There are plenty of cases where the cost and complexity of sending code to the GPU means that you use processor intrinsics to speed up your code. As far as I k

There are plenty of cases where the cost and complexity of sending code to the GPU means that you use processor intrinsics to speed up your code. As far as I know, most people still do encoding/decoding of audio/images/video on the CPU, unless there's a hardware codec available. Audio DSP is mostly done on the CPU too, I think. Audio is a bit more demanding than graphics in terms of bit depth — the "highp" precision should be taken as a minimum precision for audio work, it's not always available in the fragment shader, and it's definitely not enough precision for audio production apps like Garage Band.