2022-08-28
A passing thought about particle systems. If you have to track the distance between each particle and every other particle, you have an O(n2) problem. To mitigate this, what if you maintain a nearest-neighbors graph? Any given particle will tend to have a small number of neighbors within some distance, and importantly, if you throw 10x as many particles into the system, this number of near-by particles (per particle) won't grow, so you have an O(n) situation. Immediate problem being, you have to tune the limit of what 'near' is carefully, and depending on particle dynamics, in regions where density increases too much, you get an unbounded increase in the number of near neighbors. Also, whereas you can deterministically grind away at the 'dumb' implementation (where you just compare everything to everything), with a dynamically updating nearness graph, there's going to be a lot of 'IF' flying around, causing havoc with instruction pipeline parallelism. Blarf. Probably better to throw the whole mess into a GPU and do it the easy way. Nvm.

2022-01-31
Imagine this: a custom CPU chip comprised of a 6502 core clockable at >= 1GHz, with a full 64K RAM (that's all a 6502 can handle, kids, w/o horky banking schemes) on board, and a full 64K flash. internally connect those memories onto the address and date busses, but bring their chip-select and write-enable lines out to pins so you can decide how to map them. This would save about 90% of the writing needed when cobbling together your typical retro SBC, and make it more reliable, too, especially at higher clocks. Would it have any real application at scale? Idk, but it'd sure be fun to play with.

2022-01-30
Interesting computer stuff placeholder. C64, PET, TRS-80, 6502, 68000, good times.

Possibly also the Burroughs Datatron 205, as seen on TV (before my time, but ultimate blinkenlights)

{Home}