Unconventional AI CEO @NaveenGRao explains the GPU energy problem: 99.9% of energy goes into moving data rather than computing, and non-von Neumann architectures could eliminate that bottleneck "Every computer that's been built in the last 80 years or so works on the principle of having an explicit state store we call memory. We pull data out of that memory, and we do something to it and write it back. That works great and it's a generalized machine, but the problem is almost all the energy in a computing system goes into moving data around." "99.9% of the energy's about data movement. It's not about the computation. The way these machines work, their explicit operation requires movement of data always" "There are ways to build computers, actually looking at how nature's built computation, where data sits more still. It doesn't have to move around between memory to computing elements." "We can have memory and computing in a single element. This is what we call a non-von Neumann architecture. Using the time varying physics of those elements to actually do computation for AI." @unconvAI