Tweet
performance
I want to chime in on the performance complaints with 1.21f. It is _not_ just the save data compression, as I do have autosave off. The between turns time in 1.21f is a minimum of about 2 seconds longer on my computer vs 1.17. This is a real pain mostly early in the game where the turns used to be basically instantaneous, but now take around 2 seconds per turn. The extra time seems to be fairly constant though, as the later game does not get significantly worse (between turns are still around 30ish seconds after 2000.)
There's no excuse for performance this bad on a game of this kind. Implement threading and do background processing during the players turn (duh). Check your algorithms and look for n-squared or worse performance, and write optimized cases for the most common situations (eg short circuit path searching).
I play on 80x80 maps. That's 6400 tiles, with less than 100 cities, and less than 1000 units. Even with every unit evaluating each tile in some horrible algorithm, there's just no way that should be adding up to 30 seconds of processor time. In fact, even with every unit evaluating every other unit, and every tile in the process, that barely comes to around 30 seconds of processor time. One would almost have to imagine an n-cubed algorithm being run for performance to be this bad.
I want to chime in on the performance complaints with 1.21f. It is _not_ just the save data compression, as I do have autosave off. The between turns time in 1.21f is a minimum of about 2 seconds longer on my computer vs 1.17. This is a real pain mostly early in the game where the turns used to be basically instantaneous, but now take around 2 seconds per turn. The extra time seems to be fairly constant though, as the later game does not get significantly worse (between turns are still around 30ish seconds after 2000.)
There's no excuse for performance this bad on a game of this kind. Implement threading and do background processing during the players turn (duh). Check your algorithms and look for n-squared or worse performance, and write optimized cases for the most common situations (eg short circuit path searching).
I play on 80x80 maps. That's 6400 tiles, with less than 100 cities, and less than 1000 units. Even with every unit evaluating each tile in some horrible algorithm, there's just no way that should be adding up to 30 seconds of processor time. In fact, even with every unit evaluating every other unit, and every tile in the process, that barely comes to around 30 seconds of processor time. One would almost have to imagine an n-cubed algorithm being run for performance to be this bad.
Comment