Voodoo 2/3 vs. TNT/2 [March 22, 1999]
        - Written by legion

One of the things that people love to talk about in regards as to which graphics card is the best is the performance issue. That is, what kind of frame rates can I expect to get with such and such a card? The weaklink is that these frame rates can not be obtained without benchmarks. The only way to get these benchmarks is to buy the cards and run some tests with it. Unfortunately, most of us are not made of money so we have to rely on others to do the tests for us.

Now here is the problem. These people have machines that most of us don't have. Even if some of us are lucky enough to get a high-octane 400Mhz Intel chip, there are even fewer of us who can afford to get a high-octane monitor. I see people doing benchmarks as high as 1600x1200 resolution in Quake II. My monitor can't support such a resolution. Moreover, my monitor refresh rate is not as high as some of the lucky few. This means that even though we have enough horsepower of a CPU and a kickass graphics card to go with it, our monitor may be letting us down by limiting our frame rate. Instead of getting 120 FPS at 1024x768, I "only" get 85. Instead of 90 FPS, I'd get just 60 FPS.

The benchmark tests themselves are also the weaklink. As the Average Joe would like you to believe in Bad Conclusions from Premature Data, Tom Pabst of Tom's Hardware does not know how to perform "good" 3D cards benchmarking tests. In fact, the Average Joe would have you believe that Tom Pabst is only telling you half the story. Average Joe will gladly tell you, for instance, that the TNT2 gives you a 40% drop in performance in going from 16-bit rendering to 32-bit rendering using the Forsaken "Ship" demo. The conclusion then is that for low-CPU intensive tests, the TNT2 is simply a faster cousin of the TNT. The average Joe, however, fails to mention that the TNT2 still delivers 70+ FPS (Sharky Extreme) in Forsaken at 1024x768 using 32-bit rendering. He fails to mention that at 16-bits, the frame rates are about 120 FPS in Forsaken "Ship" demo. He also fails to mention that his beloved Voodoo III: 3500 (Sharky Extreme) is actually a shade slower than the TNT2--120 FPS for the TNT2 vs 118 for the Voodoo III. So it is true, the TNT2 is a faster cousin--a MUCH FASTER cousin. It is so fast that it can still deliver frame rates greater than 60 in 32-bit rendering. If the Voodoo III: 3500 is so fast (as the benchmarks indicate) and it is roughly as fast as the TNT2 in 16-bits, it is a wonder why 3dfx is not including 32-bit rendering support? Their fastest card will cost roughly $40 dollars more than the TNT2! Why should we buy it, then?

But let me return to the topic at hand and talk about the benchmarks. Did it ever occur to people like the Average Joe that the so-called tests were designed to slant the results in 3dfx's favor? Let us look at the different APIs available for the TNT. There is Direct3D and OpenGL. When I play Half-life using the TNT, I see a marked difference in performance between OpenGL and Direct3D. This means that the API plays a role. When people use Quake II as a benchmarking device, what API do you think they are using? When they use the TNT, they use "default OpenGL". When they use 3dfx cards, they use "3dfx OpenGL". When I look into my Quake2 directory, I see a file called 3dfxgl.dll. I do not see a file called nvidiagl.dll or anything like that. Instead, Quake II uses the same "default OpenGL" for the TNT as it does for the i740 cards. The 3dfx cards get treated special. If I can see a difference between Direct3D and OpenGL, then why shouldn't I expect a performance difference between OpenGL and miniGL?

When one compares the TNT to the Voodoo II SLI while using Quake II as a benchmarking device, they are actually comparing apples and oranges. There is more than just two different cards being compared here. These tests are also comparing two different APIs. Look at the table below and you can see what a more fairer tests shows. The TNT is actually faster than the Voodoo II SLI. It is only when you specifically use drivers meant for the 3dfx cards do you see kick-ass high frame rates. This is no different from saying that using glide with 3dfx cards gives you faster performance than using Direct3D with 3dfx cards. The TNT cards do not use "specific" drivers. It gets treated like any other OpenGL cards such as the i740.