15.3. Trade-offs

As previously mentioned, three methods can be used to generate noise values in a shader. How do you know which is the best choice for your application? A lot depends on the underlying implementation, but generally speaking, if we assume a hardware computation of noise that does not use texturing, the points favoring usage of the OpenGL Shading Language built-in noise function are the following.

• It doesn't consume any texture memory (a 128 x 128 x 128 texture map stored as RGBA with 8 bits per component uses 8MB of texture memory).

• It doesn't use a texture unit (texture units are a scarce resource if the hardware supports only 2 or 4 of them).

• It is a continuous function rather than a discrete one, so it does not look "pixelated" no matter what the scaling is.

• The repeatability of the function should be undetectable, especially for 2D and 3D noise (but it depends on the hardware implementation).

• Shaders written with the built-in noise function don't depend on the application to set up appropriate textures.

The advantages of using a texture map to implement the noise function are as follows.

• Because the noise function is computed by the application, the application has total control of this function and can ensure matching behavior on every hardware platform.

• You can store four noise values (i.e., one each for the R, G, B, and A values of the texture) at each texture location. This lets you precompute four octaves of noise, for instance, and retrieve all four values with a single texture access.

• Accessing a texture map may be faster than calling the built-in noise function.

User-defined functions can implement noise functions that provide a different appearance from that of the built-in noise function. A user-defined function can also provide matching behavior on every platform, whereas the built-in noise function cannot (at least not until all graphics hardware developers support the noise function in exactly the same way.) But hardware developers will optimize the built-in noise function, perhaps accelerating it with special hardware, so it is apt to be faster than user-defined noise functions.

In the long run, using the built-in noise function or user-defined noise functions will be the way to go for most applications. This will result in noise that doesn't show a repetitive pattern, has greater numerical precision, and doesn't use up any texture resources. Applications that want full control over the noise function and can live within the constraints of a fixed-size noise function can be successful using textures for their noise. With current generation hardware, noise textures may also provide better performance and require fewer instructions in the shader.