Bitgrid

Probably not meeting your criteria, but anyway, I'd build a BitGrid ASIC[1]. I think it's possible to democratize access to Petaflops, while vastly reducing the power requirements. So... yea, faster Mandelbrots, but for way less power? [1] https://github.com/mikewarot/Bitgrid

It's my idea... I've been writing about it for over a decade. I hope to get a small version of it made via tiny tapeout later this year once I'm out of the precariat. The idea is that you can decompose any non-cyclic code, for example, a matrix multiply, FFT, etc.. into a directed graph of bitwise operations, then map those operations out into the grid of LUTs. Pipe inputs in one side of the grid, and get the outputs out the other side, and all of the bitwise operations happen in lock step parallelism. (Unlike an FPGA which is asyncronous) If you can decompose one stage of an LLM down into a bitwise graph of computation, you can easily map it into the grid. If there's a bad cell in the grid, you can map around it. Because all of the high speed logic lines are short (to the neighbors) you don't have the long lines / high capacitance issue

I believe in Embedded Growth Obligations[1], which we need to come to terms with. -- I believe that computer security is a solved problem, but very few people understand the solution, capability based security.[2] I further believe that because of this insecurity of our computers, people avoid straying from "strange websites" and thus aggregate in the walled gardens. I further believe that this insecurity will be the cause of losing the war for general purpose computation, and thus losing democracy, what little we have left of it. It's quite possible this is all part of a plan. Multics was left to die on the vine, and isn't mentioned much any more, because secure computing would make the NSA's job harder. -- I believe that Von Neuman computing is not the path to cheap Petaflops, but BitGrid[3] is. (It's my personal hobby horse, I may be b