Real-time audio processing is bound to CPUs forever. It's the opposite of the data that works with GPU computations : very small chunks of data (generally betwe

Real-time audio processing is bound to CPUs forever. It's the opposite of the data that works with GPU computations : very small chunks of data (generally between 64 and 512 floats or doubles) that you have to process sequentially. For those, SIMD is particularly useful - I got a ~7* increase in throughput recently when porting an effect from naive processing to AVX2 intrinsics - which means that artists can then put much more instances of the effect in parallel.