Note that you must separate DSP into two things.
- Digital Signal Processor (a DSP)
- Digital Signal Processing (to perform DSP)

Any processor can perform Digital Signal Processing.

A Digital Signal Processor on the other hand is a processor that is designed with special capabilities. Adding a MAC will improve the processing abilities, but is not enough to make something into a DSP.

Compare. There once was the 8087 FPU. A special co-processor for floating point. But on the other hand - the 8086 could compute floating point values too without any FPU. But it had to rely on software emulation instead of hardware for the job.

Same with a GPU. Hardware designed for graphics - but a standard prcessor can also draw graphics directly in RAM without any GPU.


Correct. But 3GHz and 9 ticks of pipeline is 3 ns. 100MHz and 1 tick is 10 ns. And a number of microcontrollers have unbreakable instructions of varying clock length meaning that they don't guarantee that they start the interrupt handler directly on the next clock cycle, but may have a jitter of one or two clock cycles depending on the current instruction when the interrupt arrived. A number of GP processors also have multiple register banks or sliding windows of registers, allowing them to start handling an ISR without saving registers, similar to the 8051.

So in the end, the guaranteed worst-case absolute time for taking care of an interrupt can many times be lower in the high-end GP processors than in a microcontroller.

The big problem with GP processors is that when they are mated with GP operating systems, they are often run with interrupts disabled for very significant amounts of times. It's the OS that is the main reason for the loss of real-time behaviour.