The Video RAM

In the early days of PCs the amount of information displayed was much, much less than today. A screen of monochrome text, for example, needs only about 2 KB of space. Special parts of the upper memory area (UMA) were dedicated to holding this video data. The processor would compute what needed to be displayed, would put it into this area, and then the video card would read it and display it through the RAMDAC.
As the need for video memory increased into the megabyte range and more, it began to make more sense to put the memory on the video card itself. In fact, to preserve existing PC design limitations, it was necessary (there simply isn't any more space in the UMA to hold the bigger screen images). The memory that holds the video image is sometimes called the frame buffer. A big advantage of having the memory on the video card is that it can be customized to the task at hand for greater efficiency, instead of using regular system RAM. The memory on the video card comes in many different sizes and flavors, and new technologies to improve performance are being invented all the time.
More recent video cards also use the video memory for computation work. In particular, cards that support 3D functions such as filtering need to use memory for this work. Different video cards use different methods to decide how to allocate memory for these functions: some will use whatever memory is "left over" after the necessary space for the frame buffer has been set aside, while others will pre-allocate the calculation memory and then limit the size of the frame buffer accordingly.
Since some cards use their video memory for both the frame buffer and additional calculations, they may have more than one type of video memory. For example, the dual porting advantages of VRAM are of course very helpful for the frame buffer, but of no value in memory being used internally for calculations. So, since VRAM costs more than single-ported memory technologies, manufacturers put some VRAM on the card for the frame buffer, and some "regular" memory on it for other purposes.
There are also two situations under which the video card can use the main system RAM as well. First, some discount PCs are designed to use part of the main system memory for their frame buffer, saving cost but leading to poor performance. Second, newer AGP video systems can use the main system memory for doing 3D processing or other work.

It is such an important component of the video card, and indirectly the entire PC, that several new memory technologies have been created specifically for it. The goal: to improve the speed with which information can be pumped into and out of the video memory, to keep system performance high as the video system tries to do more and more. So, it is important to take a detailed look at the various memory technologies now being used on video cards. They are:

Standard (Fast Page Mode) DRAM Extended Data Out (EDO) DRAM Video RAM (VRAM)
 
Window RAM (WRAM) Multibank DRAM (MDRAM) Synchronous Graphics RAM (SGRAM)
 
Synchronous DRAM (SDRAM) Double Data Rate memories (DDR RAM)



Fast Page Mode DRAM:

The oldest technology used in video card memory, fast page mode (FPM) memory is now considered "standard" DRAM as it has the fewest performance-enhancing capabilities of the different types of memory on the market. FPM DRAM is a technology used primarily for main system memories (even there, it is now considered a poor performer) and is not really well-suited for high-performance video applications.
FPM is the least expensive type of memory available for video, and is used today mostly on low-end or generic cards (as well as older cards of course). For many applications they can be quite satisfactory; however, they reach their limits quickly when trying to use high resolution modes, especially in true color.
The limitations of standard DRAM are due to two primary effects: it is single ported (which means it can only do one access at a time) and it runs at a relatively low speed and access width. Newer technologies improve performance by dual porting the memory (VRAM), increasing the bandwidth of the memory (SGRAM, MDRAM) or both (WRAM).
top
 

Extended Data Out DRAM:

EDO DRAM is the same as standard FPM DRAM except for a slight modification in the access cycle that gives it a small performance boost. With EDO DRAM, one read to memory can begin before the last one has completely finished; this yields a raw speed improvement of between 5 and 20 percent, depending on whom you ask.
Originally used only for main system memory, EDO DRAM became more popular on video cards because it provides slightly improved performance over standard DRAM at the same cost. EDO is still, however, a low-cost and low-performance solution compared to other types of video memory, and is not used on today's cards. (5-20% faster than "slow" is still "slow" in the video world.)
top
 

Video RAM:

The traditional, standard DRAM used for video cards typically does not have enough bandwidth to handle the demands of running a card at high resolution and color depths, with acceptable refresh rates. The main reason why is the two competing access factors for the video memory: the processor writing new information to the memory, and the RAMDAC reading it many times per second in order to send video signals to the monitor.
To address this fundamental limitation, a new type of memory was created called video RAM or VRAM. As the name implies, this memory is specifically tailored for use in video systems. The fundamental difference between VRAM and standard DRAM is that VRAM is dual-ported. This means that it has two access paths, and can be written to and read from simultaneously. The advantages of this are of course enormous given what the video card does: many times per second a new screen image is calculated and written to the memory, and many times per second this memory is read and sent to the monitor. Dual-porting allows these operations to occur without bumping into each other.

