Hacker News Reader

'roughly' is really poorly defined. CPUs leverage things like branch prediction and context switching, and also CPUs and GPUs both need to be designed for approximately arbitrary use cases, which is a much harder thing to do than designing for random input.

If you know for certain that the operations you get will follow some distribution, you can customize towards that distribution in ways a cpu or gpu can't.

From another comment here, X16R?

https://ravencoin.org/wp-content/uploads/2018/03/X16R-Whitep...

An interesting question.

One potential issue perhaps would be that it would presumably violate the memorylessness principle, which really changes the mining dynamics.

Is there a difference between what you're proposing and the more memory-intensive hash algos like Scrypt, etc?

I think so. I'm suggesting that small programs which use most CPU features be generated from the input, and used to transform parts of it.