If you’ve ever walked into a conversation revolving around 3D card specifications and felt lost, then this is the article for you. This should also be good timing for those looking to pick up a new 3D card for Christmas.

There are a number of features to be considered on a 3D card when judging what sort of performance to expect. Throughout the article I’ll make reference to the Radeon HD 6870 which is a nice new card and a good example of a high-end mid-range product – something most gamers in the market for a new GPU would consider buying.

Overview

Because GPUs can’t operate independently of the rest of the PC, a quick overview of how it all fits together is necessary. When a game is launched, the data is loaded from the harddrive (HDD) into system RAM. The CPU can process data much faster than an HDD can serve it, hence the need for fast RAM. This is one of the first potential performance bottlenecks one might encounter. The CPU will then decide what data needs to be sent to the GPU for specific graphics processing tasks. This data is sent via the graphics BUS, typically PCI Express x16 these days.

Video RAM 

The amount of memory available on the 3D card is extremely important. Once your CPU has decided what data needs to be sent to the GPU, it does just that. The VRAM is where this data is stored while the GPU does its rendering thing. The more memory available on the 3D card the better, since it won’t have to constantly dump existing data and request more from the CPU. The HD 6870 has a cool 1024MB of VRAM, which is the ‘I’m serious about gaming’ standard these days.

Memory Bus and Memory Core

This is probably the least advertised element of 3D cards, and perhaps one of the most important. The Bus is the electrical pipeline along which data is sent to and from the VRAM and GPU. The width of the memory Bus indicates how much data can be sent at once. Most high-end cards should have a Bus of at least 256 bits wide, as is the case with the HD 6870.

Memory clock speed is also important (and another number often skipped by the marketing types) as this dictates how quickly the memory can send and receive each 256 bit data packet. All modern 3D cards use some type of DDR VRAM which can send two packets of data for each memory clock cycle.

Despite its age, Crysis is still one of the most GPU intensive games on the market, and it’s also one of the most visually impressive

We’ll avoid the mathematics behind it all, but suffice to say that the Bus width in combination with the memory clock speed determines the overall memory bandwidth of a 3D card. The higher this number is, the better – assuming the GPU can keep up of course. The HD 6870 has a rated memory clock speed of 1050 MHz, and an impressive memory bandwidth of 134.3 Gigabytes per second (GB/s). 

A good rule of thumb here is that if the packaging and marketing materials of the 3D card you are looking at doesn’t make a song and dance about the memory bandwidth and memory speeds, then they probably are trying to draw attention away from its inadequacies.

GPU core

The beating heart of your 3D rendering beast – the GPU core. Like the CPU clock speeds most people are familiar with, the GPU also has a core clock speed. In the case of the HD 6870 the GPU core speed is 900 MHz. This might not sound impressive at first, given that CPUs can easily hit 4GHz, but GPUs work differently.

Another factor that determines overall performance is the number of stream processors on the GPU (yet another unadvertised element). Stream processing is a highly technical form of parallel data processing, which is great for data intensive applications such as 3D rendering. 1120 stream processors coupled with a 900 MHz GPU core (and all the other elements previously mentioned) means that the HD 6870 can push out 2016 gigaFLOPS (2016 billion floating point operations per second). 

The GPU is essentially the graphics card’s brain

In anyone’s language, that’s a lot of FLOPS. However, GFLOPS capability isn’t a definitive way for gauging 3D card performance as there are many complicated processes going on under the hood, but it is a good way to measure cards up against one another when considering a purchase decision.

Fillrate is another measure of performance. This refers to the number of pixels a GPU can render and write to VRAM each second. In the case of the HD 6870, it can process 28.8 Giga-pixels per second and 50.4 giga-texels (textures) per second. Fillrate  has become less of a performance bottleneck and considering the number of stream processors is more important.

Fabrication Process

We’ll avoid a lengthy explanation of the GPU core fabrication process here. Suffice to say that the smaller the fabrication process, the more power efficient and cooler the GPU will operate. When things are running cool there is less electrical resistance, and the GPU can unleash its full clock potential.Overclockers love this sort of thing. The current standard is 40nm, and provides power efficiency not possible in previous generation graphics cards.

Application Programming Interface (API)

This indicates which modern APIs the GPU is capable of utilising. APIs offer various 3D effects capabilities and it makes sense to go with the most up-to-date 3D card available in order to squeeze as much life as possible out of the unit. The APIs typically supported are DirectX, OpenGL, and OpenCL. Most current generation graphics cards support DirectX 11, OpenGL 4.1, and OpenCL 1.1.

Conclusion

This simple guide gives an overview of the elements to consider when upgrading your 3D card. These bits of hardware kit are by no means considered cheap by most gamers so it pays to do some research into 3D cards before making a final purchase decision. Benchmarks, reviews, and specifications can help narrow down any potential list to within most budgets. Bear in mind performance bottlenecks that may occur in a system, such as system RAM and CPU speed.

Discuss graphics cards in the forums