Designing the Perfect PC

The GPU (Graphics Processing Unit, or graphics card)
GPUs are the driving force for graphically intense applications such as games, 3D modelling/CGI and video editing. Recently they have been increasingly popular for calculating large amounts of data in place of existing CPU-based processing clusters, mainly due to their superior parallel processing nature.
The main components of a GPU include:

VRAM: VRAM is onboard memory used to store textures and data waiting to be processed by the GPU core, similar to how the CPU accesses RAM on the motherboard. VRAM is based on the GDDR architecture. Later architectures have larger bandwidth allowing higher transfer rates between the core and memory.

Core: The GPU core is the main onboard processing unit. Similar to that of a CPU, but primarily used to calculate floating-point calculations for graphics tasks.

Pixel Shaders: Pixel shaders convert 3D co-ordinates and instructions into 2D data, or ‘pixels’, for display on the monitor. The more shaders the better as they act as parallel processors. As with most components, in most cases the performance gain is dependent on the unit as a whole, such as overall architecture, rather than specific elements such as this.

Clock Speeds: The VRAM, core, and shaders each have separate clock speeds. Older GPUs often link the shader and core clocks so that they are in the same ratio; increasing the core will increase the shader. The higher the clock speed the better when comparing the same model amongst different brands. Comparing different models by clock speed is not a true indication of performance; do not do this.

Scalability: Modern GPUs allow two or more cards to be linked to work as one. Some models may have enough connectors to support three or four cards at once instead of the typical two card collaboration. SLI and Crossfire are the main technologies for NVIDIA and ATI cards respectively; most motherboards only support one or the other. Some high end GPUs come in dual core and dual PCB (Printed Circuit Board) models, which incorporate inbuilt scalability between two onboard GPUs.

Choosing a GPU
As with most components, the choice of GPU depends on the budget and intended usage. It’s wise to look out for the following factors in each computer category:

Performance/Enthusiast/Overclocker: High performance GPUs generally have high clock speeds, large memory capacity, and often incorporate dual core/PCB solutions. These cards are best for high end systems which need the extra power. In particular, overclockers prefer smaller manufacturing processes when choosing a GPU to allow higher overclocks with minimal heat.

Server: Generally servers are CPU intensive and don’t require much GPU power. Integrated solutions are fine for file, print, and web servers.

The models to look out for include ATI’s 4870 and 4870X2, and NVIDIA’s GTX280 for high performance systems.

What to research
There are three main GPU manufacturers; NVIDIA, ATI and Intel. NVIDIA and ATI make expansion cards while Intel focuses mainly on integrated solutions. The manufacturer websites provide a list of models they offer, so it’s just a matter of searching online for comparisons of models in your price range. Larger monitor resolutions require more VRAM, so be sure to compare benchmarks using the same resolution as your monitor.