode would show much lower scores for write operations since it basically negates the effect of the L2 cache buffering.

Apple 'Kansas' MotherBoard Note:
As reported here previously on my 1998 Mach 5/9600 page, the 'Kansas' motherboards (8600/250 and up, 9600/300 and up) seem to have lower memory performance than even previous designs such as the 8500 and 9500 (or even sub 300MHz 9600's which use a different motherboard). This has also affected graphics card performance to some extent.

Memory Bench Results:

As shown in my previous G4 reviews and articles, there is a dramatic difference in even the L1 cache speeds of a G4 upgrade in an older Mac versus a G4 upgrade in an Apple G3 system. L1 cache speeds of a G4 upgrade in a B&W G3 for instance show over twice the bandwidth of the same G4 CPU running in an older Mac. Thankfully this huge delta is not seen in actual applications use.

The images below show the results of tests in the 9600/350 with the PF G4/350 (with 1MB of backside L2 cache at 175MHz) compared to the stock 350MHz 604ex CPU (with 1MB of in-line [on card] L2 cache at 100MHz).

I'm told G4 CPUs prior to the 2.7 stepping normally run 4 wait states on the L1 cache. Regardless, L1 cache speeds in older Macs are much lower than seen in B&W G3s with the same CPU stepping running this particular benchmark. I can conclude that the slower main memory bus must be affecting L1 cache performance (it's the only logical conclusion). Again this dramatic difference does not show up in any real world applications tests.

Newer Tech's Gauge Pro Results:

Newer Tech's Gauge Pro utility also includes a memory bandwidth test feature. (Gauge Pro should ship with their new G4 cards but was not publicly available at their web site at the time of this review). Sample screenshots are shown below for the PF G4, stock 9600/350 and an XLR8 G4/350 upgrade for comparison.

.

Note G4 CPU Stepping (CPU Version) is 2.2

XLR8 G4 stepping was a later v2.6:

Stock 9600/350 results are about 8MB/sec lower than the G4 upgrades.

For more comparisons of memory bandwidth including an Apple G4/450 (aka 'Sawtooth' w/MPX bus) and upgraded B&W G3s, see my XLR8 and Newer Tech G4 ZIF upgrade reviews on the CPU upgrades page.

These 'pure' benchmarks are interesting, but not directly related to performance in most real world applications as shown on the next page of the review.

Summary: Since most benchmarks don't take advantage of the G4's 'Velocity Engine' (Altivec registers), much of the power of the G4 CPU is idle in these tests. However G4timedemo, even with the fill-rate limited Rage128 in the Kansas motherboard of the 9600/350 shows an example of the difference Altivec can make. Hopefully more benchmarks will be made Altivec aware in the future.
Benchmark performance is interesting for technical reasons, however I prefer to rely on real world application performance since that's what really counts. Application and 3D Game tests are covered in the next page of this review. Or you may use the links below to jump to a specific page.

Index of PowerForce G4 350MHz Review
Intro | Benchmarks | Appl. Tests | Software Controls | Installation | Specs/Design | Summary

- or -
Back to WWW.XLR8YOURMAC.COM

Copyright © Mike, 2000.
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.

