Advertisers love numbers because they are a simple and straightforward way to convey the idea of improvement. For example, version 2.0 is always better than version 1.0, a clock speed of three gigahertz simply must be faster than two gigahertz, and four gigabytes of RAM are better than three gigabytes. Rarely will somebody challenge the universally-accepted truth that more is better.
Unfortunately, the real world is a lot more complex than the simple numbers suggest. Sometimes version 2.0 loses the elegant interface that made version 1.0 so compelling. Sometimes 3 GHz clock speeds are slower than 2 GHz if they are based on an inferior architecture. And sometimes, more RAM doesn't make a difference.
Graphics card manufacturers have been exploiting the amount of RAM as a marketing tool since the very beginning. Back in the day, you needed a certain amount of RAM on the graphics card to simply run a resolution like 1024x768. As time went on and 3D accelerators emerged, RAM on the graphics card was employed to store textures and allow for features like anti-aliasing (AA), post-processing, and normal mapping.
The focus of this article isn't to dig into the minutia of where your graphics card RAM is being used. Instead, we're more interested in looking at the tangible impact that different amounts of graphics card RAM will have on your gaming experience. Our goal is to let you know exactly what advantage, if any, you can expect from a graphics card that has more RAM on-board.
Having said that, there are a few important concepts we'll need to cover before this will make any sense, so let's get started.
RAM Bandwidth Versus Quantity
The biggest misconception out there is that more graphics RAM will increase gaming performance. But, for the most part, this is untrue. While the bandwidth of the graphics RAM might have a direct effect on gaming performance, the quantity of RAM doesn't have a direct impact on how fast a graphics card performs. The quantity of RAM does have an indirect impact on performance sometimes, and we'll talk about this in a little bit.
The bandwidth of graphics card RAM is affected by two main factors: the clock rate and interface width. Clock rate is measured in MHz, just like a CPU. The width of an interface is measured in bits, such as 128-bits wide. Without going into too much detail, you should know that 200 MHz memory should provide twice the bandwidth of 100 MHz memory, assuming everything else is equal. Similarly, a 128-bit memory bus should provide twice the bandwidth of a 64-bit bus.
It gets a little complicated because we're talking about two variables here: clock rates and bandwidth. For example, 200 MHz, 64-bit memory should offer approximately the same bandwidth as 100 MHz, 128-bit memory. There are other factors, such as memory latency. Technology also plays a role, as GDDR5 offers twice the throughput of DDR through GDDR4 at a given clock speed. But, for the most part, all you have to remember is that more bandwidth should have a direct impact on your gaming performance.
However, we won't look too hard at memory bandwidth today. Instead, the quantity of graphics card RAM is what we'll focus on, which is measured in megabytes (MB) or gigabytes (GB).
As we've said, the quantity of RAM that the graphics card employs doesn't have a direct impact on game performance, but it can have an indirect impact. Graphics card RAM will only negatively affect performance if there isn't enough to handle what a specific game title requires. The point is that all other factors being equal, a graphics card with 2GB (2,048 MB) should perform exactly the same as a graphics card with 512MB as long as the game's graphics memory requirements are below 512MB. If the game's settings and resolution require more than 512MB of graphics card RAM, the 512MB card will demonstrate a performance penalty compared to the 2GB card.
With this in mind, we're going to see exactly what happens when we test a number of popular games at different settings and resolutions with three different graphics cards armed with 512MB, 1GB, and 2GB of RAM. All of these models center on ATI's Radeon HD 4870 GPU, and we will make sure they are set to the same clock speeds so that the bandwidth and engine throughput are identical. Let's have a quick look at the cards.
Three Radeon HD 4870s: 512MB, 1GB, And 2GB
Powercolor AX4870 512MD5-G Radeon HD 4870 512MB
We'll begin with the 512MB card, a PowerColor Radeon HD 4870.
This card is overclocked from the factory with a GPU speed of 770 MHz compared to the reference 750 MHz, so we'll lower this to the reference speed for our testing to keep everything on an even keel. The GDDR5 memory runs at 900 MHz, which is the same as the standard Radeon HD 4870 reference speed. This card retails for about $135 on Newegg.
