Wavey: What's your take on the tiling architecture that GigaPixel employs? Its certainly quite a radical departure from 3dfx's history in 3D graphics - is it sufficiently different to other Tilers (PowerVR) to make it more into the mainstream? Will it overcome the traditional Tiler issues when low poly counts are reached?

It is radical in the sense that it is region-based, but in many other ways, it is not a departure from 3dfx's history. The multitexture pipeline within the GigaPixel design was very similar to what we were designing. The GigaPixel architecture was designed from the beginning to be compatible with existing APIs, so there are no issues with it being mainstream. I believe your last question was meant to ask about high polygon counts - and that was answered in a previous question.

Wavey: Were 3dfx 'caught with their pants down' over bandwidth saving solutions? The very success of 3dfx started when there was a large drop in memory prices to the point that Voodoo Graphics could become commercially viable in the consumer market place, and it appears with the likes of V5 (especially the 6000) you have been dependant on those prices staying low - now that ATI have already released 'HyperZ', nVIDIA will soon have 'Occlusion Detection', it seems like 3dfx will not have anything in the bandwidth saving area until the chip after Rampage. Was the GigaPixel deal partly because 3dfx realized they were behind in this area?

We have been working on memory bandwidth savings for many years. The FXT1 texture compression technology was created to reduce memory bandwidth. In addition, we created scalable architectures so that we could overcome the bandwidth limitations of a single chip. We also researched many other memory savings techniques including region-based architectures. We thought buying GigaPixel was the best move for us, greatly reducing the time-to-market for this kind of technology. The HyperZ technology that ATI has is a good start, but it doesn't go nearly as far as what GigaPixel technology can do, the savings are rather small percentage wise. The "Occlusion Detection" technology is not shipping yet, so I can't say anything about that until it can be run through the wringer, as they say.

Wavey: It been said that the GigaPixel tech will not have any effect on 3dfx hardware until the chip after Rampage. Does this mean that Rampage will be kinda 'on an island' where the development is concerned? We know that Rampage is a completely new architecture, and presumably anything built with GigaPixel will be as well - with the Voodoo series seeing quite a successful life time, and presumably GigaPixel being in 3dfx's future, won't that leave Rampage out on its own? (If so, could this not have issues with subsequent support etc.?)

As I said in one of my previous answers - the tiling architecture of GigaPixel technology is radically different than previous 3dfx chips. However, this is transparent to applications, except that they get a big boost in performance. The sections of the design that are not transparent to applications, e.g. the multitexturing pipeline, are very similar and so there will be an amazing amount of continuity between products. Our goal is to have zero discontinuity.

Reverend: Video card reviews on the Web evidently are very important to a company like 3dfx (or NVIDIA, ATI, etc. for the matter) - what's your personal opinion on the general quality of such reviews? Would you approach reviewing video cards differently if you were a reviewer? How so? On a related note, would you care to comment on this little rant of mine, which is essentially about me approaching reviewing video cards with the primary concern/emphasis on games instead of the hardware?

I am disappointed by the quality of journalism, including reviews, on the Web. The Web is great in that it allows everyone to be heard and communicate with each other - the ultimate free speech medium. But one of the problems it creates is that everyone can become an "instant journalist", resulting in a tremendous amount of misinformation and sensationalism. Unfortunately, it seems that sensationalism sells and that truthfulness of information is often judged by how often it's repeated and linked to, rather than its contents. Thus, misinformation that is sensational has a tendency to become credible. The root problem is that there is no quality control.

Getting back to video card reviews, I don't mind our products getting bad reviews when the review is done fairly and objectively and diligently. However, there are many reviews that are just shoddy journalism. They are not diligent in their investigation, they are biased in their views, and they don't attempt to show both sides of the story. My favorite example is that when we get beat by 3 fps (frames/second) we "get our clock cleaned", but when we lead by 3 fps in another benchmark, we "just barely beat" the competition. It's not so much the data that is inaccurate as the spin that is put on the data. We also go through great pains to request that reviewers benchmark certain modes that we feel are representative of actual game play - for example high-resolution with FSAA (full-scene-anti-aliasing) enabled, e.g. 1024x768 or 1280x1024. Yet many review still focus on 640x480 without FSAA. In general, I am very disappointed by how many reviews ignore FSAA or ignore the different types of FSAA. Even if a reviewer doesn't value FSAA himself/herself, they have a duty to report FSAA results to readers who do value FSAA, and trust me, there are many out there.

The primary job of a 3d video card in terms of 3d is to run games. Therefore, I think your emphasis on games is warranted. For example, instead of being concerned about whether a card supports a particular feature or not, e.g. hardware T&L, just review the games. The final result should be judged by the overall user experience, not whether a function is performed in hardware or software. If software can perform a function faster than hardware, more power to it! In addition, some reviews focus only on one game or one benchmark, which isn't very thorough. It wouldn't be bad if this was pointed out in the review, but most times this is simply ignored and conclusions are drawn based on a fairly poor sampling of data. Driver and card stability, image quality, and smoothness of game play should also be taken into account. As you state, the overall experience is what matters, individual game benchmarks are just one part of that.

Dave: Would you care to rant about anything? (yeah, Gary… this is your chance!! - Rev)

I think I already did in answering the above question .

Wavey: What is the M-Buffer !!!???

Good question! Let's just call it our next generation technology for now.

Reverend: Any last words?

Sure. 3dfx has gone through some very difficult times in the last year or so, but we have survived. Unfortunately, in the last four years there were dozens of other companies that didn't survive. We are now very focussed on the basics which made us great - providing the best 3d graphics for the PC and gaming, period. This involves many aspects - from great performance and price, to superior image quality and features that really make a difference to the game developer. I am very excited by the products we have under development and some of the business opportunities we have. This next year should be a great year for 3dfx!
 

Thanks, Gary! Lots of good (alluded) info in your answers (with some understandable "side-steppings"). We appreciate the time you took to answer them and look forward to the next "Ask Gary" monthly session.