NVIDIA GeForce 256 Review
3D WinMark 2000 Benchmarks
New to Ziff-Davis' release of 3D WinBench 2000 is support for Microsoft's DirectX 7 and includes tests specifically for transform and lighting capabilities. This allows the GeForce 256's hardware accelerated transform and lighting processor to be measured under DirectX 7.
Transform and lighting can be disabled or enabled in 3D WinBench 2000 when tests are run (see screenshot below), which is what these benchmarks are based on. The benchmarks are also based on the 3D WinMark 2000 series of tests.
I asked NVIDIA's Nick Triantos if he could elaborate on the results for these benchmarks and he gladly offered his expertise:
MikeC: I was wondering if you could help me out with some analysis of the 3D WinMark 2000 T&L tests. More specifically, if you look at the results, you'll notice there are certain tests, where T&L is enabled, that provide significantly better performance than when T&L is disabled.
For additonal 3D WinMark 2000 results from a GeForce 256, check out ZDNET's Benchmark Results Page from their November 15, 1999 graphics card roundup.
Look at the results for tests 1 (speedway), 2 (hangar), and 9 (chapel) to see a huge difference. However, if you look at other tests, such as 3 and 5, there is virtually no difference in performance.
What makes tests 1, 2, and 9 perform so poorly without T&L, while 3 and 5 are running about the same? Does it have to do with the way these tests are using specific features of DirectX 7 ?
NickT: Sure, it's actually quite simple. Those 3 tests you say run great on the GeForce are more of the scale of apps that will be created on DX7. They have between 18K triangles (test 9), and 65k triangles (test 1). While their fill requirements are also large, they really stress the heck out of transform and lighting.
Test 3 was designed to measure multitextured fill limit. It only has about 1800 triangles in the scene. Test 5 has only 2500. Both of these are, obviously, carried over from WinBench99. With that few triangles, it seems plausible with a "reasonable" CPU (i.e. P2/450) to transform the vertices fast enough that the test really only tests the fill limits of the card.
Speedway - Test 1
Hangar - Test 2
Hangar - Test 2
Disable/Enable T&L Hardware Acceleration
In addition to generating a frames per second result for individual tests, 3D WinMark 2000 also provides an overall result. The frames per second for each individual test are totaled and the average frames per second is calculated. This result is called the 3D WinMark 2000 score.
The following settings were used for the tests:
The results show that transformation and lighting is alive and well in DirectX 7 and that the GeForce 256 is doing its job very nicely.
- 32-bit color
- 32-bit z-buffer depth
- Double buffering
- Flip rendering mode
- Full screen
- 75Hz monitor refresh rate
- Overclocked to 135MHz/180MHz
The 3D WinMark 2000 results database can be downloaded here - 10K.
3D WinMark 2000 - 800x600@16bpp
3D WinMark 2000: T&L Disabled - 69 T&L Enabled - 106
3D WinMark 2000 - 800x600@32bpp
3D WinMark 2000: T&L Disabled - 54 T&L Enabled - 78
3D WinMark 2000 - 1024x768@16bpp
3D WinMark 2000: T&L Disabled - 55 T&L Enabled - 82
3D WinMark 2000 - 1024x768@32bpp
3D WinMark 2000: T&L Disabled - 41 T&L Enabled - 56
3D WinMark 2000 - 1280x1024@16bpp
3D WinMark 2000: T&L Disabled - 41 T&L Enabled - 57
3D WinMark 2000 - 1280x1024@32bpp
3D WinMark 2000: T&L Disabled - 28 T&L Enabled - 36
Next: TLCMark Benchmarks