[ExI] Smallest human-equivalent device

[Eugen Leitl]

On Sun, Oct 13, 2013 at 09:16:53AM -0700, spike wrote:

> Heat transfer problems never go away.  At the end of the miniaturizing game
> we have enjoyed for the last half century, heat dissipation is what finally
> stops our fun, and explains most of why processor clock speeds have stopped

Heat dissipation is what already stopped our fun with frequency doublings
almost a decade ago. 

> increasing.
> 
>  
> 
> Agreed there is room for further progress, but what I see is more efficient
> use of smaller processors.  I am thinking about checking out a local ARM
> processor conference this week:
> 
>  
> 
> http://www.linleygroup.com/events/event.php?num=24
> 
>  
> 
> I get the feeling the future development is in these small energy efficient
> processors with open architecture.  We might be able to get a bunch of these
> things working together to form a kind of intelligence.

The exascale machines are effective embedded SoCs with stacked memory,
and a mesh fabric router. The actual challenges of exascale is graceful
handling of unreliable transport and unreliable components (if you have
millions and billions of widgets, you will need to keep on trucking
when there's failure here and there). I don't think we'll get spintronics
in the first generation of exascale, but it's the only way to go on
after that.
 
> When we talk about a human equivalence intelligence, we may find it is
> impossible to decide exactly what that means.

If you have an isofunctional system, this is effectively your benchmark.
However, the only way to construct isofunctional system is for isolated/small
scale systems, and many that are evolutionary old.

I mentioned retina, but retina doesn't do high connectivity. So it
is highly suitable for embedded-like SoC computation.
 
> I will make this speculation: the next generation of humanoid robots will
> have a bunch of interdependent processors.  Future sports will be robots
> racing and playing against humans and against each other.  This stuff will
> be big entertainment and possibly big employment in the near term future.

The big advantage of autonomous cars is that this not just kills trucker
and delivery jobs (the poor bastards), but it vastly reduces the number
of cars built. So it actually hits car makers far, far more than it
hits truckers.