Click for Thunderbolt upgrades!
Rage128 vs 3Dfx Banshee Video Card Comparison
Review date: 7/13/99 (Updated 7/18/99 for Q2 multitexture lib scores)
After over a year of begging Apple to license OpenGL from Conix, my prayers were answered at the S.F. Macworld keynote speech earlier this year. With the addition of OpenGL as a standard component of the MacOS, the game gates have opened and we're seeing titles being released and developed for the Mac that we could only dream of a year ago. Quake 2 finally arrived, Quake3test was released first for the Mac and even such mega-hits (game of the year) as Half-Life are being ported to the MacOS. I don't think this would have happened without Apple making OpenGL a standard part of the OS. Although performance may lag many PCs (in large part due to the lack of PCI bus 'write combining'), with the right Mac configuration (plenty of RAM, a fast CPU and good 3D graphics card) game performance is more than adequate and will only get better as OpenGL, drivers and game code are optimized. Mac gamers have a lot to be thankful for and the best is yet to come.
I also want to publicly say Thanks and Great Job to Miklos Fazekas for his work on 3Dfx Mesa GL for the Mac. Without it 3Dfx card owners on the Mac would not be able to run games like Quake2 and Q3test. They've done an amazing job and continue to provide updates often to improve performance, add features and address issues. I know of no company that is as prolific. I think VillageTronic or 3Dfx should hire this guy ASAP.
To compare game performance I used some of the most popular past and present titles: Quake 1 (RAVE version for the Rage128, 3Dfx/Glide version for the Game Rocket since it has no RAVE support), Quake 2 (OpenGL, using Mesa3DfxEngine for the GameRocket), Quake3Test (OpenGL) and Unreal (RAVE and Glide). Due to disk space limitations, Falcon 4 performance is only shown with the B&W G3. I've included both benchmarks and image quality ratings from the two cards. Screenshots in OpenGL games were a problem on the Game Rocket as the files had a totally blue cast that I could not correct. This was not the fault of the Mesa3dfxEngine extension, as Voodoo2 card screenshots did not have the same problem.
Since many readers (and I) may be wondering, I've also included comparisons to a 3Dfx Voodoo2 Game Wizard and a RagePro card in many of the tests. Note: Any time you see 'Voodoo2.ini' in the table of results that means the Voodoo2 card was using my tweaked (but not overclocked) ini file - see my Voodoo2 Tweaking page for more info or to download the file.
Note the OpenGL performance of the Game Rocket is disappointing and the author of Mesa3DfxEngine says he is working with VillageTronic to improve this. The lack of multitexturing in a single pass (dual texture engines - present in both the Rage128 and Voodoo2) is not the sole reason, as Quake 2 does not support this feature and the Game Rocket trailed the Rage128/Voodoo2 in that game as well.
For information on how to test framerates in games like Quake1, Quake2, Q3test and Falcon, see the Game section of my Frequently Asked Questions.
To avoid people thinking that my overclocked 9600/350 has a G3 upgrade, I've listed it in the table as a an 'OC 9600/350', since I modified the Apple 604E CPU card to run at 400MHz.
Quake 1 is getting long in the tooth now but still popular since it needs less horsepower than many of the new games like Unreal, Quake2, etc. I ran tests in all 3 systems at resolutions up to 1024x768. The Voodoo2 and earlier 3Dfx cards can't run 1024x768 mode as currently Mac Glide is limited to 960x720. I'm hoping 3Dfx's soon to be released (I hear they are complete) Voodoo2 mac glide drivers do have SLI mode support, as this would allow two Voodoo2 cards to work in parallel for 1024x768 resolution and higher performance in some cases. (See my SLI page for more info on how it varies by CPU speed and Game, based on PC tests.)
For RAVE tests, Shadows, Flames and Filtering were enabled. All tests used the Quake v1.09 update.
Performance: The following tables are sorted by fastest to slowest card/system combo.
There are so many data sets/cards/systems that tables seemed clearer than a graph which was very cluttered.
800x600 Results: As the resolution increases things get interesting as some players change positions performance wise.
1024x768 Results: At 1024x768 the 3Dfx cards can't compete - as Mac Glide is limited to 960x720. Without a RAVE driver they can't run Quake at this resolution. (I didn't have Microconversions' Rave Voodoo2 driver on hand to test and I'm not sure the Voodoo2's frame buffer allows 1024x768 RAVE. At press time there were no Rave drivers available for the Banshee based Game Rocket.)