This XFX version of the Radeon HD 4870 comes with 1,024MB (1GB) of RAM, and runs at the reference 750 MHz core and 900 MHz GDDR5 memory speeds. This model costs about $145 on Newegg.
This Sapphire card is equipped with four times the memory of the 512MB card. While it sports the same 750 MHz core and 900 MHz GDDR5 memory speeds as the reference Radeon HD 4870, it comes with a unique aftermarket cooler. This model costs about $227 on Newegg.
Now that you've seen the contenders, let's have a closer look at the test system and move on to the benchmarks.
Microsoft Windows Vista Ultimate 64-bit 6.0.6001, SP1
DirectX version
DirectX 10
Graphics Drivers
Catalyst 9.9
Benchmark Configuration
3D Games
Crysis
Patch 1.2.1, DirectX 10, 64-bit executable, benchmark tool Test Set 1: High Quality, No AA Test Set 2: Very High Quality, No AA
Far Cry 2
DirectX 10, in-game benchmark Test Set 1: Very High Quality, No AA Test Set 2: Ultra High Quality, 4x AA
World In Conflict
Patch 1009, DirectX 10, timedemo Test 1: Very High Details, No AA/No AF Test 2: Very High Details 4x AA/16x AF
Left 4 Dead
Version 1.0.1.4, custom THG benchmark Test Set 1: Highest Settings, No AA, No AF Test Set 2: Highest Settings, 4x AA, 16x AF
Grand Theft Auto IV
Patch 1.0.2.0, in-game benchmark Test Set 1: Med, Med, Med, Med, High, 30,70,51,10 Test Set 2: High, Very High, Very High, Very High, High, 50,70,51,10
Game Benchmark: Crysis
To demonstrate a memory usage difference in Crysis, we're going to benchmark first with the High quality setting and next with the Very High quality setting. The Very High settings will use higher-resolution textures, increased shadow resolution, and advanced post-processing that might demonstrate a difference using three levels of texture memory:
First, the High settings:
Only at the high 2560x1600 resolution is the 512MB card experiencing a slight performance penalty. Let's see what happens when we up the ante to the Very High quality settings:
Once again, there is no difference in average frame rates until the 2560x1600 resolution, but the game is running so slow at this point that the results are purely theoretical. However, if we look closer we can see that the minimum frame rate is consistently lower on the 512MB card, so the increased amount of video RAM is providing a benefit.
Crysis remains one of the most performance-adverse PC games on the market. The fact that the 512MB card achieves average frame rates similar to the 2GB card at 1920x1200 suggests that Crysis is somewhat conservative with video RAM use for such an advanced graphics engine.
Game Benchmark: Far Cry 2
Far Cry 2 isn't as demanding as Crysis, so we're going to set details to their highest levels right off the bat (Ultra Quality setting). For our second run, we'll use those same settings, but add 4xAA in order to increase the memory requirements.
Here are the benchmarks for the first round, with no AA applied:
We are seeing a very slight disadvantage for the 512MB card at 1920x1200, which increases slightly at 2560x1600. As with Crysis, the 1GB card remains right on par with the 2GB card. Let's add AA to see if we can learn anything new:
Now we're getting somewhere. We see that 4xAA is causing a very notable performance hit with the 512MB card, which only increases with higher resolutions, while the card is effectively crippled at 2560x1600. However, performance isn't hit too hard until 1920x1200, and the game even remains playable at that resolution.
Game Benchmark: World In Conflict
With World in Conflict, we'll try the same strategy we used with Far Cry 2: we'll run once with the highest settings available, and then once again with 4xAA and 16x anisotropic filtering (AF) in order to increase texture memory usage.
Now for the first run results:
There appear to be no significant differences of which to speak. Now, let's add 4xAA and 16xAF:
It's a very close race. The 512MB card does seem to suffer a minor performance penalty at 2560x1900, but at 1920x1200 there's nothing significant to talk about. World in Conflict is likely more CPU-bound than anything.
