The Amiga outsourced a lot of its work to custom peripheral chips, so doing something like a mass move of memory from one location to another, on a bit boundary, didn't rely on the CPU. Likewise, you could change the addresses of video bitplanes, which could give the illusion of graphics animation without doing huge memory moves. All of that accumulated to mean an Amiga had a better human interface than a PC, at least in that era, even if the PC could do more math.
Yeah but it's an architecture that couldn't make it past the late 80s.
For one because it couldn't keep pace with Moore's law and economies of scale. Custom chips are expensive and the talent to design them hard to source.
But most importantly the bus multiplexing on systems like that -- sharing video memory between CPU and dedicated coprocessor -- makes no sense in the 90s and beyond when memory became many multiples slower than the CPU.
The whole assumption of the Jay Miner design is that the coprocessor could do things faster than the CPU.
... but then CPUs just became way faster than main memory, and those fancy coprocessors would have been sitting there waiting on the bus just like anything else.
When GPUs came on the scene, they took a very different path. They have their own memory, and do very specialist things, and for a long time were very limited in the kind of bandwidth they had back to the CPU, too.
Anyways, Amiga had a nice system for 1986. By the 90s it was pointless and mass produced ISA video cards outpaced it, and they did so with a design entirely different.
Amiga and systems like it are not "paths not taken", they're design dead-ends.
With later systems, the coprocessors (blitter, at least) were often slower than the CPU. I remember having an A3000 and installing "CpuBlit" (see http://aminet.net/package/util/boot/CpuBlit ) Text scrolling speeds were vastly improved!
Exactly; there was a similar situation with Atari corp. Initial ST shipped without the planned but not finished yet "blitter" chip. With the STe they rolled that out, but a couple years later and it never got good market penetration.
But by the time they got to the late 80s with the TT030 and the Falcon 030 it become pointless because main CPU was in most cases faster than hardware blitting.
That's the problem with custom chips. An approach that was invaluable in the 70s and 80s for the 8-bit machines, but became a handicap by the 90s.
