Tweak3D - Your Freakin' Tweakin' Source!
NVIDIA Ultra TNT2 (Page 2/3)

Posted: April 18, 1999
Written by: Dan "Tweak Monkey" Kennedy
Estimated retail price: $199.99


Test system
  • Pentium II 450
  • ABit BX6 mainboard
  • 128 MB PC100 SDRAM
  • NVIDIA TNT2, Voodoo3 2000, Voodoo3 3000, Canopus Spectra 2500 (NVIDIA TNT), STB Black Magic 3D 12 MB Voodoo2 / SLI
  • Drivers: NVIDIA TNT2 - pre-release drivers (3-9-99); Voodoo3 - pre-release drivers (3-20-99); NVIDIA TNT - Detonator (2-17-99); Voodoo2 - 3.01.01 (2-1-99)
  • 19" Optiquest V95 monitor
  • VSYNC DISABLED with all cards
  • Diamond Monster Sound MX200
  • Windows 98
To test the performance of the TNT2, I ran my favorite set of benchmarks: Quake II, GLQuake, and Turok 2. All tests were run three times, and the best score was taken. VSYNC was disabled for all cards. The TNT2 OpenGL 1280x960 results (Quake II demo1, Quake II Crusher, GLQuake, and also the averages) were from a Windowed display, so these results will by all means improve when the product is released, because it will run 1280x960 full screen. All other programs were closed for testing. No autoexec.cfg or config.cfg files were present. Anti-aliasing was disabled.

Note - Benchmarks results are subject to change because the product has not yet been released...

Turok 2 T2Mark test (turok2.exe -benchmark) [first transition only]

Additional settings: Detail set to "High", Trilinear filtering enabled.

Turok 2, a Direct3D based first person shooter, was used as a benchmark for numerous reasons. I admit that it's a boring and slow game, but the engine supports several hardware features and 32-bit rendering. It's one of the newest and most up to date Direct3D benchmarks available.

The TNT2 takes the number one spot for every resolution in Turok 2. For the most part, as the resolution increases the TNT2's lead over the Voodoo3s extends. At 1280x960, the 16-bit TNT2 is almost 20% faster than the Voodoo3 3000. At 1600x1200, it is 15% faster. The TNT2 using 32-bit rendering keeps neck and neck with the Voodoo3 2000 through almost all of the tests. The 32-bit rendering hit is not easily distinguishable at low resolutions, but at 1280x960 and higher it's easy to see. Overall, the TNT2 is the king of Direct3D, at least for this game.

The visual quality of the two "classes" (Voodoo and TNT) is not very different with this test. For the most part with Turok 2, all looks about the same on the TNT2 as it does on the Voodoo3. The 32-bit color version offers more realistic looking explosions and much better lighting, however.

GLQuake Timedemo 2 (timedemo demo2) [-bpp 32 for 32-bit rendering]

Additional settings: Joystick disabled, 16-bit color for all platforms, except 32-bit TNT2.

GLQuake is a classic, yet excellent, benchmark because it always shows a variety in performance and visual quality. Also, it is not very CPU dependent, so the results are based primarily on video speed.

The TNT2 rendering 16-bit was right between the Voodoo3 3000 and 2000 for all the tests, except the windowed 1280x960. As the resolution increases, the TNT2 again gains on the others, coming within .5 FPS of the Voodoo3 3000 at 1600x1200. The TNT2 with a 66% higher clock speed than the TNT manages to score almost twice as high in nearly every test. The Voodoo3 3000 remains on top for all the tests, but the lead could go back and forth with new drivers and optimizations from each company. The 32-bit TNT2 trails right behind the Voodoo3 2000 in all the resolutions except the somewhat negligible 1280x960 windowed test. The TNT2 tests and Voodoo3 tests indicate that these cards would all be tolerable for running GLQuake at 1600x1200.

The visual quality with GLQuake is accounted to brightness and detail. The TNT is darker than the Voodoo cards, and consequently offers a more vivid and rich output. The detail differences are more obvious at lower resolutions. At 640x480 and 800x600, the TNT cards easily show a crisp picture, but the Voodoo cards show a blurry and somewhat distorted image. These differences are negligible at high resolutions of 1280x960 and up.

Quake II Timedemo 1 (map demo1.dm2)

Additional settings: Joystick disabled, 16-bit textures, high quality sound, 16-bit color for all platforms, except 32-bit TNT2.

Quake II's least intensive benchmark, demo1, is perfect for a comparison of what single player gameplay is like at various resolutions. The action is not intense, but it offers a great variation in performance because of this.

Quake II shows slightly different results than GLQuake. Now the TNT2 (16-bit) is struggling to keep up with the Voodoo3 2000. As the resolution is increased, the TNT2 moves in for the kill, but is stopped at 1600x1200, only 2 FPS short of the Voodoo3 3000. At 1600x1200, the TNT2 again scores twice as high as the TNT. The 32-bit TNT2 is down near the TNT, with fairly low benchmarks. However, as the resolution is bumped up, the TNT2 (32-bit) draws closer to the others as a result of its much higher fill rate. This test shows that the Voodoo3 is going to be king of the OpenGL Quake engine game at least for a bit longer. Maybe new drivers will make the difference, but it's hard to say at this point in time.

The visual quality is magnificent on the TNT2 with Quake II. Overall, it is better than the Voodoo3 on many levels. It looks cleaner, less blurry, and more rich on the TNT2. A little gamma adjusting is needed to acquire a better picture though. I recommend using: gl_modulate "2.5", vid_gamma "0.9", and intensity "2". If there was a declared "winner" for this specific test, I would say the TNT2 is better because the benchmarks are very close, and the visual quality is superior with the TNT2. Besides, who needs an extra 10 FPS at 800x600 when you're already at 90 FPS?

Quake II Crusher demo (map crusher.dm2)

Additional settings: Joystick disabled, 16-bit textures, high quality sound, 16-bit color for all platforms, except 32-bit TNT2.

The Crusher demo, designed by Brett "3 Fingers" Jacobs, is probably the most intensive Quake II benchmark available. The demo simulates a massive multiplayer match, and the action is constant. The reason I chose to use this benchmark is because it shows how bad things can get for the card. These results are about as bad as it gets for Quake II.

The Crusher shows the evolution of the TNT2 as it increases resolution. The 16-bit TNT2 crushes the Voodoo3 2000 in all tests except 800x600, and manages to significantly beat the Voodoo3 3000 at 1600x1200. The 32-bit TNT2 surfs right behind the Voodoo3 2000 and even passes it at 1600x1200.


Note - This is an average of the previous tests. Results were averaged to show the estimated performance at various resolutions, and in general (Total). Voodoo2 and Voodoo2 SLI results were excluded because they did not include all resolutions.

The averages show one important concept: the TNT2 can compete with the Voodoo3. I predict that by the time the TNT2 is ready to ship, it will beat the Voodoo3 on just about every level.

3Dfx brushed aside the Riva 128, fought hard against the TNT, but will have to battle vigorously to defeat the TNT2. This chip not only takes away the one previous 3Dfx advantage, (performance) it adds an incredible set of features that could possibly beat 3Dfx into submission.

[ Search for the lowest Prices of this Product! ]

Next Page

  • News
  • Forums
  • Tweaks
  • Articles
  • Reviews