What is the DSP processing power of the older hardware like a 480? You can usually do more with dedicated chips. Modern dedicated DSP vs modern CPU the DSP clearly wins. But what is the delta of older dedicated DSP vs modern CPU?
This subject came out when the first VST applications came out. In terms of pure number-crunching power, there is no debate. Even in the late 90's, any laptop had much more raw power than any dedicated digital audio hardware.
However, computers are plagued with so much housekeeping duties it impairs their capacity, in particular in regard with the constraints of real-time operation. Managing display priorities and many other background tasks interfers with actual real-time processing power.
That is why the Apple computers of the time had a significant advantage for audio and video applications, because of their different management of these. Actually the Atari, although beinglagging in terms of raw power, was a winner in that respect.
Dedicated DSP hardware still offers many advantages, in particular by establishing direct pipelines that a typical von Neumann archtecture does not provide.
I always thought fixed point was a better way to do audio.
This was certainly true for a time. It looks like floating-point processors now have overcome their limitations.
Floating-point has the disadvantage of introducing rounding errors, that are constant in percentage, where fixed-point would introduce absolute constant quantization error, but is a clear winner in filters, particularly at low fequencies, where the accuracy of coefficients close to unity is paramount.
Many filters rely on division by 1-N, where N is very close to 1. So it can be written as K=1/(1-N) or K=1/epsilon, where epsilon is very small, so in fixed-point, the quantzation error can be be quire significant. Floating-point allows minimal error on these numbers.