Image Quality Tests

Image Quality Test 1 - Engineering Image Quality

This image quality test consists of 7 separate tests. All 7 tests output a 256*320 .BMP file rendered with the currently selected 3D Accelerator in the selected Color Depth and Z-Buffer depth. The top 64 pixels of the image are used to store information about the rendering platform ensure image quality tests from different platforms can’t be mixed to each other. The rendered test area is the bottom 256*256 area of the bitmap. All source bitmaps for the alpha blend tests are stored in 24bit format.

Reference Images:

You can compare the result rendered with your 3D accelerator to the reference images provided with 3DMark 99. The reference images have been rendered with Microsoft Direct3D Software renderer. If you have run the benchmark with several different 3D Accelerators, you are able to compare the images against each other in the Result Browser. The Reference Images can be found in your installation directory, under the Quality\Reference folder, such as:

C:\Program Files\3DMark 99\Quality\Reference\

Use Result Browser's Image Quality Viewer for comparing the Image Quality pictures.

3DMark Quality Test 1: Sub-pixel accuracy

The lack of sub-pixel accuracy is is visible as twitching edges in slow motion or as visible seams between the polygons. This test draws a circle using thin polygons. The result should be a clean circle with easily distinguishable moire patterns.

3DMark Quality Test 2: Mip-Mapping

This test shows the difference of per-pixel mip-mapping over per-polygon mipmapping. If there is a visible seam towards the distance in the polygon and the other side of the seam is sharp but the other is blurry, the hardware can’t do per-pixel mip-mapping. This is mainly a problem in games that have large floor/wall/roof polygons with continous textures. The per-polygon mip-mapping makes the triangles pop out when they reach a certain distance from the camera.

Gradient textures

Overall, these examples show the maximum bit-depth for the textures (usually 16bit or 24/32bit). The combinations show how good the rasterizer is. All three gradient tests use a base texture with a spectrum and grayscale gradients. These images visible here in the Help files are downscaled and compressed .jpg pictures. The textures used in the benchmark are in non-lossy truecolor format.

3DMark Quality Test 3: Multiplicative Blend.

Original Texture Alpha Map Result

* =

Image 1. Image 2. Image 3.

Image 1 gets Image2 combined with"multiplicative alpha blending". The result is visible as Image 3.

3DMark Quality Test 4: Additive Blend.

Original Texture Alpha Map Result

+ =

Image 1. Image 2. Image 3.

Image 1 gets Image2 combined with "additive alpha blending". The result is visible as Image 3.

3DMark Quality Test 5: Alpha Blend.

Texture 1 Alpha Map Texture 2 Result

=

Image 1. Image 2. Image 3. Image 4.

Image 1 and Image 3 are combined with Image 2 as a transparency texture. The result is visible as Image 4.

3DMark Quality Test 6: Bilinear filtering

This test zooms into a small (3*3) pixel area in a texture. The main purpose is to show differences in 16bit and 32bit color modes. In 32bit color there should be no visible edges in the filtering. The view is also rotated 30 degrees.

3DMark Quality Test 7: Texture complexity

We are using a 512*512 texture. If the hardware isn’t able to handle large textures or uses texture compression, the result yields to a more blurred image.