The "default" settings are the default settings of Kingpin when I initially installed the demo. The miniGL provided with the demo is the one used in the benchmarks. The timedemo used was the one provided by Planetkingpin. The column factor tells you how much faster or slower a particular trait is in comparison to the default. For example, at 640x480 using the Mesa drivers, turning off fog gives you a 20% performance boost better than the default. Turning off fog with the OpenGL Beta 2.1 gives you a 48% performance boost at that same resolution. Turning off fog with the miniGL does not do squat.

As you can see, the best performance numbers were provided by Voodoo II with SLI enabled and the Voodoo II with SLI diabled, both solutions using the miniGL. The OpenGL Beta 2.1 is a horrible performer in comparison to the miniGL. It is, however, a better performer than the Mesa. Keep in mind that I am comparing the OpenGL Beta performance to the miniGL. It would be nice if the 3dfx zealots actually tried to use some perspective here. The Voodoo II with the OpenGL Beta 2.1 would still be the third fastest single-card solution in the market, losing out to the TNT and the ATi Rage 128. (If you count the Voodoo III, it would be the fourth fastest card in the market.)

In any case, Mesa is far, far, far more stable than the OpenGL Beta. In addition, the fog support with Mesa is much better even though it is still a little buggy--the models look ghostly when using Mesa. The OpenGL Beta provides horrible texture support when fog is enabled. It is highly unlikely anyone would want to play Kingpin with fog enabled while using the OpenGL Beta. With that in mind, anyone who provides benchmarks using the OpenGL Beta without mentioning the incredibly horrible image quality is either (a) not being honest and wants to prove that his or her 3dfx religious icon is better than the competition or (b) has a later revision of the OpenGL Beta 2.1 that is not publically or openly available. Remember what I said about giving people the benefit of the doubt. Remember that drivers/patches/demos have been leaked before.

  
OpenGL Beta 2.1 Fog Support Enabled

If you don't have this latest revision of the OpenGL Beta 2.1, how do you get 50+ FPS with the currently available OpenGL Beta 2.1 and prove that it is faster than the TNT2 Ultra? Follow these steps to greatness.

• If you have only one Voodoo II, buy yourself another Voodoo II and make sure SLI is ENABLED.
• Turn off fog, shadows, specular, and sync every frame. Turn them all off--set everyone of them to "no".
• Turn down texture quality to the lowest possible setting.
• Turn down pain skins to zero.
• Drop down to the console and type "props 0" and press ENTER.
• Then exit out of the game to save the settings.

When you are finished, run a timedemo test using the demo provided by Planetkingpin. Take a screenshot like the following:

Kingpin (800x600x16) at 54 FPS using OpenGL Beta 2.1

It should be noted that the benchmarks above provides some interesting patterns. For instance, enabling or disabling fog while using the miniGL has no effect on the performance. This should not be surprising since the miniGL does not have fog support at all. The OpenGL function glFog, for example, is ignored by the miniGL. On the other hand, disabling fog provides a massive 48% performance boost while using OpenGL Beta 2.1 at 640x480. It provides a 61% performance boost at the same resolution while using SLI. This same pattern is seen even with the TNT. Turning off fog provides a large performance boost--36% performance boost at 640x480, 25% at 800x600.

Remember 3dfx's propaganda compaign against supporting 32-bit rendering? They claim they won't support 32-bits if the performance suffers too much and if the frame rates are below 60. Well, we already know that nVidia's TNT2 can provide frame rates greater than 60 at 1024x768x32. So the question is: does 3dfx also have a policy against fog support? Is poor performance the reason that the miniGL does not support fog after all this time? Judging from the early numbers of the yet unreleased TNT2 Ultra, it seems that the TNT2 Ultra can not provide frame rates greater than 60 at 800x600x16 with fog enabled in Kingpin! Apparently, getting high frame rates with fog enabled is a difficult thing to do. It seems that after 32-bit rendering, getting fog support is 3dfx's next big hurdle or is supporting texture sizes greater than 256x256 their next big hurdle? nVidia apparently does not have this silly rule against adding features. They have fog support for a long time--it comes with their OpenGL ICD. They have support for textures sizes greater than 256x256. They have support for 32-bit rendering. And their latest product supports 32MB of memory.

It seems quite apparent that somethings can be kept fast--very, very fast--by keeping it very, very simple. If you don't add something, you don't have to worry about performance being hurt by it. It is that simple. It is the KISS--Keep It Simple Shithead--philosophy at work.

