Microchip: 7x faster, 30x less power

rice.edu

24 points·superkarn··5 comments

The streaming video example is pretty to understand. But I don't quite follow the encryption one. Don't you need to be precise with the calculation to encrypt/decrypt? Or would the advantage here would be mainly for generating random numbers?

generating random numbers seems right.. maybe they switch chips when accuracy matters, or maybe they don't care if some of the output gets mangled since the human can 'ignore' the mistakes.

I wonder if you have to program them any differently...

sometimes_print("Hello World! I guess, at least in this case...");

Maybe you could have 3 processors running the same code. You could repeat any steps that resulted in disagreement between processors. If the 30x less power claim is true, you would still be saving ~10x the energy(Less than 10x due to repeated calculations and inter-processor communication).

Wow this sounds very intriguing but I have no idea where to start learning about this. I tried reading some of the papers and was quickly overwhelmed. Anyone have any suggestions for how to start experimenting with this? Are there any VHDL like kits for doing probabilistic logic?

Is there any hope of there being dev kits made that would allow one to experiment cheaply? I want to try and see if this gives any advantages for monte-carlo algorithms.

Reply on news.ycombinator.com