T-Buffer Physiology

Let's do another take

In simple words, the T-buffer works by storing and combining multiple renders before a scene is displayed. If you want a certain effect, you run it several-times and the T-buffer will perform additive writes into the back buffer for each successive pass. When complete, flip the back and front buffers to display the final image, and repeat for the next frame. If you're familiar with how stereo glasses such as the Elsa Revelators work, you've got an idea of how some of the T-buffer functions work.

Motion and movement

For effects such as motion blur, soft shadows/reflections, and field of view, the T-Buffer will draw between 4 and 8 renderings from different perspectives (remember it's not just positional perspectives) to create the appropriate effect. For motion blur, the T-Buffer will store images of the object's motion through time to create a "multiple exposure" effect, which at high speeds can effectively simulate motion blur.

Subltle motion blur

Fast motion blur

The same general principal is used for depth of field effects. By storing multiple renders of a scene through different focal settings, the T-Buffer creates an effect where out-of-focus areas appear to be blurred with misaligned additive layers, while the in-focus distance remains sharp.

Depth of field near

Depth of field far

General blurs

Now that we have a basic knowledge for the kind of effects the T-Buffer can create, it's easy to see how it can implement effects such as realistic shadows and reflections. Touted as "soft shadows and reflections," the effect is prevalent on smooth or semiglossed surfaces. Reflections and shadows close to the surface are sharp in crisp, and they become increasingly feathered and blurred as the distance to the surface increases.

Using the T-Buffer, we can assume that a similar multiple-render technique is used for both reflections and shadows - by adjusting the transparency and angle for each reflection/shadow pass, the additive effect near the surface will be sharper and clearer, and the effect will taper off as the additive translucent layers separate by angle.

Pre-rendered example of soft-shadows

What power is needed?

Wait…doesn't a multipass technology such as this take up cycles a lot of rendering cycles? Well of course. What that does intimate is how powerful a card 3dfx's next-generation technology must be. Quite simply, a custom configuration of 8 Voodoo2 cards was unable to play back the Rocket Burger demo at 1024x768 with T-Buffer effects at anywhere near 60fps.