There's No Such Thing as a General-purpose Processor (2014)

> If you remove the ability to run an operating system from a processor that is considered general purpose, then the result is usually described as a microcontroller.

I respectfully beg to differ. A microcontroller is distinctive in its integration of onboard peripherals; the processor is a primitive element of a microcontroller.

I'm not really sure I understand the author's point. I also disagree with his definition of general-purpose.

My car is a general purpose transportation device. Does that mean it is an all-purpose transportation device? Of course not.

It solves almost all of the problems I have related to transportation, though. But there are times I explicitly want to pick a different mode of transportation:

1. I need to cross an ocean: Choose a plane/boat

2. I need to go where there are no roads: Choose legs

3. I need to haul significant quantities: Train/Truck

4. I need to prioritize speed: plane

5. I need to prioritize attention: choose a subway/bus

Does this mean the car is the best form of transportation? Also no. But it's still very much a generalized means of transportation within our current context. It handles a user's daily and most consistent needs.

This is also true of our current processors.

The author's definition of "General-purpose" appears to be impossible, which makes his title both true and useless.

The Singer System 10 was a somewhat-successful early 1970s computer without an operating system - it ran from firmware. Yet it seems that it was a general purpose processor (as limited as it was). Had either 13 or 16 instructions in its CPU set. see http://www.fano.co.uk/history/sys10b.html

There is ProjectOberon http://www.projectoberon.com/ which has an FPGA based CPU (which requires only money to turn it into a real CPU), using 20 instructions.

The PIC32MZ micro-controller runs a MIPS32 core at 200Mhz, with MMU and FPU. This core or an earlier one like it, was used on the SGI Iris Indigo workstations back in the day ...

> For example, most modern processors lack the ability to cleanly express the memory model found in Multics, with fine-grained sharing and transparent virtualized access to memory-mapped I/O devices.

General-purpose does not mean all-purpose.

"On the design of display processors" has some great observations on the evolution of specialized/general processors http://cva.stanford.edu/classes/cs99s/papers/myer-sutherland...

The article completely misses the point of the idea of "general purpose".

"General purpose" means I can do what I want with it, as opposed to the manufacturer lets me do what they want me to do with it.

Makes sense to think of it as a general purpose processor when speed is growing quickly enough compared to gains of using a specialized design that using one device for everything is the most economical. Now that that has come to an end the model is breaking down and we need to consider the details more carefully.

The very talk of space division multiplexing rather than time division multiplexing just sent a shiver down my spine.

Caches are going to get super thrashed in those situations and just cleaning and reloading for the next application will destroy performance.