 The Source for Mac Performance News and Reviews |
|
 Review: PowerForce G4 350MHz G4 CPU Upgrade for PCI Macs By Mike Published: 1/23/2000 |
|
|
|
|
|
|
|
| Applications and Game Performance | |
| This page lists test results in common Mac applications like Photoshop 5.5, SoundJam MP, Infini-D 4.01, After Effects 3.1, Premiere 4.2.1, Virtual PC 3.0 and popular 3D games like Quake3 Arena and Unreal. Only Soundjam MP and Photoshop 5.5 (with the Altivec plug-ins) take advantage of the G4's 'Velocity Engine'. OpenGL 1.1.2 does have some Altivec optimizations however, so games usually get boot from a G4 cpu if the video card is not the bottleneck. (At higher resolutions at least, a Rage128 card is often fill-rate limited for 3D games and therefore may not show any benefit.)
As always, a clean reboot was performed before running each test. All but the Premiere movie tests used the application's built-in timing function to eliminate human error variations present from starting and stopping a stopwatch. For comparisons in many of these applications to more Macs (604 and G3 CPUs of many types) see my Real World Apps Performance page. The CPU Upgrades page lists reviews of many other upgrades (G4/G3/604e) in other Mac models. |
|
| SoundJam MP 1.6 MP3 Encoding Tests | |
| Casaday & Green's SoundJam MP 1.6 (full version) has Altivec support and was used to test the boost provided by the G4 CPU versus the same speed Apple 604ex CPU in the 9600/350 system.
As you can see, the 'Velocity Engine' support of the G4 delivers a huge performance gain compared to the stock 604ex CPU of the same speed. (Due to having found the solution to problems running Soundjam MP 1.6 with the G4 only just as the review was being written, tests with the PF G4 at 400MHz and other G4 cards are not shown in this review.) Since past G4 reviews used an earlier demo version of SoundJam MP (v1.1), their results are not shown here since they may not be comparable. |
|
| Photoshop 5.5 Tests: | |
| I also ran a complete PS5Bench (21 filter test) series using Photoshop v5.5 with the Altivec extensions (active for G4 CPUs only). PSBench settings are 1024x768, millions colors, VM off, Interpolation set to bicubic (better) and Photoshop should be allocated enough RAM to avoid any swap file activity from the 10MB test image filter actions. (I allocated 140MB to Photoshop 5.5 for this review.)
Unlike my early PSBench tests which used v5.02, to eliminate swap file/disk activity I saw even with a 10MB image file and 140MB of RAM allocated to Photoshop v5.5, I set the 'History' settings from the default '20' to '1' and unchecked the 'automatically create snapshot' option. This dramatically lowered the filter times and removed all signs of disk activity during the filter tests (each filter is run 3 times during the test). Photoshop 5.5 showed much higher PSBench times without these changes than what I saw with Photoshop 5.02 from earlier tests. PS v5.5 seemed to hit the scratch disk much more quickly than I'd seen with previous versions, at least with the default history settings. Altivec Extensions Note: All G4 CPU tests had the 4 OS 9 Altivec extensions active as well as the Adobe current release of the Altivec Core and Lighting Effects Filter. The updated/public release fixed a distortion filters bug and promises increased performance on some functions over the original version that shipped with Apple's G4 systems. The most improved filter performance I saw in the new release was Lighting Effects (about 13% faster than the original version). I've included results with an AMD Athlon 600MHz CPU based system running Windows 98. Unlike the Pentium III and G4, there is no support in Photoshop 5.5 for AMD's Altivec/PIII-like enhanced instruction set (called 3DNow!), but it still does well in many filter tests using brute force alone (triple superscalar FPU engine, highly pipelined design and 600MHz clock speed). The Athlon's SIMD registers are only 64 bits wide however, unlike the G4 and Pentium III which are 128 bit. The "Benefit" column in the table below notes how many times faster the G4 CPU upgrade was that the original 9600/350 system in these filter tests. The most dramatic gains were seen in the Lighting Effects filter which was over 4.5 times faster with the G4 CPU than a 604e CPU of the same speed. I also noted that disabling the Altivec extensions with the G4 CPU installed increased the Lighting Effects filter times by about a factor of 4. This shows the performance potential of Altivec instructions. |
