Reverend: One of the criticisms of 3dfx's Voodoo5 FSAA implementation is that the FSAA images seem to be comparatively "blurry" (another word used is "less detailed looking") compared to either with FSAA disabled or compared to its competitors' FSAA implementation. Is this simply a "side effect" of 3dfx's Rotated Grid Super-Sampling's "jittered" FSAA implementation? Going on further, would such other 3D effects like DOT3 bump mapping (which basically means more detailed and realistic looking surfaces) lose "detail" or look "slightly blurrier" when FSAA is enabled at the same time?

I haven't seen this in person, so it's hard to say what they are referring to here. But I will take a guess. First let me say that FSAA is not done by "blurring" the image. My guess is that if there is any "blurring" that is noticeable, it probably has to do with texture filtering. Since each subsample is textured independently, the effective result is a filter kernel slightly larger than a pixel. This has a huge benefit in reducing texture aliasing, at the cost of a slight amount of blurring. Let me emphatically state that "blurring" is not always bad - bilinear filtering and mipmapping offer great improvements in visual quality yet some describe these features as "blurring". The same for edge anti-aliasing, whether done via edge based methods or super-sampling or multi-sampling. In general, aliasing is reduced via higher sampling rates and filtering. The filtering often reduces the final displayed frequency and people often see this as "blurring", which it really isn't. It is "blurrier" than the aliased image, but that is because aliasing has been removed by eliminated frequencies that are too high to accurately display (they are beyond the Nyquist limit). This is a good thing, not a bad thing. Aliasing is bad, don't forget that! Some people still prefer point-sampling over bilinear filtering, but hopefully they are in a minority now.

Other effects, such as DOT3 bump mapping, would not be effected any differently than regular texture mapping. That is, if there is any "blurring" going on due to a larger texture filtering kernel, it would effect the results of DOT3 bump mapping the same as everything else. We provide LOD bias settings for users to adjust the image sharpness to their liking. Some people like sharp aliased images, some prefer smoother anti-aliased images

Reverend: Can you comment on what has been going on with DirectX8? Are its features (as they stand now... Lord knows how dependent Microsoft is on IHVs for DirectX) as solid as many make it out to be?

I can't comment just yet - as we are still under NDAs. Please monitor our Web site for information once DirectX8 is released. All I can say, is that we are working extremely closely with Microsoft in defining DirectX8 and we are very pleased with the results.

Wavey: Now that we're soon to have DirectX8 knocking on our doors, where do you think the next advancements will come from in the hardware? How long will it be before we have chips capable of 'Toy Story' in our PC's?

I believe it will take many years for developers to fully utilize the programmability that is coming in the next round of products. Every year, hardware will be a little better, and the applications will be a little smarter, and so you will see steady improvements in image quality. I imagine it will be like consoles which have fixed hardware for many years - except in our case the hardware is advancing as well as the software. So if you compare games in a few years to today's games, there will be quite an improvement. What I am saying is that programmability is a huge advancement and it will take quite a few years, perhaps many, many years, to fully show what it can do. As for the next advancement beyond that, that's a tough question, but I will stick to my guns and say pixel quality improvements. I feel there's still a need for some improvements in FSAA quality. But the larger need is along the lines of what John Carmack wants - bigger and better pixels.

I won't touch the second question, for fear of upsetting our friends at Pixar.

Kristof: T&L and the Rasterizer are now moving from a fixed layout to programmability - are you happy with it?

Yes. However it raises an interesting "philosophical" type question to ponder. And that is - what is the fundamental difference between a Geometry Engine and a CPU? Or what is the difference between a Raster Engine and a highly parallel CPU? As I said, these are "philosophical" type questions….

Kristof: How long before we start to see games that use the new programmability features?

You will start to see the first games or game demos next year. I believe most of the leading game developers who have pushed the state-of-art in features in the past, will continue to do so in the future.

Dave: An issue that arises when designing a programmable architecture is that of performance. You can make things completely programmable, but you end up with a loss in performance. If you make everything completely set, you then seriously lack needed flexibility. How do you go about deciding what balance needs to be made between programmability and speed?

Simplistically, we start with a speed requirement and work from there and see how much programmability we can design in. In the Voodoo1 and 2 days, we used mostly fixed function datapaths with dataflow (mux) selects to get the performance we wanted. Now, with technology advances, we can afford more gates and can add some programmability. This does cost performance, or costs gates to keep the performance the same, but is worth it. Game developers want programmability, and the effects they will achieve with it are well worth the effort.

Wavey: GigaPixel! Its been said that 3dfx virtually hemorrhaged a bunch of engineers over the past year or two - was this deal more about the quantity of (high class) engineers could get their hands on, or was it actually about the technology?

It was about the technology. We were also overjoyed to get a group of great engineers, but if it were just about the engineers we would have pushed the technology into a dark corner somewhere. Instead, the technology is an integral part of our future.