I will toss out the idea that the moderns are simpler and more orthogonal to teach than the retros.
For example Classic Retro Motorola had an interesting register model where 8 bit registers A and B are ALSO simultaneously 16 bit register D, found on my favorite 6809 but also the 68hc11 microcontroller. This is mystifying to new programmers (so I wrote to register A which messed up register D what?)
The craziest thing about RISC-V register model is if you try to save a value to register x0 you will be very unhappy when you read zeros from x0 (because x0 is an infinite source of zeros on RV32I ...)
Likewise things get weird on retro CPUs where certain adds and moves can be written in assembly but can't be assembled by the assembler. I distinctly recall a test question where some new-at-the-time chip example was asking if you can inc its stack pointer or add hex 1 to its stack pointer or just pop it. It depends on the chip of course but generally the newer the chip the more generic its registers and rules.
For example Classic Retro Motorola had an interesting register model where 8 bit registers A and B are ALSO simultaneously 16 bit register D, found on my favorite 6809 but also the 68hc11 microcontroller. This is mystifying to new programmers (so I wrote to register A which messed up register D what?)
The craziest thing about RISC-V register model is if you try to save a value to register x0 you will be very unhappy when you read zeros from x0 (because x0 is an infinite source of zeros on RV32I ...)
Likewise things get weird on retro CPUs where certain adds and moves can be written in assembly but can't be assembled by the assembler. I distinctly recall a test question where some new-at-the-time chip example was asking if you can inc its stack pointer or add hex 1 to its stack pointer or just pop it. It depends on the chip of course but generally the newer the chip the more generic its registers and rules.