Note: That is not a typo - I re-ran the PowerCenter Pro 210/Rage 128 tests twice to verify it. I can only guess that perhaps OS 8.1 and far less extensions than the OS 8.6 equipped B&W G3 (which has tons of USB and Firewire extensions) is the reason why a 210MHz 604E with 1MB cache and 60MHz ran as well as the same card in a 33MHz PCI slot of the B&W with a 400MHz G3 with 1MB of backside cache and a 100MHz at RAVE Quake at 1024x768. Note that at lower resolution tests of RAVE Quake the G3/400 was substantially faster. Also note it ran a hair faster than the Kansas motherboard (50MHz bus) 604E at 400Mhz with a 100MHz 1MB inline cache, running OS 8.1. I'd assume from these results the card or RAVE itself is maxed out at that framerate regardless of CPU speed. Note that OpenGL games didn't show this trait, so perhaps the limit is in RAVE.
Image Quality: Quake Rave has come a long way, appearance was almost identical to the 3Dfx version in my opinion. It would be hard for the average user to spot the difference but I still think the 3Dfx version looks a bit better in some respects (impossible to show in screenshots).
The Game Rocket looked great in Quake 3Dfx but there were some areas where I saw clipping issues I did not see with the Voodoo2 or any other 3Dfx card I've reviewed. In some cases it was not repeatable but below is an example of one area where it was 100% repeatable with every machine I tested. Simply running demo1 and 2 would often occasionally show 'leaks' (tranparent areas which show as aqua/blueish usually) on the top surface/edges of water areas for instance. The screenshots below show one area where I first saw it during gameplay; moving down and back in that hallway would make the problem area grow and shrink on the side of the wall. Image problems like this were rare however and may be addressed in a Glide extension update or perhaps it was due to some specific software extension I was using, but since the Voodoo2 card did not show the same problem I had to assume it was Game Rocket specific.
I must say that unlike the Voodoo1/Voodoo2 cards, the screenshots from the Game Rocket don't look as good as the actual game. In case you're wondering the image quality of the full size screenshots are not due to JPEG compression, as I used very little as you can tell from the 130-160K file sizes. Ignore the poor quality of the screenshots as image quality was excellent in every game I tested on the Game Rocket. I'm wonder if the thousands color mode Banshee issue is why these screenshots are don't look as good as the actual game screens. Warning - the full sized screenshots are very large files so they may take some time to download.
Game Rocket Quake 3Dfx Screenshots
RAVE Quake Screens: The screenshots key (F12), even when remapped does not work in Rave Quake. I thought I had a workaround last year but can't remember it. I'll try to download SnapZ Pro and grab a few shots to post here. In general Rave Quake looked comparable to the 3Dfx version, and again most users would not be able to tell the difference. If you're running both a 3Dfx and RAVE card, sometimes it's best to delete the files from the GL_Quake folder to force a rebuild when switching modes. This improved RAVE texture/image quality on some of my macs here that have mixed cards.
I installed Quake 2 (full install) and ran timedemo tests all 3 systems with each of the video cards. Tests in the B&W G3 were run with the Rage128 and Game Rocket in both the 66MHz and 33Mhz PCI slots. All tests used the same game settings
Update: I retested the Voodoo2 card with the new MesaQuake2 and gl_ref.lib updates for Quake 2 that adds multitexturing support for Voodoo2 cards (and Voodoo3 when it becomes available). Currently there is no Rage128 multitexturing support in Quake2, although Rage128 the chip and Apple's OpenGL does support it. For links to down the files see my Game News page sidebar software updates section.
The table below shows the of Quake 2 tests with the original release (except where noted on tests with the Voodoo2 which used the new Q2 libs):
Note: That's no mistake - the Game Rocket moved up a notch at 1024x768 mode, however its overall OpenGL performance was disappointing (even the author of Mesa3DfxEngine is trying to figure out why). With the release of multi-texturing libs/support for Quake 2, the Voodoo2 and Rage128 chips that support this feature (single pass multitexturing -not supported in the Banshee chip) should see a 10% or so speed boost. I'll update these tables when the official patch is out.
All tests above were done with the default Quake 2 shadows off. The command to enable shadows is 'gl_shadows 1' (gl_shadows 0 turns them off). You can add 'set gl_shadows "1"' to your config.cfg file to make it a default option. As noted at the Game News page last week, I ran tests with and without shadows enabled in Quake 2 with a Rage128 Orion, Game Rocket and Voodoo2 card to see what performance effect shadows had when enabled. To save space on this page I did some tests to see what performance effect shadows had with the Voodoo2, Game Rocket (Banshee) and Rage128. The table below shows the results. The Voodoo2 scores used my Voodoo2.ini file shown on the Voodoo2 tweaking page. The 7/18/99 update adds the results of tests with the Voodoo2 and the new Quake2 multitexture support update.
All 3dfx cards (Voodoo2 and Banshee) used the B3 Mesa3dfxEngine. Both Rage128 and Game Rocket were in the 66MHz PCI slot of a B&W G3. When the multitexturing support for Quake2 arrives the Rage128 and Voodoo2 scores should improve.
