The CPU of the Raspberry Pi is a slow moving, general purpose device. Don’t get me wrong, it is an amazing combination of price, low energy use and flexibility. But, The Graphical Processing Unit (GPU) in the Pi is downright amazing. It can convert a compressed video and pump it out to the screen at the same resolution of a Blu-Ray player. But the GPU is limited to specific number crunching activities and it isn’t very flexible.
Up until this point the GPU has been a black box. The features were restricted to graphics processing for items on screen. But no longer. Andrew Holme has created a Fast Fourier Transform (FFT) library using the GPU. Most engineers know what a FFT is and they know, computing a FFT is mathematically complex. If you’re not an engineer, the FFT takes a signal (sound for example) and splits it up into the frequency parts. It is similar to the spectrum analyzers found on stereos. The sound is split into the different frequencies and the level or volume of each frequency is shown. The left is usually the low end and it is easy to see the beat of the bass drum as the left bars shoot up. Using the GPU for FFTs on the Pi gives between a 5x-12x speed improvement.
Unless you are planning on building a software based radio or adding a spectrum analyzer to you streaming media server, this may not sound like big news but the FFT library is only the start. Unlocking the GPU means that more libraries are coming. To start, I expect that the Pi will get some audio altering filters so your playback can get effects like “concert hall” fond on high end receivers. Also, I expect that someone will build a GPU based transcoder allowing the PI to recode audio ad video without having to move that task to a higher powered device. Lastly, everyone knows the best bitcoin miners use GPUs. Look for new bitcoin libraries soon.
Instructions to install the FFT routines can be found here.