By: RichardC (email@example.com), April 8, 2017 6:15 pm
Room: Moderated Discussions
> Like, to take Richard's point again, if we had some kind of new memory - maybe a memory that was
> able to match more closely the speed of the processor, and was low latency - then the RISC processor
> might make sense again (I mean, a CPU architecture, other than the one has is most common today in
> 2017). But we have hit a wall in terms of clock speed of the processor, and it does appear as though
> the speeds of memory will close the gap at some stage. Maybe that will become a contributing factor,
> that will lead some architecture like ARM to replacing x86 for a lot of things? Or put that another
> way, maybe it was the lack of availability of fast memory (relative to very, really fast processor
> clock speeds), that gave the x86 systems the seeming advantages that they enjoyed.
One huge change from the 1980s *and* 1990s is that we now have much better tools and
techniques for writing parallel and distributed software systems. That has moved the
goal of cpu designers away from single-thread performance and towards throughput-per-$
and throughput-per-watt (not exclusively, but to a large degree). So it isn't likely
that we would see a repeat of the 1980-1995 race for ever-faster single-thread performance -
a more likely field of competition is in architectures which can expand highly-parallel
GPU-like processing to a wider variety of applications: it seems that GPU cores have been
evolving to become more general-purpose, and AMD HSA is a step to reducing the overheads
of mixed CPU+GPGPU apps.
But anything could happen if a new technology and/or new markets spring up. That's what
makes it a fun business ...