Introduction

Editor's note: Both NVIDIA and S3 have stolen some thunder from 3dfx this past week with their announcements on new video cards. What does 3dfx think of these new developments, and what new details are available on 3dfx's upcoming card? We sent the following questions over to the kings of Voodoo. 3dfx's CTO, Scott Sellers, was gracious enough to answer all of our questions for us. Questions are in bold, answers are in normal text.

Have you heard of Nvidia's GeForce 256 announcement? What are your first impressions?
Nvidia is a capable competitor and we never take them lightly. I found it interesting that they marketed this part with the auspicious "The world has changed" (or something like that) tagline. Sticking a geometry engine on top of a raster engine with little more fillrate and raster features than currently available products does not, however, deserve a "changing the world" status, in my opinion. They have done a nice job on the geometry side, though.

What is your opinion on the term "GPU?" (A GPU by Nvidia's definition is "a single-chip processor with integrated transform, lighting, triangle setup/clipping and rendering engines that is capable of producing a minimum of 10 million polygons per second"
Seems like everyone in this business is trying to come up with new and cool marketing name for new technology. I think nvidia's definition of a "GPU" is somewhat arbitrary and clearly slanted to GeForce's features. Certainly I don't think you'll see the term "GPU" become the de facto naming convention for on-chip geometry acceleration, however.

What is your take on Nvidia's decision to move to on-board T&L?
It is certainly an interesting position to take, but somewhat not surprising given the background and experience of many of their architects and engineers. I think what was surprising about the specifics of the GeForce 256 is how they seemingly ignored new rasterization features (cube environment mapping excepted) and certainly ignored substantial fill-rate improvements. As a result, it is our belief that they have created a very unbalanced part with impressive geometry capabilities but not enough fill-rate to really take advantage of it.

How important is support for non-square textures?
Well, our architecture has supported non-square textures since the original Voodoo Graphics so I guess they felt it was important to finally catch up with us and the rest of the industry. Rectilinear textures are used quite frequently by developers, so they are an important feature.

How significant are cube environment mapped reflections?
Cube environment mapping certainly makes for nice demos, as does spherical environment mapping which we first demonstrated back in 1996 with the launch of Voodoo Graphics. The real problem with cube environment mapping is that to generate the cube environment maps, the scene has to be rendered from six different viewpoints. There is therefore a tremendous performance penalty when enabling cube environment mapping for dynamic environments. Certainly the GeForce does not have either the geometry power,and certainly not the fillrate power, to increase performance requirements by a factor of six! So, I think cube environment mapping will be relegated to interesting demos showing static environments (notice that none of the demos nvidia showed which utilized cube environment mapping had any dynamically changing environments - they were always static environments so that the surrounding environment never had to be re-rendered six times). You'll see cube environment mapping show up as minor special effects in games in the future, but the processing power is not there yet to really even come close to utilizing cube environment mapping in a real game with dynamically generated environments.