When drawing for example lots of vegetation (with textures containing an alpha channel) you need to minimize overdraw and cost for each draw call. If you use blending for archiving the desired transparent effect on the leaves then you will need to sort all the triangles (back-to-front-order) to draw, (and sometimes this needs splitting of triangles) to get the correct look. Also when blending you might render the same pixel multiple times because you want the result to blend together. All this drawing and sorting takes a lot of time and a better way is to do alpha testing. In alpha testing (or alpha killing) you select an alpha threshold and all pixels with an alpha above (if you use GREATER as test function) this threshold will be drawn. Because no blending occurs, you don’t need to sort the triangles anymore (but preferably you will sort them front-to-back-order for optimization) . The drawback with alpha killing is that you will get sharp edges around the pixels drawn.
Alpha killing does not work with shadow volumes.

Alpha testing in OpenGL
http://opengl.org/documentation/specs/version1.1/glspec1.1/node96.html

Alpha testing in DirectX9 (in DirectX10 you must do your own implementation in a shader)
http://msdn.microsoft.com/en-us/library/bb172254.aspx

The full title of this paper is improved “Alpha-Tested Magnification for Vector Textures and Special Effects”. It’s about a technique to use vector textures when decaling for improved precision when magnificating. This kind of decal is useful for signs and such things that contains text or symbols because they can easily be represented as vector graphics. The left image below shows standard Alpha-Testing and the image on the right shows the improved version.

Link to the paper:
http://www.valvesoftware.com/publications/2007/SIGGRAPH2007_AlphaTestedMagnification.pdf

In some occasions a linear depth texture is preferred over the usual non-linear depth buffer (z-buffer). The linearized depth could for example be used when doing SSAO or depth of field. Rendering linear z-buffers is very easy and only the following lines of code is required ( a float texture should be set as render target):

Vertex shader:

varying float depth;
void main()
{
    vec4 viewPos = gl_ModelViewMatrix * gl_Vertex; // this will transform the vertex into eyespace
    depth = -viewPos.z; // minus because in OpenGL we are looking in the negative z-direction
    gl_Position = ftransform();
}

Fragment shader:

varying float depth;
void main()
{
    gl_FragColor.r = depth;
}

 
You might want to scale the depth in the vertex shader to a more appropriate range as the the distance will otherwise be the same as the distance from the camera to the vertex. The following vertex shader will map the near and far distances to 0..1 which is often a good idea.

varying float depth;
void main()
{
    vec4 viewPos = gl_ModelViewMatrix * gl_Vertex; // this will transform the vertex into eyespace
    depth = (-viewPos.z-near)/(far-near); // will map near..far to 0..1
    gl_Position = ftransform();
}

Here’s how the shader looks like if you render a cube.

Link to a page about the depth buffer in DirectX
http://www.mvps.org/directx/articles/linear_z/linearz.htm

A good introduction to the depth buffer in OpenGL
http://www.sjbaker.org/steve/omniv/love_your_z_buffer.html