Actually, if you're going to do graphics at all, and you have the
space for it, a GPU is going to be a solid win for energy efficiency.
Why else do smart phones use GPUs?  The GPUs they use are very small,
with like 1 to 4 shader cores, so they don't take a lot of area or
power.  But they are optimized for graphics, making them vastly
superior to a CPU in terms of joules/pixel.  Even for simple 2D
graphics.  There's an added bonus of better performance, but the main
benefit is energy.  Any specialized logic circuit is going to always
be more energy efficient than a general-purpose CPU, and a GPU is part
way there.  This is why smart phones also include MPEG decoder cores;
even if the CPU could manage to do the decoding fast enough, it would
use vastly more energy; even if videos are played occasionally, it's
still a win, and the only cost is some extra die space (because the
MPEG core would be power gated when not in use).

One of the reasons for the failure of Larrabee is that they tried to
do graphics on a multi-core CPU, which didn't perform that well, and
it was power hungry.  It really would not have made a very good
mass-market GPU.  Sure, we're having a parallel discussion about using
Epiphany to do graphics, but we don't have Intel's resources, and
we're not trying to compete on performance or energy, just openness.

On Mon, Oct 29, 2012 at 3:56 PM, Dieter BSD <[email protected]> wrote:
> The most reliable, smallest, lightest, least expensive,
> and most energy efficient component is the component that
> isn't there.
>
> So... for computers that:
> a) run a very simple window manager
> b) don't run fancy 3D CAD
> c) don't run scientific simulations (massive amounts of floating point)
> d) don't run games
>
> Why would such a computer need a GPU at all?
>
> Does anyone make simple cards with framebuffers and modern
> VGA/DVI/HDMI/Displayport ports?
> _______________________________________________
> Open-graphics mailing list
> [email protected]
> http://lists.duskglow.com/mailman/listinfo/open-graphics
> List service provided by Duskglow Consulting, LLC (www.duskglow.com)



-- 
Timothy Normand Miller, PhD
http://www.cse.ohio-state.edu/~millerti
Open Graphics Project
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to