>> How many Xeon transistors per clock tick? Any idea?  I recall
>> estimating .001 of neurons were firing at any given time (although
>> I no longer recall how I reached that rough guesstimate.)  And
>> remember, the Xeon has a big speed factor.

Mark> The Xeon speed factor is just less than 1E7.

Mark> Using your numbers, .001 of neurons is 1E7 to 1E10 if you meant
Mark> percent and 1E9 to 1E12 otherwise.

Mark> Also, you are treating synapses as binary bits with no memory
Mark> (which many experiments have proved is not correct).

All this I knew. But you didn't answer the interesting question,
how many Xeon transistors are active per clock tick? Anybody out there
know?

>> Well, on this we differ. I can appreciate how you might think
>> memory bandwidth was important for some tasks, although I don't,
>> but I'm curious why you think its important for planning problems
>> like Sokoban or Go, or a new planning game I present your AI on the
>> fly, or whether you think whatever your big memory intensive
>> approach is will solve those.

Mark> All of those "planning" tasks require recognizing salient
Mark> features from previous experiences.  In the case of Sokoban or
Mark> Go, this process may well be compiled enough that the memory
Mark> bandwidth is unimportant at game time but, in those cases, the
Mark> memory bandwidth is still required at compile time.

Yes, but is this more than we have in machines? Its devising the
compilation algorithms that's the bottleneck. We can build machines
with adequate memory bandwidth-- if Moore's law holds for long,
we will surpass humans-- but we will never get near the amount of
processing power evolution put in to algorithm design. We may
sidestep it and succeed anyway, or we may fail.

-------
To unsubscribe, change your address, or temporarily deactivate your 
subscription, 
please go to http://v2.listbox.com/member/[EMAIL PROTECTED]

Reply via email to