Not so long ago, the relationship between the central processing unit (CPU) and the graphics processing unit (GPU) within a computer or workstation was very much like that between Batman and Robin. The dominant partner getting to do all the impressive work and take all the glory, with the sidekick filling in where needed... anti-shark spray anyone? According to NVIDIA however, this is getting more and more skewed as people realise the potential of the parallel processing power of GPUs. Batman and Superman is probably a little bit nearer the mark with the way things are headed.

Before we delve too much into the electronics of the concept it is important to reflect on what this will bring to the role of the designer. Performance gains of an order of magnitude are on the cards, especially in the realms of CAD and rendering. According to some simple facts and figures from NVIDIA, some rendering activities that could have taken days using a CPU-dominant approach will soon be possible in hours by harnessing the power of the GPU.

A typical workstation-sourcing exercise would comprise finding the most powerful CPU you could afford, backing it up with as much RAM as possible and then finding a graphics card that would suit your needs. Ben Berraondo at NVIDIA UK argues that this is all set to change. "It is likely that the GPU's capabilities are going to play such a dominant role in the future. The way that GPUs are evolving means that the CPU can be relegated to the simple tasks it is good at – doing them one after the other as single core processes – leaving the GPU to do all the hard work when it comes to processing the CAD data and then representing it on the screen."

So the big question is where is all this heading? Berraondo says that it is all heading towards custom tailoring workstations to a user's exact needs. Rather than basing everything on a core spec, users will be able to pick base-model workstations and then beef them up with the graphics card – it is the card that will differentiate one system from another, not the CPU. Reliability is also a key factor in NVIDIA's approach – with a 24-hour Tech Support aimed, as the company puts it, to "improving your working life".

Back in the day, Moore's law dictated everything. But CPU manufacturers soon realised that to get more power from single cores would mean greater heat generation, so, as we have all seen, multi-core processors are becoming the norm. Indeed, the ability to run in parallel has completely changed the programming environment. If we put this into perspective, traditional graphics cards have evolved with parallel processors from the outset. With over 240 cores in the high end, NVIDIA realised that its cards could do far more processor-hungry tasks than just rendering graphics. Berraondo elaborates that it is in fact easier to speed up parallel applications using the GPU than it is using the CPU.

The first result of this was the company's CUDA approach, a general-purpose parallel computing architecture that utilises the parallel compute engine in Nvidia GPUs to solve many complex computational problems in a fraction of the time required on a CPU. Over 100 million Cuda-enabled cards have been sold and the power they offer is steadily beginning to dawn on many within the computing industry. What is interesting is that rather than developing a proprietary technology that tied people to a certain programming format, Cuda drivers act like OpenGL and DirectX graphics drivers – opening the application base up massively making development easier and ensuring that cross-platform integration is possible.

To give an idea of what this GPU-based approach can achieve, in the world of high definition television encoding a 1080P film to another format or resolution using the H.264 codec would typically take a CPU-based system eight to nine hours. GPUs can offer speed increases of 10 fold meaning a two hour film might only take 20 minutes to encode. In many instances, it is NVIDIA's belief that it is now the CPU that is causing the bottleneck, not the peripheral devices. Another interesting demonstration is the Stanford University Folding@Home programme. Those of you with a PlayStation 3 may have seen this as an option on the PS3's main control screen. Folding@home uses the processing power of idle, internet connected processors to undertake protein folding calculations that are ultimately used in the battle to fight cancer. An unofficial friendly rivalry has arisen as many organisations try to make it into the top ten of the number of folding calculations undertaken

We mentioned above the DirectX graphics driver – a driver approach developed by Microsoft, for graphics-intensive operations on PCs – mainly gaming related. The next generation of DirectX – version 11 – will also contain GPU compute elements which are easily portable from CUDA. Indeed the move to GPU processing is not lost on the CPU manufacturers. Intel will be releasing Larrabee, its own GPU which should be available on sometime next year. It is pretty much guaranteed that we are going to see some major changes over the next few years.

Software developers are obviously moving down the GPU compute route, including many of the leading CAD houses. AutoCAD already has a very good relationship with NVIDIA hardware. On installation, it will recognise the hardware and immediately open up a variety of options that would otherwise not be available. AutoCAD will run happily on a good graphics card, but it will run even better on a similarly priced Quadro card. Indeed many other major CAD suites have Quadro enhancements; this is due in part to NVIDIA getting involved in the software's development stage. Engineers from NVIDIA often work with the software vendors to ensure that the mutual performance benefits are plain to see to the user.

Next on the horizon from NVIDIA is its Ion platform, a lower-cost GPU option that is designed to work in harmony with the popular processors within netbooks. MechanicalDesignOnline was shown a reference box which housed an Intel Atom processor and an Nvidia 9400M chip. This combination, claims NVIDIA, will offer a performance boost up to 10 times that of just an Atom processor, meaning that netbooks will be able to handle high definition pictures and 5.1 audio, while maintaining the same overall envelope. Many OEMs have seen ION and the first design wins are expected to be announced in the first half of this year.

It is clear to see that some fairly major things are afoot in the world of processing within PCs and workstations. The best news is that, as a design engineer, you will start to see the benefits very soon. Your Workstation is going to get significantly faster or significantly smaller, maybe even both. Either way, parallel processing is the way to go. Moore's law will soon lose touch with technology and the next generation of processing power will be upon us. Watch this space.

As a slight aside, if you want to see a very simplistic demonstration of how GPU parallel processing works – take a look at this video on YouTube.