8. Creative labs was very honest about their upcoming product in an interview (taken from http://www.3dhardware.net/reviews/cl_geforce/) , 2 extracts from it :

" With most of the demos, we saw a rather constant frame rate of 30FPS - with the default features on and at default polygon rates. In the fire truck demo, which contains 100k polygons, the GeForce256 tugged along at roughly 15FPS, not the most impressive score albeit anything else on the market would do that demo much slower."

Ok, some simple math based on that info reveal : 100k x 15fps = 1.5M ... Hmm if this demo is an example of a real world situation (which it isn't since there is zero AI, zero physics, zero sound, ... ) that its very very far away from your theoretical numbers... actually a factor 10?! Are those numbers incorrect or is it just a poorly programmed sample?

I suspect that the demo now runs much faster, but to be honest, I haven't seen the firetruck demo. Just run the stuff for yourself and see how well you think it performs.

9."This time around, NVIDIA have not supplied manufacturers with the solution on how to overclock the GeForce256 as they did with the TNT2. What's the reason for this? They don't want the same thing to happen with the GeForce256 as with the TNT2, meaning that many companies shipped TNT2 boards at ridiculous clock speeds while just getting a small yield.... One thing is certain though, the GeForce256 runs very hot, and you barely survive with the on-board fan and heatsink. So if you want to try and overclock a GeForce256 card make sure you have efficient and powerful cooling."

Now this kind of sends me the message: don't count on overclocking the GeForce256... it almost introduces the scare for instability on some systems... Any comments about this (why, how, ... )?

To be honest, I don't think overclocking is a good idea. However, I think that people will try to do it, and many of them will be pleased to see higher clock speeds work well.

10. The tree demo is nice but it also reveals an upcoming problem: opaque overdraw... Traditional renderers like the GeForce256 (?) do "useless" work: they draw invisible polygons... in the case of the tree demo many leaves are hiding each other causing a drop in performance due to the fill-rate limit. In the future this will only get worse: games will get more and more details and thus more and more object and thus more and more opaque overdraw. How will you tackle that problem?

There are technologies like scene graphs that perform several functions to get around this problem. A good scene graph library can, for example, cull away geometry that is definitely behind a non-blended closer object. In addition, the scene graph can sort the data so that front-most non-blended objects are drawn first, which helps on all fill-bound renderings to reduce the amount of color- and depth-buffer overdraw. Scene graphs are beyond the scope of what I want to talk about here today, but you can find lots of info about them on the web.

11. The theoretical peak of the T&L hardware is 15 million triangles, are we assuming full 100% use of strips and fans here? If that test uses strips you really only have 15 million "vertices", or thus 5 million "triangles" in reality (if "no" strips/fans are used). Correct, incorrect?

The 15M number is 15M vertices. This is the number people should really focus on when talking about transform and lighting limits, not the number of triangles, since, as you point out, poorly designed datasets will inefficiently use that budget of 15M vertices. Well designed datasets will create many more polygons from that same pool of vertices. 

12. The developer support guys told me you have a true benchmark that shows your hardware doing 11 million triangles per second. Can you describe that benchmark/demo and maybe give us some screenshots of it?

Sure. It's a small app written by a guy named Steve who works in my group. Basically it draws a bunch of rotating spheres, each of which is made of 125,000 vertices. It's actually hit higher numbers than 11M vertices per second on some new cutting edge systems, where AGP4X is available to it. I think NVIDIA will put that app (called SphereMark now, I think) on our web site fairly soon, hopefully including source code and a built app.

13. I have seen NVIDIA saying: "Right now, there is nothing that beats GeForce256 in fill rate." Too often... while this might be true "today" it will not be true "tomorrow"... what if competitors beat your GeForce256 at 1024x768 or above simply because they have more fill-rate? 1024x768 is the sweet point today can you maintain 60fps at that resolution? I ask since I saw Quake3test running on the GeForce256 on the Creative Labs stand at ECTS with the online frame counter and that counter hovers between 100 (nice) and 30 (boo). In general frame-rate jumps are seen as very unpleasant... I can only assume that those drops are due to the fill-rate since the 10.000 polygons per frame in Q3test should be no real issue.

First of all, unlevel frame rates are not a problem with the graphics card, in my opinion. The way to fix that is within an app, by balancing the number of vertices per frame, pixels drawn per frame, etc. Those are hard problems to solve, but I think that we will see those kinds of changes come in the upcoming couple of years. Now, as for people feeling that GeForce's fill rate is low compared with specs of new parts some of our competitors have announced, well, the best I can say is, GeForce 256 is shipping, those others are not. We'll have to see how well they do with a real chip, instead of with a spec on a piece of paper.

14. Elsa has reported stability problems with the reference design, they talk about lowering the voltage etc... Any comments? Will we get power supply problems like with TNT1 and the Asus Motherboards?

Don't know anything about it. I'm a software guy. I do know that there's many motherboards out there that do not supply the spec-required current to the AGP slot, and my guess is that those motherboards will have problems with any next-generation AGP card like the GeForce or even TNT2.

15. Lightmaps and dual texturing allow great effects, T&L now introduces new types of lights... will vertex lighting replace lightmaps? Can you achieve the same effects and quality? Or should they exist next to each other and be used depending on the situation? One problem with vertex lighting is that it doesn't take into account shadows... 

I'm not sure how this will all pan out. Look at demos like Whole Experience (http://www.wxp3d.com), which is a D3D title that uses GeForce's lighting for its lighting effects. From the screenshots you can't really see how amazingly cool this demo really is. I think there's a lot that game developers will learn about how to use real OpenGL or Direct3D lighting, now that it's available to them. This will be the same kind of learning curve that they all hit when the first games to use lightmaps came out, like Quake1. Over time, though, I think people should expect amazing visual effects that are created using the lighting in the 3D APIs.