There are Super Nintendo games like Star Fox that had a second accelerator chip on board. But it would still be cool for the game to be the computer and exist in custom silicon. Development would be super slow of course but the result would literally be the best it could be.
Wonder if the advantage would be worth it. Is anyone trying?
I read somewhere that the chip used in Star Fox (SuperFX) is less an "accelerator" but rather "becomes" the main chip because it hogs the main bus, so the SNES's own CPU can't really do a lot. OTOH the SuperFX is not really a "game in silicon" but just a better CPU that has built-in fixed-point arithmetic including multiply/divide. All this is just from memory, so please take it with a grain of salt.
The SuperFX doesn't really become the main CPU, and has some practical caveats that keep it from doing so. Part of that is that the thing's instruction memory is only 512 bytes.
There are other accelerator chips that do take a more general approach though, like the SA1.
To be fair, the SuperFX was the foundation of an actual honest-to-god 32 bit embedded CPU architecture named ARC (Argonaut RISC Core), produced in the billions, because it was apparently quite efficient.
All the more impressive considering the guys making it had no prior experience in CPU design.
Contrary to popular belief the SuperFX and ARC cores are about as different as can be. RISC back then was a buzzword and applied loose and fast; today we wouldn't call SuperFX a RISC chip if not for the decades of describing it as such. ARC is very much a RISC design though.
Totally agreed on how impressive it was though, despite the differences in nomenclature.
Wonder if the advantage would be worth it. Is anyone trying?