Advantages of dedicated DSP vs. generic CPU

[vco1]

I'd like to know if there are advantages of using a dedicated DSP versus doing the processing in a generic CPU.

Companies like miniDSP and Hifiberry offer dedicated DSP solutions for audio processing. While there are also solutions that use the/a CPU (e.g. CamillaDSP). Does using a dedicated DSP offer any benefits/advantages over doing the audio processing in the CPU? Especially with what seems to me relatively low-cost/low-power DSP's.

As an example: is there a quality difference by using a Hifiberry DSP in an RPi over doing the processing on the relatively powerful RPi4 CPU?

 

[tomtoo]

If you mean for the quality of the outcome? No, its math. Both can do the math right.

 

[ZolaIII]

I'd like to know if there are advantages of using a dedicated DSP versus doing the processing in a generic CPU.

Companies like miniDSP and Hifiberry offer dedicated DSP solutions for audio processing. While there are also solutions that use the/a CPU (e.g. CamillaDSP). Does using a dedicated DSP offer any benefits/advantages over doing the audio processing in the CPU? Especially with what seems to me relatively low-cost/low-power DSP's.

As an example: is there a quality difference by using a Hifiberry DSP in an RPi over doing the processing on the relatively powerful RPi4 CPU?

Modern CPU's are all hibrid designs (Fon Numan and DSP end). They all have 64 bit integer, MCU unit and SIMD (128, 256 or 512 bit) and (some have and deticated DSP integrated) can cover pretty much what ever is needed. Modern DSP's are usually up to 256 bit's per unit.
Advantage of using deticated hardware are numerous but hardware is only as good as software for it is and how much it suits your needs. Among main advantages would be ensuring and controlling that it allows function optimally (no stalls od delays in the chain), power usage, cost and similes integration. But to my better knowledge we don't have a SoC or SoM up to date which would serve as good DSP development platform (Kirin 980 got close but Huawei never finished Linux kernel mainlining of it [because of Tensilica DSP on it, its capabilities and documentation and tools available for it]).
To sum it up you use what you can and which works the best for your use case.
Some things where never rewriten in modern manner (MCU - SIMD SMP) not to mention for a specific DSP so you will have to use implementations which you have. For example PCM to DSD conversion is written to use only integer and with SMP up to four giving that in consideration it won't work flowles on two A72 core's on Pi 4 (needs at least four of such core's) if you for some bizarre reason want to do it over DSD 128 and even for DSD 128 depending of the rest of processing chain, best regards.