[AnimalMuppet]

On the other other hand, if you've got a game state, and you want multiple cores to do the update, and you're not using mutation, then when core 1 makes a change (from state 0 to state 1, say), then when core 2 makes a change, it has to start from state 1, not from state 0, so that core 1's change doesn't get lost. So then you have the problem of updating all the cores (and all the threads) with state 1 before any of them make any further changes.

The way I would do this imperatively with N cores is, I would have the game state be an array, and I would have each core update 1/Nth of the array. That might thrash the cache, but it wouldn't need any locking at all.

Functionally, it could probably be done better to have each core produce a new sub-state, and then have one thread combine the N sub-states into the new state, and update all the threads with the new combined state.