Image Quality: I was impressed with the image quality of both the Rage128 and Game Rocket cards. I needed to adjust the gamma settings in the game for all the 3Dfx cards as they were very dark. As with all OpenGL games I tried, screenshots from the Game Rocket were totally blue-cast (see below) so I can really only supply Rage128 images here. Nothing I tried resolved the Game Rocket/OpenGL screenshot problem (VillageTronic and IXMicro drivers made no difference, nor did 3 different revisions of the Mesa3DfxEngine extension). Image quality in the game was good however.
For Quake2 articles from my game editors (Ruffin Bailey and Randall Markarian) and I, see the sidebar on the 3D Game News page.
One thing I noticed on the 9600/350 when using the Voodoo2 card was that when playing the timedemos, the audio seemed delayed from the action. When I changed the Monitors and Sound audio rate to 44KHz from 22KHz it seem to help quite a bit. Quake2 was set for low quality audio but I reran the timedemo tests and noticed an improvement. I also disabled CD audio (and no CD was in the drive).
Quake2 Rage128 Screenshots:
Since the Game Rocket screenshots in Quake 2 (and Q3test) come out with a severe blue tint (I even tried the RGB swap setting - no help), I've only included Rage128 pictures for now. Clicking on the thumbnails will load a 800x600 full size version (about 70KB JPEG files).
I used Q3Test 1.05 (latest currently available) for all tests on each system and card combination. If the card supported it I ran tests up to 1024x768 resolution. Graphics settings in the game were :
I had a lot of trouble getting Q3test to run timedemos on the Game Rocket. I'd say I spent over a day alone on just trying to get Q3test to run without 'server timeouts' and incomplete runs. This happened occasionally on other cards but seemed worse on the 3Dfx cards than the Rage128. I also noticed a lot of variation in the GameRocket scores, especially on the PowerCenter Pro 210. Once it would report 6fps at 1024x768, then 9 fps. As noted on my Quake3 Tips page, according to ID the timedemo in Q3test is 'broken' but has been used by Thresh and others as a way to compare video card performance. Consider these numbers just a gauge, not a real frames-per-second and they are for reference only. I can say that the 9fps reported with the Game Rocket/PCP210 was probably close to being accurate, as it felt very sluggish at that resolution in that system.
The tables below show Q3Test performance with each of the card/system combos:
(I inadvertently omitted 1024x768 tests in some systems, and will try to fill those in.)
Image Quality: As with Quake2, screenshots of the GameRocket had a totally blue cast, but in-game images were good. The Rage128 also looked great in my opinion, except for the sky texture (which didn't look much better on the Game Rocket). Chris Bentley (3D programmer) of ATI tells me this is due to the format of the sky texture in Quake3 and will be addressed in a later release. Since the Game Rocket screenshots were literally all blue, below are screenshots from the Rage128 only:
Quake3 Screenshots (Rage128) Here are a few screens I grabbed from my B&W G3 (click image to see full size versions):
Example of Blue Screenshots with GameRocket in OpenGL games
The pausing in some areas of Unreal I noted on my original B&W G3 Game Performance page appears to have been addressed as I did not see it during the tests. The flashing walls and flickering crossmember in the opening 'castle flyby' scene in Unreal seen on the Rage128 is due to the game code according to ATI. They say incorrect 'hither and yon' (?) settings in the game are the reason. ATI said if this was a driver issue they would have addressed it already. I've written Mark Adams of Westlake for his response on the matter and here are his comments:
"Flashing walls in the fly-through is a ATI driver bug, that isn't fixed in any version I know of. The crossbar flickering of stuff in the distance is a problem with the way Unreal sends Z coords to the driver, and we haven't been able to resolve it yet. I am trying to fix it for Unreal Tournament and if I do it will get fixed in Unreal..
I used high quality detail settings for all Unreal tests with resolutions up to 1024x768 for cards that supported it (3Dfx cards running Glide can't as Macglide is limited to 960x720). I used ver 219 of Unreal with the 1.02B4 patch. I didn't use the latest Unreal v224 due to bugs reported among other things but primarily due to the fact I had used that the previous version in other graphics card tests.
Since Unreal uses Glide for 3Dfx cards they can't run 1024x768 mode. The Rage128 was run in 16Bit and 32Bit color modes. As you can see there was little difference in the Rage128 scores with either color depth.
Image Quality: The previously noted castle flyby oddities and some dithering in fog and flares (even in 32bit mode) seen with the Rage128 are primary reasons why I think 3Dfx cards look better in Unreal. Since multitexturing support was added Fog looks similar on the Voodoo2 and Rage128, and there are some areas/textures that if you look really close you'll see some benefit to 32bit mode on the Rage128, but most people will never notice it during gameplay. The dithered look when exiting the fog and the dithered look to the flare's glow on the castle flyby is the most noticeable difference in Rage128 vs 3Dfx card visuals.
