The documentation made note of an 8X FSAA mode, how is this achieved and is it Multi-sampling?

The 6XS and 8X modes are a combination of Super-Sampling and Multi-sampling.

On GeForce4 there was a total of 4 Z units per pipeline, which enabled you to output 4 MSAA samples per pipe per clock, is this still the same for GeForce FX?

Yes it is.

Are they still the sort of ordered grid sampling schemes that we've seen on GeForce4 or does GeForce FX have any jittered / programmable grids?

The 6XS is a skewed grid, the 8X mode is an ordered grid.

There was a quote in the launch information about adaptive Anisotropic texture filtering, but that's slightly confusing since recent report have stated that GeForce4 already does adaptive texture filtering.

What we have is a new technology. We're not disclosing the algorithms that we use, but in the control panel we will have sliders. On one side of the slider will be 'Conservative' and on the other will be 'Agressive'. So if the user sets the slider to conservative then it will give the the standard filtering for whatever the user request, whether its Bilinear, Trilinear, 8X Aniso or whatever it is, it will do the full algorithm. As you move it towards aggressive we have a variety of adaptive algorithms that look at both the content of the texture and the content of the scene and will proactively choose different sampling patterns to maximise image quality whilst also maximising performance. So for benchmarking, if you want a pure apples to apples then set the sliders on conservative, but if you're a user and you're playing you favourite app you can experiment with those sliders and if you find you enjoy the quality with the slider on aggressive then you're going to get more performance out of it.

What degree of Anisotropic filtering does GeForce FX go to? Is it 16X?

We do up to 8X, but we calculate 8X to be 8 Trilinear samples.

There's been talk of extra overdraw removing routines / occlusion culling for GeForce FX, what more does it have over GeForce4?

We have a faster pixel culling rate. When we went through and enhanced the overall architecture we wanted to make sure it was tuned for the next generation of content and as such we upgraded the number of pixels it can reject based on Z comparison.

Did you not think about combining some of the work that 3dfx and GigaPixel have done with NVIDIA architecture? Did you consider going deferred rendering or even to 'semi-defer' and cache some geometry data, sort it and then render it?

The challenge with some of those other architectures is with today's level of performance, today's capabilities, and today's requirements for software compatibility no one has really solved the compatibility issues, the corner cases if you will, of those different techniques. Without talking about future products, because someday we may solve those corner cases, we still use a more traditional pipeline in this GPU because what we want is absolute, solid compatibility and stability.

We did acquire technology down the paths of those other architectures and we've got a lot of smart people looking at it. When the time is right to include some of those other ideas, we'll be there, but its not the time for that yet.

Does GeForce FX have any advancements over the 'Lightspeed Memory Architecture II' that GeForce4 featured?

In the past we've lumped our memory technologies under the name Lightspeed Memory Architecture, for this GPU we're really talking about Intellisample technology and we're lumping on some of the Antialiasing and the adaptive texturing in with the improvement in our memory architecture, which include for the first time from NVIDIA colour compression. So not only are we compressing Z, we're compressing colour.

Is that full time colour compression, i.e. does it operate when AA is not enabled?

It pays dividends mostly for the AA modes because those are the ones that are memory bandwidth bound. If you're just doing aliased rendering, with the bounds of our pipeline and memory you are generally not memory bound, but it can colour compress without AA.

GeForce FX was obviously supposed to be your fall product, so what was the cause of the delays?

There were a couple of reasons.

First of all we wanted to tie this technology to DX9 and 0.13µ process technology, to make sure we could deliver the heart and soul as we didn't want to compromise. 0.13µ took us longer to get implemented from our side than we would have liked. We had originally targeted this product for this holiday season, but what was more important to us than hitting that season was making sure that when the GPU was available that it truly created a new platform, so we didn't compromise it. We could have cut some corners here and there and got it out for the season but when we got it to market it was fundamentally a different architecture than other DX9 class products.

Finally, what would you say that GeForce FX has that R300 doesn't?

Unbelievable levels of programmability that will enable the developers and artist to deliver stunning visual experiences in real time.

So, there we have it, GeForce FX explained.

Much has been made of the 3dfx connection here, and if that doesn't come from any GigaPixel like overdraw reduction routines what influences were there? Its likely that all engineers had a hand in all elements of the architecture, but from a marketing perspective and some of the demo's shown, there is an immediate link with the 'Cinematic Computing' and the T-Buffer effects 3dfx were keen to talk about some time ago. However, what outwardly seems like one of the biggest departures from previous GeForce architectures is that of the Vertex Shading pipeline. Given NVIDIA's past moves in this area, and just incrementing the number of shader units in parallel, it was unexpected for them to implement what sounds like something similar to 3Dlabs Shader array.

There's still a few more things to tie up with GeForce FX, such as board variants in the range, but for now we await the benchmarks.

• If you wish to comment on this article then feel free to do so here.