Aalchemy

So why would a Voodoo5 have a longer life span than a card with T&L? One reason is because it's feature-proof, as strange as that may seem. Hardware T&L cards like the GeForce have their feature set pretty much frozen. For example, the GeForce supports lerping between 2 matrices from what I understand, and no more. Developers basically cannot use this feature, and require 4 matrices at a minimum. When APIs support this, the GeForce will have to fall back to software T&L or not expose the feature. The Voodoo5 uses software T&L all along and thus can implement any feature that comes along. Basically hardware T&L will accelerate (and even that is questionable) today's features only. Once new features come along, these cards will either not support the feature or fall back to software t&l. Note: this is for gfx chips that are hardwired, ie. not microcoded, heavily microcoded chips can implement most new features in microcode. So I would argue that hardware T&L does not increase longevity and can actually shorten it.

I agree with Gary on point #1, that the current DirectX7 / GeForce approach to "two matrix blending", which is shown off in NVidia's "bending cylinder" demo, isn't useful for skeletal animation. For skeletal animation, you need support for many matrices, with per-vertex *indexes* as well as blending factors.
I don't believe that simply having 4 matrices is enough; for realistic animated characters, you really need "lots" (i.e. 12 to 16 matrices), and up to 3 or 4 per-vertex matrix indices. In other words, any card that implements the DirectX7 spec is not going to be useful for accelerating skeletal animation in a realistic game scenario.

The new DirectX8 features add a huge amount of programmable flexibility that enable one to do implement skeletal animation with 100% hardware acceleration, and far more. Supporting this will require one of the upcoming 3D cards which support the new DirectX8 feature set.

Accumulation Buffer support



ATI supports an accumulation buffer to support multi-sample effects like Motion Blur and Depth-of-Field. Undoubtedly everybody already knows what these effects are, if not check out our 3dfx T-Buffer articles.



The accumulation buffer works by rendering the image to an off-screen buffer (kind of like rendering to a texture) and then you apply an accumulate step in which you copy and combine the off-screen buffer data with the data already in the framebuffer. By repeatedly rendering to the off screen buffer and combining this data into a final image inside the framebuffer we can create the multi-sample effects like Motion Blur. Imagine rendering a sphere to the off screen buffer and then blending it into the framebuffer, if you move the location of the sphere for the various steps you'll end up with multiple copies of the sphere at different locations blended into one final image… this is motion blur. Accumulation buffer based multi-sample effects are supported by OpenGL and DX8.