There are various GPU grids, some of which support running at least some parts of compiled java.
I'm looking for, or maybe to build if it doesnt exist (anyone interested in that kind of project?) a network of GPUs which each have internet addresses or you could call them memory addresses within the same program which just happens to be running on a computer thats spread across the Earth.
I'm looking for 2 or 3 such grids which can each consume 0 money and computing resources until told to do something and within 1/10 second rev up to a speed of a small animal brain, then a few seconds later turn it off, and repeat as needed in such short fast bursts.
I dont think the world is that well organized yet. Or is it?
The main ops such gpus would be used for are read bit at address elsewhere in the gpu network, read scalar number depending on that bit, sum those scalars, update those scalars or other numbers, powers of e, and to take commands from slower running code on when to do these things and put such neuralnets in memory and update them etc. I cant accept any existing neuralnet software as it must be computed at the code level so I can explore variations of the algorithms, and so such neuralnets can design new variations of such algorithms.
I know how to build a new JVM or part of it if I have to, but it appears that work has already started in opensource of some kind.
--------
It would be good if they accepted some kind of cryptocurrency but its not required since there would be a few of them to keep them competing with eachother and peoples individual computers.
Or maybe they could branch a new blockchain as a localcurrency (local in the sense of those they do business with) and trade it to anyone who would use this opensource science, game, and AI network in exchange for services provided to them like statistical prediction of high dimensional vectors, is the main thing I'm going for at the low level, and specific uses of it and recursions applying variations of it to itself would be built on top. For example, in exchange for compute cycles and storing of datastructs temporarily and network bandwidth, I could eventually extend these multiple GPU networks from different providers to address into eachother, delayed of course but adjusted by prediction of that timing, so they could more easily sell computing services to others who dont want to build their own addressing or prediction systems but could if they want to.
I may also be interested in any experimental neuromorphic, quantum, brain layers on a chip (like
http://en.wikipedia.org/wiki/Physarum_polycephalum but a kind of life that thinks closer to our speed instead of thousands of times slower as the slime does), insect or monkey cyborg experiments (with respect to their positive experience and choice of joining and leaving the network to motivate the operation of the network to provide value to all involved), possibly FPGAs if they're behind a protective layer not able to damage themself by experimental programming, or just in case anyone may have anything like it... a tiny "black hole computer" somehow hooked in through statistics. I'm interested in any advanced kind of computing and hooking many of them together as a single statistical computing space.