Filter
Benefit
(G4/350
vs Stock
9600/350)PF G4/350
(9600/350) 350/175/50PF G4/350
(9600/350) 400/200/50XLR8 G4/350
(9600/350)9600/350
(Stock)Athlon 600MHz
Win98
256MB RAM
Rotate 90° CW
3.5x
0.6
0.6
0.6
2.1
0.3
Rotate 9° CW
1.28x
3.6
3.3
3.8
4.6
2.2
Rotate .9° CW
1.27x
3.3
3
3.5
4.2
2.1
1 pix Gaus. Blur
2x
1.2
1.2
1.3
2.4
0.6
3.7 pix Gaus. Blur
1.81x
3.1
2.9
3.3
5.6
2.2
85 pix Gaus. Blur
2.17x
3.5
3.4
3.7
7.6
4.2
Unsharp Mask
50%/1pix/0 level1.88x
1.6
1.5
1.7
3
0.7
Unsharp Mask
50%/3.7pix/0 level1.75x
3.6
3.3
3.7
6.3
2.3
Unsharp Mask 50%/10pix/5 level
1.86x
3.5
3.4
3.6
6.5
2.6
Despeckle
2.47x
1.5
1.3
1.5
3.7
2.0
RGB-CYMK
1.19x
5.9
5.1
6
7
4.2
Reduce 60%
2.38x
0.8
0.8
0.9
1.9
1.1
Lens Flare
1.33x
5.7
5.3
5.8
7.6
3.3
Color Halftone
1.59x
4.6
4.1
5.3
7.3
5.0
NTSC Colors
1.34x
5.0
4.4
5.2
6.7
4.7
Accented Edges
1.21x
12.8
11.4
13.4
15.5
12.5
Pointillize
1.19x
18.6
16.6
19.1
22.2
13.8
Water Colors
1.21x
26.8
23.7
27.8
32.3
25.7
Polar Coordinates
1.84x
5.0
4.6
5.2
9.2
7.2
Radial Blur
1.14x
42.4
37.4
42.9
48.3
35.8
Lighting Effects
4.59x
2.7
2.4
3
12.4
6.2
Total Time
(1.39x)
155.8
139.7
161.3
216.4
138.7
System
Benefit
(G4/350
vs Stock
9600/350)PF G4/350
(9600/350) 350/175/50PF G4/350
(9600/350) 400/200/50XLR8 G4/350
(9600/350)9600/350
(Stock)Athlon 600MHz
Win98
256MB RAM
| ** XLR8 G4/350 upgrade tests used v1.43b0 of their cache control and extension. Later revisions may show different results.
The Altivec extensions showed dramatic gains in some filters (blurs, scaling and lighting effects). When looking at the results above, consider which filters you use most often when determining how much a G4/altivec upgrade will benefit you. The overall time is of less importance if the series above includes many filters you rarely use. Weigh the gains from those that you use most often. To download the PSBench5 action, visit the PSBench home page (make sure you get the v5, not v4 action). Note you cannot compare results with Photoshop 5.0x to 5.5. I'm also baffled at the low scores some reported with v5, unless they had reduced the history settings from the defaults.
|
|
| Adobe Premiere 4.2.1 Tests | |
| The following is a chart showing the time to produce a Quicktime movie from the "Sample Project" file (duration set to the full length of the project file). Output file settings were: Video codec (max quality), 320x240, 15 FPS, keyframe every 5 frames, 22KHz/16-Bit stereo audio.
Note: Unlike the other applications tests here that have their own timing function (eliminating human error), this test required a stopwatch to record times. There could be a 1/2 second or so reaction time variation for starting and stopping the stopwatch between runs.
|
|
| Infini-D 4.01 Tests | |
|
Infini-D 4.01 does not use Altivec extensions, but is a common application I have used for comparisons of CPU/FPU performance. I used the same 'Chapter 7 completed' tutorial scene file from my past reviews. Rendering quality was set to Ray Trace, medium anti-aliasing, shadows on, patch detail low. I didn't change the default QT movie output file options.
The graph below shows times to complete the 150 frame movie (10.6MB) rendering with the stock 9600/350, the same system with the PowerForce G4/350 upgrade (at both 350 and 400MHz settings) and an XLR8 G4/350 (not tested at 400MHz). Infini-D was allocated 40MB of RAM for the tests. OS 9 and its 4 Altivec extensions were used for all G4 tests.
As noted in my previous G4 reviews, in this test a G4 CPU is usually a few seconds slower than a G3 of the same speed (proven in repeated testing). If Infini-D 4 used Altivec I'm sure the results would have been a far different story. |
|
| Lightwave 3D 5.6D Tests | |
|
Lightwave 3D 5.6D (OpenGL) was used to test rendering performance with the Raytrace benchmark file. The graph below compares the results with the stock 9600/350 CPU (350MHz 604ex with 1MB of L2 cache) and the same system with G4 CPU upgrades.
|
|
| After Effects 3.1 Tests | |
| After Effects v3.1 does not take advantage of the G4's Altivec registers. The following is a chart showing the time to render a special effects movie (appx. 8MB file size) in After Effects 3.1. Resolution was set to 1024x768, thousands colors as was common on all but the Photoshop 5.5 tests.