Since the Voodoo1 and Banshee chips do not support multitexturing (they only have a single Texture Management Unit), fog actually appears more pleasing with them in my opinion (less flashing so to speak that is seen with the Voodoo2/Rage128). Although it can't run has high a resolution and during normal gameplay the cards look very similar, all things considered I'd give the Game Rocket the edge in visual quality in this game.
Game Rocket Unreal 3Dfx Screenshots
Rage128 Unreal Rave Screenshots
I ran tests of Falcon 4 with the latest 1.06c patch only on the B&W G3/400. This was primarily due to disk space limits on the 9600 and PowerCenter Pro (less than 100MB left) and the fact that game really needs a very fast CPU for best play. The game features a built-in framerate (instantaneous, not average) counter and I noted what was an average framerate seen during an autopilot run of the 'instant action' mode. I only wish the text size was larger - the tiny text is very hard to see on many monitors and against some scenery. I want to make clear that this is just what I saw in one instant action autopilot run, not a definitive test. The Voodoo2 card did not have my Voodoo2.ini tweaked settings but later tests showed they really have no effect it seems in Falcon 4.
Image Quality: Both the Voodoo2 and Banshee cards have more realistic texturing in the game in my opinion, most noticeable on the plane's surface. The images below demonstrate the most clear examples I've seen of Rage128 vs 3Dfx image quality. The Rage128 looked just as good in my opinion on smoke and explosions as the 3Dfx cards, and overall the image quality was better than I expected actually based on initial reader reports. This game is very resource intensive, and as shown in my Rate Your Mac Games database, for best results you'll want a fast video card, G3 CPU and plenty of RAM (128MB or more).
I was impressed that the smoke and explosions looked as good on the Rage128 to my eyes as it did on the 3dfx cards, but the texture mapping on the planes seemed more realistic in Glide mode with the 3Dfx cards.
Game Performance Summary:
I saw no texture problems at all with the ATI cards other than the noted Q3Test sky which is said to be fixed by ID before the final release. I did see the Unreal dithering/banding however and the Game Rocket had the better game image quality. Some B&W G3 owners have reported filtering issues in the forums with the OEM Rage128 in some games, but overall I was impressed with the quality of the visuals in the OpenGL games I tested with the Rage128. In Glide/3DFx vs RAVE games, the 3Dfx cards just look better in most cases, always have really. However in almost every game (except Quake 1), the Rage128 was faster than the Game Rocket, sometimes by a large amount.
The Game Rocket does best in games with native Glide/3Dfx support, but Mac Glide is currently limited to 960x720 resolution, so it can't run as high a resolution as the Rage128 can in Rave mode of equivalent games. Image quality in games was very good with the Game Rocket, except for the occasional oddities I saw in Quake 1 3dfx (often not repeatable). Most disappointing is the OpenGL performance of the Game Rocket, and the fact there is no RAVE drivers, so RAVE only games like Bugdom, Nanasaur, etc. can't be run. When VillageTronic/IXMicro releases RAVE drivers it will be interesting to see how the performance compares to the Rage128. For Voodoo2 card owners, I see no reason to get a Game Rocket for games, as in most every case I saw the Voodoo2 was faster, esp. when run with my Voodoo2.ini file (see my Voodoo2 Tweaking page for details and to download the ini file). The Voodoo2's multitexturing support is also an advantage and missing from the Banshee chip on the Game Rocket.
For the target market of these cards, game performance is probably important. Overall the ATI Rage128 is a much better choice for OpenGL games of these two cards, at least with the drivers I had for testing. Its the only choice of the two for RAVE games. The recent reports of lockups in some Mac models give me pause however, despite the fact I didn't see in tests here (but I didn't run games for extended periods of time - no more than an hour or so).
For 3Dfx games the Game Rocket would be a good choice for someone that did not have a 3Dfx card and needed faster 2D in a single card. Check any game you wish to run; if it only supports RAVE, you won't be able to run it on the Game Rocket until RAVE drivers are released. Most new games (3D first person at least) require OpenGL support for best performance. The lack of multitexturing support and the significantly lower OpenGL performance of the Game Rocket compared to the Voodoo2 make it a less desirable 3Dfx card; so if you have a Voodoo2 card already, there's no reason to buy a Game Rocket for games in my opinion.
Based on my tests with several games, I rated game performance of the two cards as follows:
The next tests I ran dealt with movie playback.
- or -
Copyright ©, 1999.
No part of this sites content or images are to be reproduced or distributed in any form without written permission. All brand or product names mentioned here are properties of their respective companies.