Very interesting thread! I'm also kind of glad to see that your WildcatVP scores lower in 3D00 than in 3D01 as well (at least it isn't just me), but I'd love to know why they behave like that. 😕

Of the cards you listed for consideration, the V8 Duo Ultra is supposedly fastest (on paper), but it's plagued by awful drivers (iirc some reviews actually showed it to be slower than the single-gpu variants). If you wanted to consider "anything under the sun" as long as it's AGP-based the "best" are probably the ES simFUSION boards (2-4x Radeon 9700 boards). Since you specified 2002-2004 you could also include the Wildcat 4 boards from 3DLabs as well (3DLabs positioned them above the VP series); they're fairly cheap nowadays but they do require AGP Pro to hook up.

The REALiZM cards are reportedly *horrible* at gaming - check this benchmark for example: http://www.anandtech.com/print/1603

Also, could you provide fillrate/FPS values for the cards running through 3DMark (easiest to just take screenshots of the detailed results screen) - IME the 3DMark score is almost worthless for seriously comparing hardware performance (an example: I have two systems that both score around 20k - one has MT fillrate numbers in the mid-30000s, the other in the low 4000s; the "slower" machine scores so high because 3DMark likes that model of CPU better (their FPS values do reflect the higher fillrate system being better at gaming)). It's also helpful for comparing against the theoretical #s for the card and seeing how well they live up to specs.

A very nice test! thanks for the numbers!
Personally I use Geforce cards in my retro-builds, I like nvidia and the rest (except ATI) are "exotic" cards with hidden issues, I do prefer troublefree setups 😀

Great information, that would have taken you a while to put it all together!

I admit I haven't checked out this alternative video cards, likely never will. I guess raw speed is one thing and drivers are likely working well with certain benchmarks. But what about compatibility with games? Are there glitches, crashes?

This is the main reason I like sticking with Nvidia. With a period correct card and driver any game will work just fine.

I agree with you guys on nVidia for compatibility. Thus far my own experience with 3DLabs has been that the drivers are pretty stable, but that they do indeed have some weird compatibility glitches with certain games/applications (never had anything outright crash, but certainly have seen visual glitches).

Nice write up. + Rep.

When I benchmarked with my wildcat vp990 I got scores around 3900 for 3dmark2001se. But I think the 3dlabs driver puts vsync on by default - removing that the score was more like 6800. Different other hardware of course, though I was using an athlon 1600 so not that far off.

As you put 2003-2004 cards in there, wouldn't it be great if you include FX 5950 Ultra in your getlist? (OK, 9800XT too...)

I get 8330ish in 2001 with a stock geforce 5800, on a tualatin 1.4ghz.
Its amazing how a 4600 can keep up with it, despite the fact it uses ddr2 and runs hotter then heck.

what about X850?! The last ATI card that's not badly bridged to hell...

AlphaWing wrote:

I get 8330ish in 2001 with a stock geforce 5800, on a tualatin 1.4ghz.
Its amazing how a 4600 can keep up with it, despite the fact it uses ddr2 and runs hotter then heck.

That's around what my 5800U scores with my 2G Willamette (should be relatively close to the performance as a 1.2-1.4G P3) as well. Not so surprising that Ti 4600/4800 keeps pace in DX7 - the FX can't use its pixel/vertex shaders there. In DX8/9 the FX cards can be much faster. This review illustrates that pretty well: http://www.xbitlabs.com/articles/graphics/dis … geforce-fx.html

leileilol wrote:

what about X850?! The last ATI card that's not badly bridged to hell...

Isn't the X850 supposed to be one of the fastest (if not the fastest?) AGP cards that doesn't have bridge problems? 😀

obobskivich wrote:

Very interesting thread! I'm also kind of glad to see that your WildcatVP scores lower in 3D00 than in 3D01 as well (at least it isn't just me), but I'd love to know why they behave like that. 😕

Of the cards you listed for consideration, the V8 Duo Ultra is supposedly fastest (on paper), but it's plagued by awful drivers (iirc some reviews actually showed it to be slower than the single-gpu variants). If you wanted to consider "anything under the sun" as long as it's AGP-based the "best" are probably the ES simFUSION boards (2-4x Radeon 9700 boards). Since you specified 2002-2004 you could also include the Wildcat 4 boards from 3DLabs as well (3DLabs positioned them above the VP series); they're fairly cheap nowadays but they do require AGP Pro to hook up.

The REALiZM cards are reportedly *horrible* at gaming - check this benchmark for example: http://www.anandtech.com/print/1603

I believe that all the WILDCATS are terrible for games, it should not be a surprise, those cards were marketed for CAD applications. I just never thought they would be THAT slow...

I think that V8 Duo Ultra would be the fastest of them all 😉 I know that Ultra Duo beats DeltaChrome in some games. It would be interesting to test S8 Nitro too.
Yeah, the simFUSION is good option too but it would definitely blow away the competition, and it is more like a PRO card.

