On Sep 26, 2006, at 5:04 AM, Ben Goertzel wrote:
Speeding up this process using FPGA's is a possibility, hinted at by DeGaris's work using FPGA's to speed up GA evolution of NN's. However, right now this doesn't seem workable for the Novamente case because the Combo programs being learned inside NM make a lot of calls into NM's long-term memory store. A combination of FPGA's with some kind of super-large super-fast RAM cache might be helpful though...
Isn't this exactly what companies like DRC provide? Not only does the FPGA have full access to memory as though it was just another processor in the system, the latency is lower than for the Opteron on the same board (and the Opteron can use the FPGA's main memory banks as well). How much tighter binding do you need?
J. Andrew Rogers ----- This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
