Re: Human-level CPU power crossover date

From: Dani Eder (
Date: Tue Apr 10 2001 - 15:34:30 MDT

The big problem with
> this is that it would require an obscenely fast
> memory bus (say 8-megabits
> wide at 8-terahertz) to feed the processor core in
> the best case
> scenario. Given the tech for this kind of memory
> bus, the memory required
> (which would fit snugly in a 64-bit address space)
> and the processor core
> would already be solved problems. Massive
> distributed computing won't work
> because the effective memory bandwidth is so low
> that the actual throughput
> will be orders of magnitude less than suggested by
> simply aggregating the
> abilities of individual processors.
> That said, I think a well-engineered AI solution
> optimized for silicon
> would require substantially less hardware than
> suggested by the above.

Memory bandwith is already a bottleneck in desktop
PC's, where the CPU is running up to 10 times the
memory bus rate. The solution is to put processor
cores on the memory chip. Think of it as re-arranging
the silicon real-estate. You've got a certain amount
of silicon in the CPU, and some more (quite a bit
more, usually) in the memory chips. If, instead
you have each chip be 90% memory and 10% CPU core
the same total silicon real estate (and hence cost
to manufacture) is used.

A current 256Mbit memory chip has enough transistor
count to house a small, simple CPU. If used to
simulate a human type neural net, you can choose
the shape of the 'brain slice' that each chip is
simulating so as to minimize the inter-chip

Now, I'm not saying that neuron simulation is the
right way to build an intelligence. My 10^17 bits/
sec number is only a guideline to when we will have
powerful enough computers, based on the meatware
example between our ears. It may well be that
an efficient algorithm can do the job with 3 orders
of magnitude less processing power, but it should
be unsurprising that early attempts at AI failed
using 10 orders of magnitude less. I wouldn't be
at all surprised if it could be done with 1-2 orders
of magnitude less power, since some of the brain is
dedicated to just running the body, people can
lose large chunks of their brain and still function,
and evolution hasn't had time to really optimize
the most recently developed parts of the brain.


Do You Yahoo!?
Get email at your own domain with Yahoo! Mail.

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:36 MDT