The rise of the GPU

Posted on 10 Nov 2009 at 15:25

• 
• 
• 
• 
• 

 1 of 5

Gallery

Mike Bedford looks at the graphics processing unit's history and its new role in bringing supercomputing power to our desktops.

If we were to suggest that your PC contains a powerful and sophisticated processor you wouldn't take issue with us. But we're not talking about the likes of the Core 2 Duo, Phenom II X4 or Core i7. Instead, we're referring to another component that rarely gets into the limelight. The Graphics Processing Unit (GPU) on the graphics card works alongside the main processor to put images on the screen.

Unless you're a hardcore gamer or have other serious graphics applications, you might not be too concerned whether the graphics card is an ATI Radeon HD 4770 or an Nvidia GeForce GTX 275, and you'd be less interested in what type of GPU it uses. Even so, times are changing, and increasingly the GPU is being used for jobs other than texture mapping and rendering for 3D games.

Processing features that were designed to accelerate high-performance graphics are also pretty nifty at carrying out other tasks that require heavy number crunching, and computer scientists are starting to capitalise on this to help today's machines run more smoothly. From intricate applications for recovering lost passwords for encrypted documents to state-of-the-art supercomputers targeted at mind-boggling simulations in material chemistry, the GPU is proving itself as the supercomputer of the future.

Historical Perspective

In the early days of personal computing there was no such thing as a GPU. The main processor was responsible for placing images on the screen. The process was simple. In graphics mode (in the days before Windows, PCs also had an entirely separate text-only mode), an area of system memory was designated as graphics memory. Unlike ordinary RAM, graphics memory took the form of so-called dual-port memory, which meant that it could be accessed both by the processor and the hardware that was responsible for putting images on the screen. Each part of that memory was associated with a particular pixel on the screen. For example, to draw a line from the top left to the bottom right of the screen, the processor would work out which points were on that line and then, for each of those points, write a value to the corresponding location in the graphics memory. The graphics hardware would then light up each of those pixels to form the line.

The first PC graphics card that was capable of generating graphics rather than just text was referred to as Color Graphics Adapter (CGA). To say it was primitive would be an understatement. The highest screen resolution in graphics mode was 640x200 pixels, but this was only available in monochrome. In its colour graphics mode, which permitted just four colours, the resolution was a mere 320x200. Needless to say, the processing power needed to handle this sort of display was vastly less than what's needed today to cope with 2,560x1,600 pixels in 16.7 million colours, especially when we remember that 3D graphics as we understand the term now was just a distant dream in the days of the CGA.

The 1980s and 1990s saw the rise of the graphical user interface (GUI) where text entry was replaced by a computer desktop controlled via a mouse. As this style of computing gained in popularity, graphics were no longer considered an added extra and resolutions increased. Various technologies came along to reduce the increasing workload on the processor.

When hardware graphics acceleration appeared, for the first time the processor could issue simple instructions such as 'draw a blue line from (0, 0) to (1023, 767)' rather than having to work out the position of each and every pixel along that line. The first graphics cards used off-the-shelf components, but this paved the way for the custom-designed 2D graphics processor, which led to the 3D graphics processor that we have today. These are microprocessors in the sense that they can be programmed to execute instructions in sequence, but their hardware is tailored specifically to the requirements of handling 3D graphics.