Mau1wurf1977 wrote:

Great information, that would have taken you a while to put it all together!

I admit I haven't checked out this alternative video cards, likely never will. I guess raw speed is one thing and drivers are likely working well with certain benchmarks. But what about compatibility with games? Are there glitches, crashes?

Thanks, It took some time but there is still a lot of testing to be done.
So far I have not tested any games, besides Quake III Arena, and synthetic benchmarks.
The most unstable card was V8, but only in 3D Mark 2001se (maybe becuase of the broken capacitors?)
If you have any suggestions concerning games to use for testing then please feel free to share your ideas. (I do not know much about games released after the year 2000....)

archsan wrote:

As you put 2003-2004 cards in there, wouldn't it be great if you include FX 5950 Ultra in your getlist? (OK, 9800XT too...)

FX 5950 Ultra and 9800 XT were the “best” options in 2003
6800 GT and X850 were the „best” In 2004, don’t you think those cards are too “fast” for this comparison? I used the fastest cards from ATI and Nvidia from 2002 just for reference, and still NON of the alternative graphics, realesed even 2 years later, managed to beat those two....

Or 6800 Ultra and X850 XT. Well isn't that the point? Ultimately it will show how those companies went out of the performance business. Even today the pro CAD/CG market is dominated by Nvidia's Quadro line.

VooDooMan wrote:

I believe that all the WILDCATS are terrible for games, it should not be a surprise, those cards were marketed for CAD applications. I just never thought they would be THAT slow...

One thing to note on the VP - the setting of the "GPU Optimization" slider can make a BIG difference in how they perform - setting it to Max Texture will improve performance substantially in games and 3DMark (on my VP880 it can be up to 4x improvement); the various application profiles can also have an influence on how 3DMark runs, but I have not tested all of the available profiles. I know the default for DirectX yields better results on the pixel shader tests in 3D01 than the "DirectX Games" mode, for example. I also know that changing settings within 3DMark can have an equally dramatic effect on how the card performs - for example in 3D00 setting 32-bit can cause glitches and absolutely sinks performance on some tests; I don't know if this is just an issue with 3D00 or something to do with the VP.

And yes, the Wildcat cards are professional accelerators - they were designed to compete with the Quadro 4; based on reviews I've read they tended to do a good job of that too.

If you have any suggestions concerning games to use for testing then please feel free to share your ideas. (I do not know much about games released after the year 2000....)

In no particular order:

- Half-Life 2 (will preference ATi cards)
- Counter-Strike: Source
- Doom 3 (will preference nVidia cards)
- FarCry
- Halo PC
- Unreal 2003 or 2004
- C&C Renegade and Generals
- Morrowind
- Wolfenstein: Enemy Territory (this one is free)
- Grand Theft Auto: Vice City or San Andreas
- Splinter Cell
etc

FX 5950 Ultra and 9800 XT were the “best” options in 2003
6800 GT and X850 were the „best” In 2004, don’t you think those cards are too “fast” for this comparison? I used the fastest cards from ATI and Nvidia from 2002 just for reference, and still NON of the alternative graphics, realesed even 2 years later, managed to beat those two....

6800 Ultra was nVidia's top model for GF6. 😊

To the question at hand: You aren't really comparing the "best" from any of the 3rd-party makers (and aside from XGI, I don't think any of them ever really tried to compete with the top GeForce/Radeon cards), and some of the cards you've got aren't gaming cards to begin with. My understanding is the S3, XGI, and SiS cards are more mid-range parts from their respective manufacturers, and it'd probably be more reasonable to stack them against mid-range cards like the FX 5600 or 8500LE. The Wildcats are pro cards, and should probably be compared to something like Quadro 4 or FireGL 8x00 (and with a test like viewperf) as opposed to top-flight DX9 gaming cards like the 9700Pro. Parhelia is kind of an odd-duck - part gaming card, part pro card, and very expensive when new; I'm not sure what's actually "fair" to compare to them as a result (basically, I don't think there's any other card from another manufacturer that has an entirely comparable feature-set). None of this will change how the cards you've already tested will perform, but it might provide better perspective on their performance. 😀

I will also watch this with great interest. Don't mind mashing cards from different segments together, any comparison of little known cards is valuable. But it could be indeed more interesting to revisit Radeon 8500 with late drivers, rather than the notorious 9700. Some cards could give more data points, for example underclock the Xabre 400 into 200.
If I may have just one suggestion for the tests, go for as many unoptimized games as possible, that means rather the opposite of obobskivich's list.

VooDooMan wrote:

I believe that all the WILDCATS are terrible for games, it should not be a surprise, those cards were marketed for CAD applications. I just never thought they would be THAT slow...

did you check the vertical sync setting?

Very interesting, thanks for sharing. I used a Xabre for a while in 2003, it was pretty decent at the time, cheap and the drivers were better than expected.