The Source for Mac Performance News and Reviews
Review: PowerForce G4 350MHz G4 CPU Upgrade for PCI Macs By Mike Published: 1/23/2000
MacBench 5.0, G4TimeDemo and Memory Benchmark Test Results
Intro \| Benchmarks \| Appl. Tests \| Software Controls \| Installation \| Specs/Design \| Summary
Benchmark Tests
Benchmark tests were run with MacBench 5.0 as it is the accepted Mac standard, however the current version does not use the G4's Altivec extensions and therefore will not show the performance improvement they provide for software that uses them. I've also included results with two memory bandwidth benchmarks and with G4Timedemo, one of the only benchmarks I know that utilizes the Altivec ('Velocity Engine') of the G4. Remember the most important results are on the next page of the review which covers real world applications performance. This is where the rubber meets the road so to speak. Benchmarks have their place, but actual applications performance is far more important that pure benchmark scores. As you'll see in the G4timedemo test results (and the Photoshop 5.5/Soundjam MP results on the Applications tests page), until benchmarks like MacBench take advantage of Altivec, they are not an indicator of what the G4 is potentially capable of. However, MacBench CPU scores do seem to be a pretty accurate gauge of G4 performance in non-Altivec aware applications. Be aware that Macbench Graphics scores will vary depending on graphics mode and the installed video card. In Macbench 5.0 a 1000 score is the baseline based on performance with an Apple Beige G3/300 running millions colors, 1152x870, so consider this when evaluating any scores at lower resolutions and color depths. Likewise the disk scores will vary depending on drive and interface used (onboard or PCI SCSI, IDE, etc.), and even the free space and fragmentation present on the test disk. (The fastest areas of a disk drive are the outer tracks, so as a disk fills up it becomes slower, and data often becomes fragmented, further slowing down performance.) MacBench disk scores are highly affected by disk cache size set in the Memory Control Panel as well (improvements generally seen up to about 5-6MB cache size, where diminishing returns occur). In OS 8.5 and later, the disk cache defaults to a percentage of installed RAM. Earlier versions defaulted to very low settings such as 96K (increasing disk cache size was one of the tips in my Performance Basics article from the OS 7.6 era.) MacBench 5.0 Tests: The graph below compares the follow configurations and systems The base 9600/350 system (350MHz 604ex CPU w/1MB on-card L2 cache) The 9600 with the PF G4/350 upgrade running at rated (350MHz) CPU speed and at 400MHz (Overclocked - not recommended. Other samples may not be reliable at beyond the rated speeds.) The 9600 with past results with an XLR8 G4/350 upgrade installed. All tests used the same base system, OS 9, 320MB of RAM (interleaved), VM Off, ATI 4.2 drivers, OpenGL 1.1.2. All G4 upgrade tests had the 4 OS 9 Altivec extensions active. MacBench 5.0 Results Notes: Some explanation of the MacBench graph and test components. CPU/FPU Scores: MacBench CPU scores will disappoint most people that have heard so many marketing comments on the 'supercomputer' G4 chip. The G4's FPU scores are really no higher than the 604 at the same clock speed, as it has basically the same FPU design. (The G3 however does not have as powerful a FPU design as the 604 or G4, it's based on the 603e CPU, with a backside cache) Software: Since G4 control software often modifies the supervisor registers, it can have an effect on performance so later versions of control software may perform differently than the versions I used. The results above used PowerLogix G3/G4 Cache Profiler v1.3 and their Altivec extension v1.0. The XLR8 G4 tests used XLR8 Control Panel and Extension v1.4.3b0 (a later version exists now and may show different results). Speculative Processing was disabled to better reflect what the typical older Mac results would be (the base 9600/350 is one of the few Mac models with a ROM version that allows Speculative Processing to be enabled without any negative effects). Disk Scores: The 9600/350 test system was running OS 9 from a Ultra2 ATTO PCI SCSI card driving a Seagate Cheetah U2 4.5 GB hard drive that was approximately 80% full and not optimized. If the disk had more free space or was defragmented (optimized) scores would have been higher. Due to installing control software for different cards, the disk scores vary a bit from past G4 upgrade tests shown in the graph as the disk cannot be in the exact same state (free space, etc.) as prior tests. Graphics Scores: The 9600/350 used a OEM (rev 1) Rage128 graphics card taken from a early B&W G3. This card has a slower clock speed (about 10%) than the retail Rage128 cards or later rev 2 B&W G3 systems, so the graphics scores are a bit lower than they would be with a retail Rage128 card. ATI's 4.2 universal installer drivers along with the final release of OpenGL 1.1.2 were used. See my video cards page for reviews of faster alternative cards for these systems. G4Timedemo