Connectome and how could open source hardware make use of one?
I seen ted talks on connectome and a ted talks on IBM blue brain.
That made me wonder linux has clustering distributions that go on sd cards.
IBMs blue brain runs some kind of connectome structure by putting 1000 artificial neurons in eatch processor of 4 refrigerator sized boxes of networked not soo large processors 16000 of them to run a neocortical column simulation.
Hypotheticaly would linux be neeeded for a system like that or could maple and a fpga be one node of that system? Would it be larger and not require as many? Or would it be better suited for something like a arm8 in the beagle bone? Would it be better as pannels of processors hard wired with wires or wireless chips?
Connectome and how could open source hardware make use of one?
(2 posts) (2 voices)-
josheeg
MemberPosted 3 years ago # -
gbulmer
ModeratorHard to say without details about what software they need to run. A neuron is so simple that I can't imagine needing anything as complex as a a real OS. I could imagine them having a TCP/IP stack (maybe just UDP which would allow broadcast and topic based routing)
I do remember years ago someone doing an experiment where they put, I think, 1000 MIPS processors (comparable to ARM) on a single chip.
IBM have all the technology and their own processor designs to do that sort of thing.I believe folks have put 32+ 32-bit processors on a single FPGA (e.g. micro Blaze http://en.wikipedia.org/wiki/MicroBlaze).
I don't think 32-bit processors are essential. There was a machine called the connection machine http://en.wikipedia.org/wiki/Connection_Machine
which used processor which were 1-bit serial.
I'm not convinced its trade off of width vs clock was ideal, but a modern FPGA has 'cells' which are not too much simpler, so you could expect to have 10,000's of cells on a single FPGA.With enough on-board RAM to hold the "program", then there would be enough to do the worm, and more.
Posted 3 years ago #
Reply
You must log in to post.