Ben Goertzel wrote: > The next question is: What's your corresponding estimate of processing > power?
Thanks for the prompt. Lets use the number 2^30 for the size of the memory which will require 25 operations for each 32 bit word. 2^30 bytes == 2^28 words. We are going to cycle the thing at the 30hz rate of the human EEG so the memory throughput required for the cortex part of the application (ignoring the critical nuclei and cerebellum) will be 30*2^28 > 15*2^30 bytes/second total, 15GB/sec. (this is the most critical number). A pair of servers with 8gb/sec throughput should be plenty and preserve the native organization as well... Now the CPU: We want to do roughly 20 operations to each of 2^28 words every 30 seconds. 20*30*2^28 > 75*2^31 > ~160 GHZ. (raw) Each server is responsible for 80. Split 16 ways, the load comes to 5ghz each. If we use a more conservative estimate of the typical EEG rate, say 15 hz, the load for each processor comes to 2.5 ghz... (splitting into more servers will probably not be practical due to network constraints). We could buy this for about $500,000. I would guess the cost to devel the hardware to emulate the varrious nucleii, (many of which do nothing more than a few simple vector operations), would probably add about $150k for custom cards (probably several iterations of such.) To make a meta-notation here, I am exploring this path towards AI because it is closer to a "sure fire" thing compared to a more radical idea I have that I hope to see implemented in the next generation. This more radical approach has some serious problems which may not be resolvable. > To emulate the massively parallel "information update rate" of the > brain on N bits of memory, how many commodity PC processors are > required per GB of RAM? Well in the above I only mentioned 1GB total memory. However there is almost certainly going to be an overhead above that gb... I invite the reader to factor in a sensable overhead ratio (for pointers and misc data structures) to the numbers above... -- pain (n): see Linux. http://users.rcn.com/alangrimes/ ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