Note: Don't confuse VRAM with the generic term "video RAM" or "video memory", which just refer to the memory in the video subsystem in general.

VRAM provides substantially more bandwidth than either standard DRAM or EDO DRAM; double in many cases. It is more suited for use in systems requiring high resolution and color depth displays. The only reason that it hasn't replaced standard DRAM entirely is of course: cost. VRAM is more complex and requires more silicon per bit than standard DRAM, which makes it cost more.
top
 

Window RAM:

Window RAM or WRAM is a modification of regular VRAM that both improves performance and reduces cost on a bit-for-bit basis. Designed specifically for use in graphics cards, WRAM is also dual-ported but has about 25% more bandwidth than VRAM (since it uses bigger memory blocks), and also incorporates additional features to allow for higher performance memory transfers for commonly used graphical operations such as text drawing and block fills. Furthermore, WRAM is less expensive than VRAM to manufacture (although still more expensive than DRAM).
WRAM is suitable for use in high-end graphics cards and was first made popular by Matrox's Millennium series.

Note: Window RAM has nothing to do with Microsoft Windows.
top
 

Multibank DRAM (MDRAM):

A type of memory that attempts to address two problems with conventional video memory, Multibank DRAM or MDRAM was invented by MoSys specifically for use in graphics cards. MDRAM differs substantially in design from other types of video memory. Previous memory designs used a single monolithic "block" of memory for the frame buffer. MDRAM breaks its memory up into multiple 32 KB banks that can be accessed independently. This provides the following advantages:
MDRAM is also cost-effective to manufacture compared to VRAM and is efficiently organized to reduce waste. MDRAM is suitable for use in high-end applications and is becoming popular due to its performance-enhancing and cost- reducing features.
top
 

Synchronous Dynamic RAM (SDRAM):

Kind of memory nowadays freqently used on 3D video cards, due to it's relatively low cost. They increase the performance of previous EDO DRAM because they are synchronous with a clock signal, whose frequency increases in each generation of boards. In this technology the data bus is written once per clock cycle, at each rising edge. The main limit is the first access latency, i.e. the time needed to find the first data in a sequential vector. Although slower than WRAM, SDRAM technology based memories are massively used, since their low cost allow producers to create boards with big quantities of memory, operating at very high clock frequency.
top
 

Synchronous Graphics RAM (SGRAM):

A relatively newer RAM technology, Synchronous Graphics RAM or SGRAM tackles the poor performance of regular DRAM by increasing greatly the speed at which memory transfers take place. SGRAM also incorporates specific performance-enhancing features designed to work with acceleration features built into video cards, to greatly improve overall video processing speed: in fact, the reset of certain portions of memory (like the frame buffer or the Z-buffer) is performed very quicly, while in SDRAM memory it takes some clock cycles. SGRAM is still single-ported, unlike VRAM or WRAM, but offers performance that is much closer to VRAM than DRAM due to its advanced design.
top
 

Double Data Rate SDRAM-SGRAM (DDR SDRAM-SGRAM):

This is the newest techology available: it is based on the same principles of SDRAM and SGRAM memory, with a clock signal which temporizes all the operations, but instead of writing the data bus on the rising edge of the clock, the DDR memories uses both the rising and falling edges of the clock signal. This allows to double the memory bandwidth without changing the frequency of the clock signal, which is more difficult because of EM interferences, although it rises slightly latencies, since a 2 cycle latency makes the system lose 4 clock edges, instead of 2 edges in regular SDRAM. DDR RAM are the fastest memories on the market; introduced on the high-end segment a few years ago, they are nowadays used on almost all 3D video cards, except for low-end ones.
top