Graphics Processing Unit (GPU)
A high-performance 3D processor that integrates the entire 3D pipeline (transformation, lighting, setup, and rendering). A GPU offloads all 3D calculations from the CPU, freeing the CPU for other functions such as physics and artificial intelligence

The main new features of the GeForce is the Hardware Transformation and lightning unit. NVIDIA calls it a GPU - a Geometry Processing Unit. That´s also the Reason why the chip is called GeForce: 'Ge' for Geometry.

it's funny how NVIDIA says GRAPHICS considering they make the chip i think i'd lean towards them

The Big Picture

Here is a bit of general background information on OpenGL and 3D
hardware acceleration contributed by Steve Baker
(sbaker@link.com)

Updated by Curt Olson (9/25/2000)

When you are drawing graphics in 3D, there are generally a hierarchy
of things to do:

1. Stuff you do per-frame (like reading the mouse, doing flight
   dynamics)
   
   
2. Stuff you do per-object (like coarse culling, level-of-detail)
   
   
3. Stuff you do per-polygon or per-vertex (like
   rotate/translate/clip/illuminate)
   
   
4. Stuff you do per-pixel (shading, texturing, Z-buffering,
   alpha-blend)
   

On a $1M full-scale flight simulator visual system, you do step (1) in
the main CPU, and the hardware takes care of (2), (3) and (4)

On a $100k SGI workstation, you do (1) and (2) and the hardware takes
care of (3) and (4)

On a $100 PC 3D card, you (or your OpenGL library software - which
runs on the main CPU) do (1), (2) and (3) and the hardware takes care
of (4).

On a machine without 3D hardware, the main CPU has to do everything.

The amount of work to do each of these operations goes up by one or
two orders of magnitude at each step. One eyepoint, perhaps a hundred
objects, tens of polygons per object, hundreds of pixels per polygon.

Hence, putting step (4) into hardware is vital - you could easily need
to draw a million pixels for each time you read the mouse. Putting
step (3) into hardware is also very nice - cards like the nVidia
GeForce are now doing this.

So, if you put a cheap accelerated 3D card onto a slow-ish PC, you
still get all the benefits of not doing the per-pixel math in software
- so your frame rate will speed up. But since that per-pixel stuff now
goes very fast, you are probably limited by the speed that your old
clunker can do step (3) - the per-vertex math.

A card like the GeForce which can do step (3) will bring you even
greater performance benefits on a slow-ish PC.

It has been suggested that even a 400MHz Pentium II couldn't feed
vertices to a Voodoo-2 card faster than it could process them. That
was for a 'typical' scene though. If your polygons are very large (in
terms of screen area) then the amount of per-pixel math will increase
up to the point where even a slow PC would end up waiting for the 3D
card to complete the rendering of one polygon before it can accept the
next.

If your polygons are very small then the 3D card may not give you much
benefit. There is a certain amount of overhead in sending each
polygon to the hardware from the main CPU - and that cost could
ultimately be more than the cost of drawing the pixels in software if
the polygons only cover a few pixels each.

If your 3D card doesn't provide a speedup then make sure you have a
proper OpenGL driver installed for that card - some cards only support
Direct3D which FGFS is unable to interface to. Some 3D cards can't
function if you set the display resolution too large or the pixel
depth to something that card won't support. Some OpenGL drivers
silently drop back to software-only rendering under those
circumstances.

In summary: Will a 3D card speed up FGFS on a standard PC? Yes -
immensely.

Ok that was a very obfusicated way of explaining it :). Sorta like spaghetti-code (using goto in C should be punishable by some sort of pain!).