It is now time to tell you the biggest reason for writing this article: it is a follow-up to my previous The Truth behind TNT 1/2 vs. Voodoo II/III Benchmarks. It should not surprise anyone that the 3dfx zealots did not appreciate the article. In fact, time and time again, the zealots have proven that they know nothing about the card they worship. They spent so much time and effort defending their religious icon. If you spent $US300 on Voodoo II SLI (and for some people an additional $US50 to $US100 for a good 2D card), wouldn't you want to defend your religious icon, too?

Let us analyze the Kingpin benchmarks a little bit more closely. The miniGL does not support glFog. (I'm picking glFog as an example, the actual function call maybe different.) So a glFog function call will not have a performance drop with the miniGL. However, the TNT and the TNT2 cards are true OpenGL v1.1 compliant cards. This means that the TNT and the TNT2 support glFog. So this means that a glFog function call will cause a performance drop with OpenGL cards like the TNT and the TNT2.

Naturally, the 3dfx zealots are still lost and do not understand. The reason that the TNT and the TNT2 suffer a performance drop is quite simple. Games like Kingpin and others would make a glFog function call (and other associated functions). The TNT and TNT2 then makes several calculations and renders the result to your screen. This means that the TNT and the TNT2 will be "doing stuff" with the glFog function call. The Voodoo II using the miniGL DOES NOT SUPPORT glFog. This means that the Voodoo II will not be doing anything with a glFog function call.

It is so simple to understand. When one compares the benchmarks of the miniGL to the benchmarks of an OpenGL ICD, they are potentially comparing apples to oranges. In the above simple example, Kingpin makes glFog function calls (or something similar) that the miniGL does not support. So therefore, the Voodoo II/III will be doing less work, less rendering (aka special treatment) than an OpenGL compliant card such as the TNT. This is the benefit of a miniGL with a "reduced instruction set". The 3dfx zealots seem unable to understand this simple theory. If the OpenGL compliant cards like the TNT were not told to execute the glFog function (or something similar), then you will see the performance numbers of these cards rise. It is that simple.

Therefore, if someone told you that a Voodoo II SLI with OpenGL Beta 2.1 is faster than a TNT2 Ultra, you should know right away that the results were "doctored". That is, the TNT2 Ultra results include fog being enabled. The Voodoo II miniGL does not support fog so it does not matter either way if fog is enabled or disabled.

In the above simple example, I singled out glFog because Kingpin made it easy to single out. Are there other OpenGL function calls that we have to worry about?

I can see it now. 3dfx zealots are going to use Kingpin as a benchmark test to prove that their new religious icon--the Voodoo III: 3500--is faster than the TNT2 Ultra. They will compare high settings to high settings. That is, they will have fog enabled, sync every frame on, etc., etc. with their tests knowing full well that these options have absolutely no effect while using the miniGL. These options, on the other hand, have a negative effect on the performance of the TNT2 and other cards as well. They will have all these options turned on so that the performance numbers for the TNT/TNT2 would drop. And these are the same people who have no clue when I say that comparing miniGL performance of the Voodoo II/III to the OpenGL performance of the TNT/TNT2 is like comparing apples to oranges. The best these zealots can do is try to provide a detailed discussion saying that every card comes with its own driver. How small minded these people are! Here's a taste of their intelligence:

If driver performance shouldn't be considered part of a 3d card's performance then run yours without any and see what kind of framerate you get.

There is no doubt in my mind that 3dfx will make a killing with their Voodoo III series. Look at the type of people who are going to buy them. 3dfx could be selling a box full of rocks and still make a profit. Re-read the above quote again. See how stupid that individual is. The topic of this article and my previous article was on BENCHMARKS and COMPARING the performance of various cards based on these benchmark results.

Suppose Paul Steed of id Software is set to race against Michael Johnson in a 1000-meters race. Wouldn't you want both participants to be treated the same as reasonably possible? Afterall, you don't want Paul Steed to run the race using a 10-speed bike and then proclaim to the world that Paul Steed is faster than Michael Johnson, right? So why isn't this same stupid criteria applied to the benchmark results?

And look at how hypocritical these zealots are. They bitch and moan about how unfair Tom's Hardware is to the 3dfx cards with his benchmark tests. The 3dfx zealots bitch and moan that the crusher timedemo tests are practically worthless benchmarks. They accuse Tom and others of deliberately setting up the tests to make 3dfx cards fail. They want Tom and others to use Quake II demo1 test as the benchmark standard. Yet, these same zealots see nothing wrong with comparing TNT2 with fog enabled to the Voodoo II SLI with miniGL (knowing full well that enabling/disabling fog has no effect on the miniGL).

