You know it is said that AGP increases speed greatly,
When AGP was first released this was actually true...
How AGP works is basically to by-pass the chipset.
AGP <-> CPU
... <-> RAM
PCI <-> ChipSet <-> RAM <-> CPU
This ment that you could only transfer as fast as the PCI bus, which 2.0 set to 33MHz (as this was the FSB for the Ram)
Basically this ment that access time between the Video Chip, and CPU had to go through the ChipSet and the Ram first.
What AGP allowed was the Graphics Card to access the Ram and CPU directly; this ment that basically communication speeds were increased, often to the boards FSB * AGP Bandwidth.
At the time this was 100MHz * 1X (33MHz) giving some impressive speed increases, as alot of 3D operations were done on the CPU.
As time progressed the Bus speed of the card was increased, this didn't actually really increase the access time; all it did was allow more data to be transferred at once. Which speeds up loading time, which for 8MB cards like there were back then this ment quite a bit faster when accessing data.
See the thing now is, GPU's have taken over almost all of the CPU 3D Tasks; So unless your trying to do something your GPU can't handle you will not notice a difference in rendering speed.
Also most cards have 128MB/256MB Ram, which means entire scenes of data can be loaded and cached so that less calls need to be made to the system ram and cpu allowing less speed lags.
So basically all PCI <-> AGP now means, is loading time and how smooth in-game loading is. As some may notice running cards capable of AGP4x on a 2x mobo in DirectX produces a 'jerky' effect whenever you enter a new area. This is because it is slowing down to grab the new data, there is no difference in actual rendering speed though this is a pure output bug of the system because the card doesn't recieve the call to draw.
Although PCI-Express seems impressive with it's Bi-Directional tranfers and such... it STILL is going through the chipset, and no doubt AGP-Express will follow which will once again by-pass the system chipset. But really at this point; it's a waste of time due to how the GPU is actually the most powerful processor for Graphics in your system.
A 250MHz GeforceFX 5200 is capable of outperforming a Pentium4 3.0GHz HTEE doing 3D Operations with ease.
Really at the end of the day with these cards it just comes down to what you can put in your system - rather than what technology is on it.
AGP 8x sounds impressive, but the only advantage it has over AGP 2x is the fact that it can shift around 8x the data in a second.
But when you only have 256MB oncard and 512MB on system; what the hell is the point in shifting 34GB/sec of information?
AthlonXP 2500+ | 256MB DDR PC2700 | GeForce FX 5200 44.04 | DirectX 9.0 | Audigy2 | Crystal Clean OS