Just as a CPU, a GPU needs short-term memory. As I told you in the beginning, the graphics card is like a small computer with a motherboard, RAM and some controllers to control it all. The RAM on a graphics card plays a big role. It is the GPU’s work memory, and the more there is, the more data the GPU can save and quickly access.
That is one of the main reasons why games run bad in a too high resolution. If the resolution is set to 1920 * 1080 it gives 2 MP, and all these pictures has to be stored in the short-term memory. If there isn’t enough short-term memory to store the 2 MP data, it goes wrong. Most common, the graphics card stalls complete, and your game freezes or gets completely unplayable.
Another thing that your graphics card memory doesn’t like is what you might know as AA, MSAA or CSAA. That is short for anti-aliasing. That means that the edges in your game are smoothened. What AA does, is that is upscale the game’s resolution. 4 x AA in 1920 * 1080 gives 4 x 2 MP = 8 MP. That is four times more megapixels to control, than if AA was turned off. In games like Battlefield 3, MSAA(AA) can kill even the most powerful graphics card. That is caused by the very advanced textures of the game, and the general game world and effects.
A very common misunderstanding is how multiple GPUs control the short-term memory. Too many thinks that 2 4 GB graphics cards give 8 GB total. Unfortunately, that is not reality. Every GPU needs its own short-term memory to save the data in. Therefore, a graphics card with 2 x 4 GB RAM gives 4 GB RAM to every GPU. Therefore you have a 4 GB graphics card only.
There are more types of RAM for graphics cards; although very few are still in production today. The most common types of RAM for graphics cards are the following: DDR3, GDDR3 and GDDR5. You may have noticed that GDDR4 is missing, and there is a very good reason for that. Exactly this generation you chose to skip. That is because of the experiences the producers got with the AMD (or ATI as they were called then) Radeon HD3870, which was the only wide distributed graphics card that used GDDR4. They were very silly. GDDR4 chips didn’t clock very high, they used too much power and generated extremely much heat. All of this meant the fast and silent death of GDDR4.
On the other hand GDDR3 is extensively used today. The speed of GDDR3 is not as high as GDDR5, but it still has more speed than set of normal DDR3 RAM. GDDR3 is used in low end graphics cards, and in mid and low end graphics cards in laptops. The advantage of GDDR3 is the low generation of heat and the cheap production prices.
GDDR5 is the standard of modern mid and high end graphics cards. The advantage of GDDR5 is that they clock completely crazy. GDDR5 RAM will easily run at 6 GHz, a speed DDR3 RAM can only dream of. The high speed makes it possible for the GPU to quickly move around the data. Besides that, GDDR5 is power saving, has low generation of heat and is relatively cheap to produce.
The last thing I will talk about is the bandwidth. That is the space the RAM of the GPU gets to play on. A graphics card with a bandwidth to the RAM of 128 bits will be half the speed of the same graphics card with a 256 bit bandwidth. This is of course only if they have the same RAM. Many smaller graphics cards are limited by their bandwidth. That one e.g. seen on nVidia’s GTX460. That graphics card was made in three versions. One with 768 MB RAM and 192 bit bandwidth and then a 1024 MB model with 256 bit bandwith.
It quickly showed up that the model with 256 bit bandwidth was appreciably faster than the 192 bit model. The reason is to be found in that the bandwidth, as mentioned, is the motorway between the GPU and the BUS that controls the memory. The less bandwidth, the less data can pass.