Note: A stock 8600/250 took 5 minutes, 54 seconds to complete this test. A 8500/120 with 1 MB L2 cache took 12 minutes, 18 seconds. As shown in my MAChCarrier G3/500 review, under OS 8.6 a Genesis system with that upgrade running 500/250/50 completed this same test in under 3 minutes. Although After Effects v4.1 now has an Altivec plug-in available at Adobe.com, several readers reported it provided literally no gain in tests with their particular projects. As with Photoshop, I suspect only certain functions are accelerated and the benefit seen will vary depending on effects used.
|
|
| Virtual PC 3.0 Tests | |
| The following chart shows the results of tests with Connectix's Virtual PC 3.0 (VPC) using Norton Utilities v4.0's System Info benchmarks. Virtual PC was allocated 128MB of RAM.
Previous G4 results were from my 1MB vs 2MB G4/350 upgrade comparison. The XLR8 G4/350 card ran VPC 3.0 fine with the U2 SCSI card in PCI Slot 4. Perhaps moving the PCI card to the upper slot (PCI bus shared with the video card) and the reinstallation of OS 9, VPC 3.0 and Norton Utilities must be responsible for the difference in SI scores between the PL and XLR8 G4/350 scores. VPC 3.0 would not run with the PF G4/350 card until the SCSI card was moved to the upper slots. Prior to that during troubleshooting I had reinstalled (clean) OS 9 and Virtual PC 3.0 (which required reinstalling Norton Utilities in Windows 98 of course). The fact the video card and PCI SCSI controller share the upper PCI bus resulted in about a 30% lower disk score in Virtual PC.
|
|
| Game Performance Tests: | |
Unreal Performance: I used Unreal v224b7, the latest version available for this test. All systems were running OS 9 and QuickDraw 3D v1.6. The G4 CPUs had OS 9's Altivec extensions active. To prevent the OEM Rage128 card from being a bottleneck, I installed two 3dfx Voodoo2 cards running in SLI (scan line interleaving mode) and Unreal was set to use Glide (not Rave). The table below summarizes results from the 3rd cycle Timedemo of the 'Castle flyby' scene, 640x480 resolution. (All tests used the same high quality textures/detail settings - see my FPS Database entry page to download Unreal.ini files for RAVE and 3dfx cards.)
To see how other video cards like the Voodoo3, Rage128, and Formac Proformance3 perform in Unreal with various CPUs, see my Mac 3D Card Roundup and my past reviews of those cards. For a wide array of video card/system/game performance results, you can search my Mac Game/Video Card Framerate database. For gamers, the under $100 (list price) Voodoo3 2000 is a very cost-effective way to improve performance, but remember that adequate RAM (128MB recommended) is also needed with the latest crop of OpenGL games for best performance.
Quake3 Arena (Retail) Tests: Game settings were left at the defaults, 16-bit, Lightmap lighting, high geometry detail, etc. and the default game options page settings were used (high quality sky, ejecting brass, dynamic lighting, etc.). Due to the fact this review is late, I did not test the PF G4 at 400MHz speeds or run other G4 upgrades for comparison.
The effect of the (limited) Altivec support in OpenGL 1.1.2 is seen from these results, but it's also apparent the CPU is not saturating the Voodoo3 graphics card fully. Faster CPUs and more modern Macs deliver faster Quake 3 performance with the Voodooo3. (See my Mac Game/Video Card Performance searchable database for examples of this). |
|
|
Summary: Unlike comparing the G4 to a G3 CPU of the same speed, the G4 does show consistent performance gains even for software that does not support the G4's Altivec (aka 'Velocity Engine') compared to a 604e CPU of the same speed. The faster backside L2 cache is likely the major reason for the improved performance over a 604 CPU of the same clock speed. (BTW: The Apple 604 250, 300 and 350MHz CPU cards have a 1MB inline cache on the CPU card running at 100MHz, unlike most other 60x Macs that have motherboard L2 cache that runs at the main system bus speed. It's important for owners to be aware that these Apple CPU cards are not compatible with Macintosh models other than the 8600/250, 8600/300, 9600/300 and 9600/350). I highly suggest you keep the original Apple 604 CPU card as a spare, since they are not worth much on the open market due to the limited system compatibility and the fact they may be needed should the G4 card fail or you wish to sell the system in a stock configuration in the future. For those considering a G3 vs G4 upgrade, remember for non-Altivec enhanced applications, don't expect any real gain from a G4 compared to a G3 of the same speed however. (I.E. If you have a fast G3 CPU already, you may want to wait for faster G4s or until more applications are Altivec enhanced.) For those applications that do support Altivec, significant gains are possible as shown in the results above. Hopefully more programs will be updated to take advantage of Altivec extensions in the near future as much of its potential is untapped today. For comparisons of G3 vs G4 CPU upgrades, see my previous G4 reviews and articles on the CPU Upgrades page. In the G4 ZIF upgrade reviews, I also include comparisons to a new Apple G4/450 AGP system.
|
|
| The next page describes the software cache control supplied with the card. Or you may use the links below to jump to a specific page. | |
|
|
|
|
Intro | Benchmarks | Appl. Tests | Software Controls | Installation | Specs/Design | Summary - or - |
|
|
Copyright © Mike, 2000. Users of the web site must read and are bound by the terms and conditions of use. |
