Peter Wicker: T-Buffer

     Interview: Peter Wicker / 3Dfx T-Buffer/ Strategy plus / cdmag.com / 8.2.1999.

Rozhovor jsem ponechal v originále, beze změn. Myslím, že překladem bych mohl ovlivnit celkový význam. Pokud si na překlad sami troufáte, můžete to zkusit, poslat mi ho. Rád ho přidám.

Confused about some of the implications of 3dfx's new T-Buffer technology? We spoke with Peter Wicker, Director of Product Marketing at 3dfx, for some clarifications. Peter elaborates a bit about spatial anti-aliasing, the speed and memory requirements for T-Buffer effects, and 3dfx's commitment to fast frame rates.

With the "spatial anti-aliasing," is it possible to mark just certain parts of a scene that you want anti-aliased, or is that a sort of "turn it on and everything is anti-aliased, or turn it off and none of it is?"

Yeah, it is possible, because we can do it at the per-triangle level. But we don't see a reason, because….

Nothing looks worse…

Right, nothing looks worse. The issue, though, is that you never can tell, especially in a game that involves lots of motion, what the end user is going to be focusing on. You want everything to be anti-aliased spatially; it's just a basically good thing. That's an interesting thought, though. I guess we have the ability to do that, but that would definitely require application interaction, though..

What sort of speed hit do these T-Buffer effects incur? You say that you're still targeting a minimum of 60 frames per second, so you want to do them really fast, but are they absolutely free?

Oh, no way are they free.

So they still incur some sort of speed hit, but you should be fast enough to still deliver 60 frames per second?

That's right. One of the reasons that we're waiting until this fall to introduce them is that we're not going to deliver a feature that you can't use. We've been adamant about 32-bit rendering that you take such a hit in frame rates that it's not worth it. We chose not to have 32-bit rendering. And we chose not to introduce T-Buffer type effects until we could put out an architecture that has the fill-rate necessary to support it.

Along a similar line, then, how much memory does a T-Buffer take up compared to a typical frame buffer?

Yeah, it definitely takes more memory, because you're creating multiple copies. That's another case where we had to wait until mainstream memory moved to 64 megabit, and prices crashed. So then you could take, say, a 32MB frame buffer and do it real cheap. A few months from now, a 32MB frame buffer is going to be cheaper than a 16MB frame buffer is today.

So it's pretty safe to say that 3dfx's fall product will be available in a 64MB configuration, since you'll have to use a lot more memory.

No comment! (laughs) I can't talk about future product details! Please don't make me do it! Use your intuition…I think it's pretty good.

How much silicon does it eat up? Compared to the rest of the processor, how much space is taken up with T-Buffer logic?

Going back to our philosophy, it's got to be available at a mainstream price. So, we must remain within the economic limits that mainstream, high-volume semiconductor processes allow us. The bottom line is, there's got to be few enough gates in this technology to allow us to do a very mainstream product, and for us to, y'know, stay in business. I can't tell you how many gates it takes, but I can tell you it's very affordable.