Armageddon was held in Zemike

[puffyn]

I second that. Very neat battle, but my computer simply didn't have the power to run it at an acceptable speed.

I think the problem is with the spell-casters. I imagine that the spell-casting AI evaluates _all_ its choices every turn. With essentially everything researched this took about 10-15 seconds for each mage on my machine (1.5 GHz with only 512 MB of RAM); an acceptable delay for battles involving a sensible number of mages, but it's simply painful with hundreds of them involved. Fast-forward helps the graphic animations go more quickly, but doesn't help this spell-casting-decision bottleneck.

[Cainehill]

puffyn said:
I second that. Very neat battle, but my computer simply didn't have the power to run it at an acceptable speed.

I think the problem is with the spell-casters. I imagine that the spell-casting AI evaluates _all_ its choices every turn. With essentially everything researched this took about 10-15 seconds for each mage on my machine (1.5 GHz with only 512 MB of RAM); an acceptable delay for battles involving a sensible number of mages, but it's simply painful with hundreds of them involved. Fast-forward helps the graphic animations go more quickly, but doesn't help this spell-casting-decision bottleneck.

Those decisions are made on the host computer when the turn is run - it shouldn't be anything to do with AI that's making it slow to watch in the battle replay.

[Truper]

I think those decisions are made on the client machine when a battle is watched. As I understand it, the host passes the starting state of a battle and a random seed to the client (as well as the battle's result). If you want to watch it, the client has to duplicate the decision-making process so it can display the results. Otherwise the host would have to pass all the events of a battle to the client, down to the Last sword thrust, resulting in really huge turn files.

[PhilD]

Truper said:
I think those decisions are made on the client machine when a battle is watched. As I understand it, the host passes the starting state of a battle and a random seed to the client (as well as the battle's result). If you want to watch it, the client has to duplicate the decision-making process so it can display the results. Otherwise the host would have to pass all the events of a battle to the client, down to the Last sword thrust, resulting in really huge turn files.

I suppose you're right about the facts, though it would certainly be possible to have the turn file contain the result of all choices in a battle that were the result of a long decision-making process (it might require integrating several different "threads" in the random number generator, and other adjustments, but it would make it possible to include a real tough AI decision process in the turn generation, without either making the turn files huger or slowing down the battle replays).

[Cainehill]

Well, I looked at the battle and while there was some small amount of slowdown at times, it never got over 1 second a spell. I've no hotrod machine either, albeit 768 megs of ram and an NForce2 motherboard meaning fast memory access.

Also, given the way the computer casts spells once past the script, I hardly think it's dedicating any CPU cycles to something you could consider "AI". I mean - casting spells at the militia that won't get there for 3 turns, as opposed to the heavy cavalry that arrives in 1, for instance, or casting spells that not only kill bodyguards and other friendly forces, but the spell caster himself when there were better (less costly in fatigue, doing more damage to the foe) spells to be cast.

[WraithLord]

Very impressive battle.
I didn't watch the full of it though because it ran very slowly on my old machine

Truth is it might be a good idea to save a batch of turn files from some "famous" games.
This way new players can learn from the best.