Software Implementation and other Hurdles

What do you think developers will do regarding T&L? Are they willing to code specifically for something that isn't supported by the vast majority of hardware available?
I think like any new 3D graphics technology there is going to be a slow ramp of games which actually support hardware T&L. I am sure that by the first half of next year you're still not going to see any substantial number of titles supporting hardware T&L. This is why we like the T-Buffer and its full-scene Anti-aliasing capabilities so well, because users can immediately enable full-scene AA on existing titles. There is no waiting around for games to take advantage of the feature. It's probably the most compelling "out of box experience" piece of technology we've ever created.

What do you think of the increasing power consumption on today's video cards? Are we going to need a major revision of AGP to handle ever-increasing demands?
Power is definitely becoming an issue as we move forward. This will be yet another technical hurdle that will separate the well-run engineering organizations from those that are not so well-run and have difficulty executing. We are in constant communication with Intel regarding future AGP requirements, and certainly increasing the power budget is one of our requests.

What is the most important speed-improving demand you see in the near-term future? Long-term?

We believe in the near term that full-scene anti-aliasing is going to reset everyone's expectations of what a 3D graphics accelerator is capable of doing. While people that buy the GeForce will be waiting around for games which actually take advantage of its geometry capabilities, our customers who buy the T-Buffer enabled products will be happily playing their games like they've never experienced before. We think consumers will quickly realize this and opt for the product that substantially enhances hundreds of available titles today, instead of waiting for a handful of titles tomorrow.

Where are the bottlenecks in today's 3D accelerators?
All over the place: CPU frontside bus bandwidth limitations, memory bandwidth limitations, AGP/PCI bandwidth limitations, local memory bus bandwidth limitations, CPU floating point processing performance bottlenecks, pixel generation bottlenecks, and the list goes on and on….

What is your opinion on non-triangular polygon, voxel, or spline acceleration?
There is a very good reason that triangle-based rendering has lasted so long, and that is because it is the most intuitive, robust representation of 3D objects around. Voxels have been around for a very long time, and the problems associated with rendering voxels in real-time have yet to be resolved. Other technologies such as higher order surfaces can always be decomposed into triangle for 3D hardware acceleration.