GPU performance: Video Transcoding

Video Transcoding over the GPU

A new trend in GPU development is using the GPU for things other than rendering games. Parallel processing aka Stream processing. ATI uses ATI Stream and NVIDIA uses CUDA. As the topic now slowly becomes more familiar we feel now would be a good time to start collecting data that measures the performance of the GPU while computing and crunching numbers.

One of the most definitive functions the GPU will be and can be used for, is video transcoding. So we sought an application that can transcode video files over both the CPU and GPU.

Below you can find the first results of this new test. In this test we transcode a 200 MB AVCHD 1920x1080i media file to a 1280x720P MP4 binary.

So here we see the result of video transcoding. The numbers are the seconds it takes to finish transcoding; the lower, the better.

As you can observe, the GPU is very well suited for this process. In fact when we look at the 5670 which is utilizing ATI Stream, it even beats a Core i7 965 with its four CPU cores overclocked to 3.75 GHz. Think about that for a second and realize how insane that really is.

We are running into a curiosity though, all graphics cards end up at roughly the same performance. The software makes use of a symbiosis of both the GPU and CPU, and on this overclocked Core i7 platform the CPU just is too dominant. But sure, the GPU is assisting and helping out speeding up the process by 4 seconds.