Does anyone remember the TNT2 preview by Computers R Us? Here's a short blurb from their preview:

This is where things were starting to get interesting, because the next thing was showed was a demo by nVidia. It was a fly by scene of a city running with 24bits Z-buffer accuracy and using 2048*2048pixels sized textures, the scene didn’t look very special until Jan zoomed in a bit, and we we’re looking at the rooftop and a penthouse on one of the skyscrapers. The scene was beautifully rendered with a pool of water and furniture.

Then he zoomed in even further onto a statue of a lions head, he kept zooming even closer and as we were getting even closer I suddenly noticed a bee on the nose of the lion, he kept zooming in even closer until we were looking at one of the feelers of the bee, even that was extremely well detailed (it even had tiny "hair" on it!!).

I'd like to end the article with another controversial theory. Many of us recognize that the 16-bit vs. 32-bit graphics rendering debate will not die until 3dfx finally decides to support 32 bit rendering. Did you realize, though, that certain key elements within the 3dfx community want to pursue this debate over and over and over again? The reason: discussing the 32-bit graphics issue distracts you from discussing the 256x256 texture limitation of the Voodoo I/II/III/Banshee cards, not to mention the lack of the true OpenGL v1.1 support. Think about it. Why is 32-bit rendering and the related "22-bit graphics" taking up so much of the headlines? It is because we (including myself) have been duped into talking about it so much. Apparently, 3dfx's 16-bit rendering is finally just as good as the 16-bit rendering of their competitors (eg. ATi Rage 128, TNT, etc.). They hype up this improvement (may actually be better than the competitions') by calling it an "effective 22-bit rendering" while simultaneously brushing off the other limitations of the card. Here check out for yourself in this pro-Voodoo III Kristof Beet's article from Beyond 3D:

I mean why support upto 2048x2048 textures? Such a texture requires 2048x2048x2bytes = 8 Mb of storage! Naturally, you want MipLevels: add 33%. This means that such a texture requires a massive 10Megs... not usable is it?

It would seem that the 3dfx zealots have already forgotten that the TNT2 can hold up to 32 megs of data while their religious icon, the Voodoo III, can only hold upto 16 megs. Therefore, 10 megs (10.64 megs actually) is far more significant on a Voodoo III than on a TNT2. Remember, we are talking about 10 megs of textures, the cards need to hold other data as well. The TNT2 simply has more space to hold that other data while the Voodoo III has far less space. Thus, you will see a significant performance drop with the Voodoo III. It is like their famed fog "support". They won't include 2048x2048 texture support because their 16 megs of memory can not make optimum use of such detailed textures.

It should be obvious that a lot of Voodoo's limitations (and lack of features like fog) are all related. So if they fix one issue, a bunch of other "issues" will be fixed as well. This means that 3dfx' next generation card may be a huge leap over the Voodoo III (and I'm not just talking about speed, either.)

On a related note regarding image quality, notice that to take pictures of Voodoo III images, you need another program that has the correct filter setting. You can't use just any program to do that, not even the famed "print screen" would work. Apparently, you don't need some special provisions with TNT/TNT2 shots. As long as this special program does not do anything to the TNT/TNT2/other cards' screenshots, there should be nothing wrong. So I hope this special program knows not to apply its "22-bit rendering trick" on TNT2 screenshots and only use it on Voodoo III screenshots.

Notes: I've noticed that "sync every frame" option apparently has no effect with the miniGL. I have no idea why this is. I was under the impression that it would make a difference. This is a non-issue anyway since most people like to disable this while doing benchmarks. Also I've used the latest Voodoo2 Windows 95/98 Reference Drivers. The Mesa drivers is Mesa v3.0. The OpenGL benchmarks used the currently available 3dfx's OpenGL Beta 2.1. A later revision of this OpenGL Beta 2.1 may be out there on the internet but was not used for these benchmarks. If I don't have them, I can't use them for the benchmarks. Planetkingpin has posted benchmarks for the Voodoo cards and the TNT. You can compare their performance numbers with mine. It would seem that some "regular" gamers have kick-ass machines. For the 3dfx zealots, it should be pointed out to you people that Planetkingpin's "high settings" is quite similar to the default settings. The only difference is that shadow is set to "auto" by default instead of "yes".

[an error occurred while processing this directive]