If you want to ask such a question, you better be more specific about exactly what has been done.

Your previous post was about isolating 8 bits at a time, and I did mention 32-bit chips that did 8-bit reads without need for any separate isolation.

Having two indices is good for a 8051. Most 32-bit processors may have 8 or maybe even 32. Sometimes with support for auto-increment or auto-decrement.

You can get an ARM chip for $1, even if most people probably ends up with $3-$10 chips depending on needs.

Some of the ARM can handle 8-bit accesses from memory without extra costs. Some can't. The ARM core produces the required info - it's a question of the memory controller glued to the ARM core.

If the ARM core does it in two clocks or five clocks is fun, but at the same time some ARM may do 40MHz and some may do 400MHz. So the length of a clock can vary quite a lot.

A LPC17xx does:

unsigned char *p1,*p2;

*p1++ = *p2++ // 4 clock cycles = 1MB/s at 4MHz or 25MB/s at 100MHz. Some parts do 30MB/s at 120MHz.

The LPC17xx are a bit high-end, resulting in most models costing around $4-$5 in volume.
But you can find much cheaper Cortex-M3 chips too.

If running 100MHz, you get a cycle time of 10ns or 40ns/transfer.
If running 120MHz, you get 33ns/transfer.

Having a PLL means that the same chip can dynamically switch speed, depending on needs.

The cheapest NXP Cortex-M3 seems to be around $1.5 to $2 and manages up to 72MHz.


No magic hidden in the sentence. Every instruction you can implement in an 8-bit processor can be implemented in a 16-bit or 32-bit processor. So it all comes back to my previous claim. You can't say that 8-bitters are faster for microcontroller work. You can only say that a specific 8-bit processor can win over a specific 32-bit processor when doing a specific task. There is nothing magic in an 8-bit processor that can't exist in a 16-bit or even 84-bit processor.