Game Benchmark: Left 4 Dead
Our strategy up to this point seems to be working, so let's apply it to Left 4 Dead, with one run at highest details and the other with 4xAA and 16xAF applied.
First, we'll test with no AA and AF:
The results are close here. Left 4 Dead's benchmarking mode doesn't capture minimum frame rates, unfortunately. But by the looks of things it's so close that it probably doesn't matter. Now let's add AA and AF:
There is still no difference. The Left 4 Dead game engine is coded by Valve, a game developer with a solid reputation for getting great visuals to work on old hardware. You can bet that the Left 4 Dead engine is making very efficient use of texture memory to produce results like this at 2560x1600 with a 512MB graphic card, although we will admit it is based on the older Source engine.
Game Benchmark: Grand Theft Auto IV
We saved the most interesting benchmark for last: Grand Theft Auto IV.
It's the most interesting for our purposes because GTA IV actually reports the exact amount of graphics card RAM it needs for the graphical settings the user chooses, which is absolutely perfect for our purposes. It turns the theoretical into tangible when you can see the actual memory usage impact of choosing higher draw distances, texture resolution, shadow fidelity, and resolution.
We did have to use a little trick to force GTA into allowing us to use more video memory than what our 512MB and 1GB video cards offered. By making a text file called "commandline.txt" that contains the command "-norestrictions," the game allowed us to choose settings beyond the capabilities of the hardware.
In this case, we first ran settings that would use just under 500MB of memory at 2560x1600. We then ran a second run that used over 1GB of video memory to see what would happen to the 512MB and 1GB cards compared to their 2GB counterpart.
First, we'll try the benchmark using less than 500MB of video RAM:
There are no surprises as there is no difference. Now, we'll see what happens when we use over 1GB of texture RAM:
There's some contrast for you: the 512MB card is completely unable to run the game at this demanding setting. The 1GB card is able to run surprisingly close to the 2GB card up to 1920x1200, but it takes a notable hit at 2560x1600.
Conclusion
We've been through a lot of different games, settings, and data. What's the bottom line? How much graphics RAM is the right amount? We've seen enough to formulate some general guidelines.
For the gamer, there are three main factors that have the most influence on how much graphics RAM you'll need: resolution, visual quality detail settings, and AA. For the most part, 512MB of RAM seems sufficient to push one of these factors to the limit, and in most cases, it can handle two of them at once. But if you plan to maximize all three--the highest resolutions, visual quality settings, and AA--then more video RAM than 512MB is a good idea.
The primary consideration should probably be resolution, because this is a hardware limitation for many of us. Typically, 20" or smaller monitors will cap out at 1680x1050, so in this case, spending extra on 1GB of graphics RAM might be frivolous. Conversely, if your monitor is 21" or larger with a 1920x1200 native resolution, purchasing a graphics card with 1GB of RAM might be a wise investment for only a little extra money.
A 2GB card is probably only realistically useful for folks who push 1920x1200 resolutions and above, and who demand the highest visual quality settings in conjunction with AA. Since the price difference is substantial, a graphics card with 2GB of RAM is probably overkill for anyone who has a monitor with a native resolution lower than 1920x1200.
There are other factors to consider, too. Longevity might be a concern as games will inevitably require more and more graphics RAM as time goes on. More importantly, do not forget that the amount of RAM on a card does not indicate its relative performance--a 512MB Radeon HD 4870 or 768MB GeForce GTX 260 will always be faster than a 2GB Radeon HD 4670 or a 2GB GeForce 9600 GTS. The point is that the type of card is far more important than the amount of RAM onboard when it comes to raw performance. For a general idea about what cards are faster than others, take a look at our video card hierarchy chart.
In the final analysis, even 512MB of graphics RAM can provide excellent high-resolution service as long as the user is aware that lowering quality settings and AA will allow the card to perform at its peak potential. However, if you plan to use the highest visual quality settings in addition to AA, it's likely worth the extra money for a 1GB or even 2GB card.
