Yes, nearly every microprocessor out there uses microcode, built on what is largely a RISC-like
internal design ... but the advantage here is that all of the microcode is
on the chip. You're not sending-out to RAM nearly as often. Also, you can customize the internal design because you know in advance that
all direct access to it is coming from the high-level instruction layer as
you have implemented it. Chip designers also provide compilers, and "suggested"
(sic) ways that the high-level instructions ought best to be used. Teams such as the
gcc team of course pay very close attention. So the chip designers also have a good idea what the actual high-level instruction sequences will be.
Supercomputer applications are usually
##classified##, which of course means highly-parallel processing of maybe low-level functions such as those need by the now-famous
##classified## and
##classified## projects... But, at
##classified## dollars apiece, for each of the
##classified## projects, you can easily afford such goodies.
