hobold (hobold.delete@this.vectorizer.org) on July 27, 2012 11:53 am wrote:
> Eric (eric.kjellen.delete@this.gmail.com) on July 27, 2012 5:12 am
> wrote:
> [...]
> > And the next step going from that
> > is of course that
> if CPUs through vector extensions can deliver peak performance
> > within a
> certain fraction of that of GPUs, their major advantages in thread
> >
> control and OoOE will outweigh the advantage in peak performance of the GPUs.
>
> >
> Do not rely CPUs maintaining the extent of this relative advantage over
> GPUs. GPUs can still harvest a few fruits that aren't exactly low hanging, but
> well in reach. CPUs have a hard time even only touching more fruit with merely
> the tips of their outstretched metaphorical fingers.

I would rely on CPUs maintaining a similar magnitude of advantage. The advantage is not there primarily because GPU companies were unaware of them or unable to implement them, it is there because that's the economics of the markets in which the devices find themselves.

GPGPU has been in their sights for quite a few years now, and no revolution has occurred, no gigantic new markets emerging, no death of CPUs. Why in some future date will it make sense for GPGPU companies to significantly reallocate resources toward such things?

Maybe I'm wrong. Maybe it will happen. But in that case of course CPUs similarly have much lower hanging fruit on parallelism and compute side of the equation in which to take the market from the other side too.