Benchmark: G4timedemo is a benchmark for showing the benefit of Altivec support with G4 CPUs. I ran tests with each of the system/upgrade combos and graphed the results. As you can see, a G4 CPU w/Altivec support does make a huge difference here (60% gain with a G4 CPU of the same speed). All tests were run with the desktop video mode of 1024x768, thousands colors and G4timedemo set to 'Amazing' (max) quality. G4timedemo results seem to vary a few FPS from run to run but you can clearly see the dramatic difference a G4 w/Altivec makes in performance. I repeated the tests several times (each after a clean reboot) and noted the best score of 3 runs. Since OpenGL 1.1.2 contains some Altivec optimizations, games will usually some some benefit of a G4 CPU as well (if the video card is not already saturated and a bottleneck of course) . Memory Bandwidth Tests: I want to caution readers to not be too concerned with the results below. Although the 9600's Kansas motherboard shows rates of about 1/3 that of a B&W G3 or later Mac, it's often very close to those systems in real world applications performance if running the same speed and type of CPU. The efficiency of the L1 and L2 caches in CPUs are the reason why a 50MHz system bus speed older Mac with the same type and speed of CPU upgrade generally keeps pace with modern Macs that have 100MHz bus speeds in most applications tests. (Applications that move a lot of data over the bus are exceptions - Photoshop, 3D Games, etc.). Due to workload and the fact this review was overdue, I don't show the results of Stream tests that I have used in previous G4 reviews. Copy-Back (often called Write-Back) cache mode was used for this test, but the slower Write-Through mode may be required for reliability in some Mac models (in 9500's for instance, based on some troubleshooting reports). Write-Through cache mode would show much lower scores for write operations since it basically negates the effect of the L2 cache buffering. Apple 'Kansas' MotherBoard Note: As reported here previously on my 1998 Mach 5/9600 page, the 'Kansas' motherboards (8600/250 and up, 9600/300 and up) seem to have lower memory performance than even previous designs such as the 8500 and 9500 (or even sub 300MHz 9600's which use a different motherboard). This has also affected graphics card performance to some extent. Memory Bench Results: As shown in my previous G4 reviews and articles, there is a dramatic difference in even the L1 cache speeds of a G4 upgrade in an older Mac versus a G4 upgrade in an Apple G3 system. L1 cache speeds of a G4 upgrade in a B&W G3 for instance show over twice the bandwidth of the same G4 CPU running in an older Mac. Thankfully this huge delta is not seen in actual applications use. The images below show the results of tests in the 9600/350 with the PF G4/350 (with 1MB of backside L2 cache at 175MHz) compared to the stock 350MHz 604ex CPU (with 1MB of in-line [on card] L2 cache at 100MHz). I'm told G4 CPUs prior to the 2.7 stepping normally run 4 wait states on the L1 cache. Regardless, L1 cache speeds in older Macs are much lower than seen in B&W G3s with the same CPU stepping running this particular benchmark. I can conclude that the slower main memory bus must be affecting L1 cache performance (it's the only logical conclusion). Again this dramatic difference does not show up in any real world applications tests. Newer Tech's Gauge Pro Results: Newer Tech's Gauge Pro utility also includes a memory bandwidth test feature. (Gauge Pro should ship with their new G4 cards but was not publicly available at their web site at the time of this review). Sample screenshots are shown below for the PF G4, stock 9600/350 and an XLR8 G4/350 upgrade for comparison. . Note G4 CPU Stepping (CPU Version) is 2.2 XLR8 G4 stepping was a later v2.6: Stock 9600/350 results are about 8MB/sec lower than the G4 upgrades. For more comparisons of memory bandwidth including an Apple G4/450 (aka 'Sawtooth' w/MPX bus) and upgraded B&W G3s, see my XLR8 and Newer Tech G4 ZIF upgrade reviews on the CPU upgrades page. These 'pure' benchmarks are interesting, but not directly related to performance in most real world applications as shown on the next page of the review.
Summary: Since most benchmarks don't take advantage of the G4's 'Velocity Engine' (Altivec registers), much of the power of the G4 CPU is idle in these tests. However G4timedemo, even with the fill-rate limited Rage128 in the Kansas motherboard of the 9600/350 shows an example of the difference Altivec can make. Hopefully more benchmarks will be made Altivec aware in the future. Benchmark performance is interesting for technical reasons, however I prefer to rely on real world application performance since that's what really counts. Application and 3D Game tests are covered in the next page of this review. Or you may use the links below to jump to a specific page.

Index of PowerForce G4 350MHz Review Intro \| Benchmarks \| Appl. Tests \| Software Controls \| Installation \| Specs/Design \| Summary - or - Back to WWW.XLR8YOURMAC.COM
Copyright © Mike, 2000. 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.