What exactly does it mean for the GeForce3 to be 'programmable'?

[Kishkumen]

This little question has been tugging on my mind for a little while. From what I've read, the GeForce's GPU is supposed to be programmable like an Intel or AMD cpu. However, what does this mean from a programmer's point of view? For example, the Intel and AMD cpu's share the X86 intstruction set which makes them both compatible with most software written. However, how does it work when you want to "program" the GPU of a video card? Does it have its own instruction set that has to be compiled for each game? The way I thought it worked traditionally, is that the game wants to draw a line, so it calls the directX instuction to draw a line, directX tells the video driver to draw a line and the video driver sends it along to the right place in hardware. But say you want each pixel in that line to sing and dance like NVidia says it can, so what happens? I am guessing that you write the code to change each pixel in directX8 and it tells the driver to do this and the driver either says "No I can't do this, send it the the CPU or it says, Yes, I'm a GeForce3 or Radeon2 and I can do it" Then the driver is responsible for taking the DirectX8 instruction and interpreting it into its own GPU instruction set. At least this is how I hope it is or are game developers going to have to choose between supporting GeForce3 vs Radeon2 vs increased cost and production time to support both?

Thanks for any insight any of you have.

Kishkumen

 

[Dark4ng3l]

The static t&l of geforce's has to be programmed for. it will not help with games witch dont support it. It can only do a certain amount of operations and are limited in the number they can do. With the GF3 you can use the t&l horsepower to do witchever feature you want while forgeting those you dont use making it more versatile to what you want in your game and how you achieve it.

 

[Kishkumen]

OK, so if I understand right what you are saying, then as a game programmer, I've got this really cool effect that I want to do. Normally, I would just send it down the pipeline and let it be rendered normally thus bogging down my frame rate since it takes so many clock cycles to compute, but since I've got a GeForce 3 then I can apply the special horse power of the T&L unit to this computation instead, bypass the rendering pipeline and keep my fps high...sounds pretty cool and alot better than how I was thinking it worked. I still wonder though, why the GeForce3 was originally so dependent upon DirectX8 being released upon its launch. Maybe I will have to go digging around NVidia and Microsoft to find out.