GPU

AMD/ATI vs nVidia

Historically, ATI has always bent over backwards for Microsoft and has some advantages over nVidia w/ DirectX. Similarly, nVidia has always paid attention to the OpenGL committee & working with them at ATI's disadvantage. Granted ATI is FAR worse at OpenGL than nVidia was at DirectX, but there was still some real-world difference.

nVidia soaked up the ever-wonderful 3dfx so long ago (may it rest in peace), and has been on the ball all the way up until the GeForce 5. This when ATI, a previous wannabe, overtook the industry by suprise with the wonderfully crafted Radeon 9600 & 9700. It beat on nVidia in all things when it came to games. Sure nVidia tried to comeback, but never really could. Prior to the Radeon 9xxxx, their Radeon 8xxxx wasn't all that bad... a good competitor really to nVidia and kept them honest. But keeping another company honest in vying for some competition, and truly crushing them is a totally different story. ATI gave nVidia "a hurt'n". But alas, it wasn't meant to be... since nVidia's 7xxx series, it has mopped ATI left and right (that'd be 3x generations now).

For a very brief time, everyone thought that ATI & nVidia would be in a dead heat with ATI's strength in DirectX and nVidia's strength in OpenGL. Unfortunately for ATI, when you combine some really good production runs, great pricing, and a killer driver support team over at nVidia (they greatly improved DirectX performance)... ATI really couldn't compete with a much more complicated card to produce, higher pricing, and historically awful driver support team. Now, in defense of ATI, they have truly turned their driver development around with the Radeon 9xxx as it is very evident in the Catalyst drivers. But they just can't get OpenGL down for the life of them, they're just too far behind... and it's gonna take some time to produce some high performance drivers.

Balance of Power = Bottlenecks

CPU's are infact underpowered and is a system's usual bottleneck in relation to the GPU. Unfortunately, the GPU does demand a great overhead over the CPU... fortunately though, DirectX10 helps eliminate some of the GPU-CPU dependencies & frees up both the CPU & GPU to do their thing.

DirectX vs OpenGL

DirectX10 hardware is required to take full advantage of GPU performance increases provided by Windows Vista. DX10 will now virtualize & manage GPU memory, and inherently supercedes OpenGL in every respect. NOT that it's superior, but OpenGL & DirectX of old BOTH are used to directly access the hardware. Well......... if DX10 is now maintaining that control from boot to shutdown, that means OpenGL no longer has the direct access they BOTH are accustomed to. All OpenGL requests must now go through DX10.

DirectX is used by prob just under 50% of the games most played with OpenGL accounting for the other half in addition to workstation programs such as 3D Studio, AutoCad, & Pro/E. ID Software in particular and FPS overlord John Carmack are stark supporters of OpenGL, and consider it to be superior due to closer integration with the hardware... whereas Microsoft uses DirectX to add layers between the hardware & O/S so it's easier to develop for (one of life's lil balancing acts). Regardless of where you stand... no self-respecting engineer would ever put an ATI card in their workstation unless they were only doing wireframe on a cost basis. In workstation graphics, nVidia is undisputed king.

SLI / Crossfire

DO NOT BUY SLI OR CROSSFIRE SETUPS unless you're rich and have got money to blow. Yes, they most definately can provide better performance than single-card, but that performance is not garaunteed and the game must be supported in drivers. Also, you've gone past the point of diminishing returns... (i.e. Better to have a single 7950GT than two 7600GT's in SLI). Single-card setups are still the best for anyone but those with too-much discretionary funds (who would do quad-gpu anyways with two 7950GX2's).

Memory

The memory on a video card is for storing the textures used in games. All 256MB means is that U can store that many textures in memory w/out compression. However, the performance of a game does NOT depend on memory, but rather the speed of the GPU on the video card itself. This is why you ALWAYS choose the 'GT' version of nVidia cards since the non-GT cards will have to compress those textures anyways just so it can read and write textures into memory in a semi-speedy manor, because it's SLOWER. The memory is slower, it accesses the memory slower, and it renders the frames in the game SLOWER. If you have to choose, GPU Speed > Memory Capacity.