Rendering and Compression Power From GPUs

There's an excellent article over at ExtremeTech on the future of using the GPU on graphics cards for applications other than just displaying graphics on a display. They discuss Nvidia's CUDA language and ATI's similar, but somewhat more closed, equivalent. GPUs have tremendous parallel processing power; for example, ATI's new RV770 GPU has 800 stream processors.

The ExtremeTech article talks about how GPUs can now, and increasingly will be used in the future, for video encoding, decoding and transcoding, as well as rendering and effects in Photoshop. I believe that there's a tremendous opportunity for speeding up video editing and effects (especially rendering) in programs such as Premiere Pro, After Effects and Final Cut Pro. One big issue, of course, is that Nvidia and ATI are going off in their own directions, but as the article states, there are two standardization efforts--OpenCL, which is backed by Apple and a bunch of other companies, and DirectX from Microsoft--both of which are supported by Nvidia and ATI. Of the two, it looks like OpenCL may well be the more generalized platform for parallel application development, but we need to wait and see what DirectX 11 will be able to do.

The key here is that it will eventually no longer be necessary to throw out an entire workstation in order to get better video editing and effects performance--you'll simply be able to swap out a video card (or two or three or four). Apple is going to support OpenCL directly in its forthcoming Snow Leopard version of OS X; it's too early to tell what Microsoft will do with Windows 7, but there will undoubtedly be GPU application support, even if it comes from third parties.