Even if you will never touch a game on your PC, there are good reasons to invest in a capable graphics card. Today we will examine the GPGPU performance of recently released graphics cards, and see what the best applications are for this technology. Which card should you have to get the most out of Adobe Photoshop or Cyberlink PowerDirector?
The main reason for spending money on a high-end graphics card is gaming. The powerful chips on modern video cards can be used for much more than that, however. GPUs have transformed into a tool for other things than just powering demanding 3D graphics, and have become fully programmable processors. That is why, when you use graphics card for things other than gaming, it's referred to as GPGPU. That stands for General Purpose GPU.
We saw the term GPGPU for the first time in 2005 in a press release from Nvidia that talked about using video cards for things other than just gaming. The Nvidia cards at the time, the GeForce 7800 series, featured the DirectX 9 API which made it possible for GPUs to run their own code. And of course it's been further developed since then.
AMD, which at the time was still called ATI, was quick to jump on the GPGPU bandwagon. It was generally accepted that the growing power of graphics cards had potential to be used for more than just gaming. The design of a GPU also makes it more suitable for certain tasks than a CPU.
The foundation of a GPU is very different than that of a CPU. Normal processors contain a relatively small number of very powerful cores (typically two or four) that can perform a variety of different tasks. A CPU is great at doing things in sequence, but the number of possible simultaneous operations is limited to the number of cores.
GPUs are designed differently. They contain dozens or even hundreds of mini-cores that are individually not very powerful and can only perform basic tasks. Clusters of these mini-cores execute the same commands or calculations, but can do this on different data. That means, if you need to perform the exact same operation on lots of different data, a GPU can be better than a CPU for the task at hand.
Processing 3D graphics is a good example of parallel operations performed on lots of data. For compiling a single frame in a popular 3D game, first the location of each of the (tens of) thousands of triangles needs to be determined. The frame has to be converted to pixels, and for each pixel the lighting has to be calculated. You end up with the same operations performed again and again on different data. Not only games can take advantage of this, however.
The Nvidia Tesla cards are intended for GPGPU only and don't even have monitor connectors.