Video memory size
GPU is not the only factor affecting graphics card performance. Any processor needs a large volume of dedicated memory with high bandwidth for writing and reading textures, vertices, buffer data, etc. Even a powerful GPU can be bottlenecked by slow and insufficient video memory, therefore the latter (as well as external memory controllers) is very important.
Memory chips are usually located on a card around a GPU. Some products already contain 12 memory chips. In some cases there is no even passive cooling for them. But the most popular solution is a common cooler, and sometimes memory heat sinks. That's how memory chips look like in GeForce 8800 GTS with its cooling system removed:
Modern graphics cards have memory ranging from 512MB to 2048MB. Low-End and Mid-End products usually have 512MB of memory. High-End cards are equipped with 1024MB and more (with some exceptions). The more memory is installed on a graphics card, the more data (textures, vertices, and buffers) it can store, so there is no need to use slower access to PC RAM. Textures occupy lion's share of memory. Let's examine screenshots taken in Call of Duty 2 with different texture quality settings:
Call of Duty 2 adjusts texture quality automatically based on memory available to textures (video memory minus frame buffer). For example, the Extra mode is enabled automatically for graphics cards with 320-1024MB; High or Normal -- for 256MB cards, depending on resolutions and AA levels; Low -- for 128MB. Even if you manually set maximum quality, a graphics card with insufficient memory will have to use some PC RAM, which will significantly reduce performance and gameplay comfort.
Below is the table with average video memory usage for some outdated 3D games at maximum quality settings, 1024x768, AA 4x (where possible). The information is taken from our technological game reviews and other articles.
As you can see, even outdated games have high video memory requirements. There are already no games, which can run well with just 256MB. They use up to 600-700MB. It does not mean that all game resources should be uploaded to video memory of a graphics card. Games often let the API manage resources, especially as Direct3D 10 uses video memory virtualization. Nevertheless, there is an apparent tendency to increase requirements to video memory size on graphics cards.
The minimum required video memory size is 320MB, but it's available only for GeForce 8800 GTS. The optimal memory size has become 512MB, especially as memory requirements will change in higher resolutions and future games, so you should choose a graphics card that has some memory reserve for the future. Even now some games require 768-1024MB of video memory.
Sometimes memory size published on a product box does not match the total size of installed chips. It used to happen with Low-End graphics cards that worked with a part of system memory, employing NVIDIA TurboCache or ATI HyperMemory technologies:
For marketing reasons, specifications of such graphics cards mentioned total amount of memory, including shared RAM (e.g., 128MB) instead of physical 16-32MB. You should be careful about that and read our reviews not to be cheated in future. But actually you don't have to worry, because such products are obsolete now, being pressed away by integrated chipsets.
Now that we've sorted out video memory sizes, we have to remember that it's not only size that matters. Inexpensive graphics cards are sometimes equipped with more memory for the sake of impressive specifications (and better sales). However in a Low-End graphics card such memory size makes no sense, because it still cannot provide acceptable framerate at texture- and geometry-intensive settings.
Salespersons sometimes boast about memory size like it's the main graphics card feature. This often confuses inexperienced users unfamiliar with the real state of things. Let's compare performance of cards with different memory sizes on the example of two similar NVIDIA GeForce 8800 GTS solutions that differ only in terms of video memory size. The first card has 320MB of memory, and the second card has 640MB. Any shop assistant will tell you that the second graphics card is much faster, except for situations when the store has only 320MB products in stock (we do not take into account those rare honest and competent shop assistants). But is that true? Let's find out:
As you can see, video memory size has little effect on performance in most modes. Performance drops only at very high resolutions and maximum quality settings, when a 320MB card is noticeably outperformed by its more expensive modification. On the whole, we can say that video memory size does not affect performance that much, so you shouldn't compare products by this feature only.
Note that we speak of memory sizes above 256MB for Low-End cards and 512MB for the rest! In this case, memory volume rarely affects performance. But if we compare a 256MB card and a 512MB card, the conclusions would be different. You see, game developers try to fit game resources into video memory of popular cards. Today it's 512MB already. Although high resolutions and maximum graphics quality may require 1 GB. A smaller amount of memory in your graphics card would cause modern games to either run slowly (because of PC RAM usage), or prevent you from playing at maximum quality, or both.
Game developers are constantly increasing the estimated video memory size. While several years ago a graphics card could do well with 256MB of memory, today there are projects requiring over 512MB. They are few, but they exist, for example, GTA IV. So if a couple of graphics cards differ in memory sizes (being identical in clock rate and bus width) and are priced similarly, you should buy the one that has more memory. However, do not "chase numbers" -- remember that no Low-End product can benefit from 2 GB of memory. This memory volume is currently excessive. Thus we can state that today 512MB is optimal for Low-End and Lower Mid-End cards; and 1024MB is currently optimal for Upper Mid-End and High-End products.
Memory bandwidth in more detail
Another important feature we have already mentioned is memory bandwidth. It depends on memory clock rate and memory bus width. This feature determines the amount of data, which can theoretically be transferred to/from memory per unit time. In other words, it is speed at which a graphics core can write and read data to/from video memory. So, the faster a card reads texture and geometry data and writes computed pixels, the higher performance is.
Peak memory bandwidth can be easily calculated -- it's a product of effective memory clock rate and the amount of data transferred per cycle (memory bus width). For example, memory bandwidth of GeForce GTX 285 that has a 512-bit bus and memory clock rate of 1242(2484) MHz would be:
1242 MHz * 2 (double data rate) * 64 bytes (converting bus width into bytes) ~ 155 GB/s
While memory clock rate is usually written on boxes and in specs, memory bus width is not always specified. Therefore you should pay special attention to it. Most modern graphics cards use 128-bit or 256-bit memory buses. But some inexpensive cards might be equipped with just 64-bit buses, and this would not be advertised. It's cheaper for a vendor and allows to scale graphics card performance in a series. Two identical graphics cards with the same clock rates but different memory bus width will demonstrate different performance. The one with the higher memory bandwidth will be able to process more data than the card with the narrower memory bus.
So, when you choose between graphics cards with different memory sizes, you should always pay attention to clock rates, bus width, and prices! For example, if there are two Low-End cards -- with 256MB and 512MB memory -- and the latter is much more expensive, there is no point in buying the 512MB product, because such a Low-End graphics card will not gain much performance from this amount of memory. Another example is graphics cards with different memory sizes and clock rates. In this situation your choice will depend on the graphics card level and clock rate difference. And you shouldn't forget about prices, of course.
For example, if you must choose between a 1GB overclocked High-End solution and a similar, but more expensive 2GB card with reference clock rates, the former product would be more expedient at the moment, because it would provide higher performance nearly in all modes except for the highest resolutions. Higher frequency plays a much more important role than more memory in most graphics cards' modes. The doubled memory size affects rendering speed only at the highest resolutions.
Write a comment below. No registration needed!