I've observed the AI self-destructing it's T1 units when it hits the limit. I think the slow down, especially on water maps, is due to something else. Pathfinding is my guess.

I appreciate that. I just thought the results were interesting.



Seems to me to be a path finding issue. Also I've strangely found that putting two AIs on one side on a water map ends up cranking sim speed into the floor while just using one does not. Weird. Reproduced a bunch of times. I wouldn't have thought it'd be much different, same sort of effective number of units and all. Some interaction between teamed AIs and path finding?

The fact the sim speed doesn't get any/much better when most stuff ends up dead I think points to a pretty serious bug which GPG need to look at. There's nothing more frustrating than not knowing if the game will play alright or not. Sometimes it does, othertimes just as you're getting into it the whole thing crashes to a snail's pace.

Until you get so frustrated you quit the game. Bah.

Hi,

I installed the AI mod in my gamedata folder but I don't know if it is working.
I tried a game against a cheating AI and it acted as if he had omni all the time.
How can I make sure that the mod is correctly enabled ?

MasterZ, try watching the replay from the AIs point of view. If it has full omni then the mod is not installed correctly.

Also make sure you dont have any other AI mods installed.


Anyway I hope to work on this mod today to look at the slowdown issue.

0.8 released.

I had a look back through the various comments people made about the sim speed drop (both in this thread and sorians) and I dont feel the issue is related just to water maps. Its just more common on them.

I think the key issue is when the AI reaches its unit cap. So with this in mind I have made it so that when the AI is within 50 units of its cap it will self destruct all T1 mobile units. This is a bit radical but hopefully it will prove/disprove my theory.

Give it a go and let me know if it helps.

Setons snails pace after just a few min in. I've never really bought the unit limit thing. I've seen games go -10 and know for a fact that none of the AIs are near the unit limit.

Still, that change doesn't seem a bad thing to put in anyway.

Hmmm setons... I dont really see a slow down til around 12 to 15 mins and its still positive sim speed until about 20mins.

What kind of slow down do you see? Can you try some other maps of a similar size too?

Also are you playing the game at a faster than 0 sim speed? Try setting it back to the default sim speed. I find if I accelerate the game it gets bogged down on some maps?

! AI will now self destruct mobile t1 units when it reaches the unit cap. This should improve long game sim speed performance.

Duncane, I think, instead of self destructing mobile t1's, it should launch them at an enemy.

Yeah that's probably about right. By slowing down I really mean that sim speed was still positive but not high.



I just started playing at maybe +2 and after I'd basically expanded and built mexes and started shoving some units up front and building a forward firebase sim speed hit 0 at which point I binned it.

Setons is the one that really bogs, anything with no water seems to run pretty well even with a couple of AIs with large amounts of units. Sort of pure water maps also don't bog down. Any other hot spot maps?

I tried some 5x5 water map with islands and the AI clogged up the narrow water channels building experimental aeon battleships. But it still ran okay. Put two AIs on it and it went simspeed -10 just as it was hotting up. Backing up what was said previously.



Yeah that's interesting. I am actually, well, I basically build a factory and then whack sim speed up to 10 until it's near built and then play on at +1 until things get hairy or sim speed drops on it's own.

Lurkah,

Setons is a big map. What kind of CPU do you have?

And when you say a 5x5 bogged down can you confirm the map and AI config? (eg how many and what types).

It seems the slow down issue maybe more a hardware or driver issue (you've probably seen this thread: http://forums.gaspowered.com/viewtopic. ... 6&start=50)

... sorry I've just read your updates in that thread... I can feel your frustration. I will try a few setons test tomorrow and if I see similar kinds of slow down I will look again.... (I would still like to know what 5x5 map you are talking about)

Last night my buddy and I played a map, also the evening before we got one in. He has a P4 2.8 Ghz. In a typical 10X10 map with 1 cheating AI vs the 2 of us (that's all his system can handle) anyway on stock AI this typically takes his PC to as much as -4 when I do a ren_shownetwork.

With this .8 AI the last 2 nights his PC never dropped below 0.

Hmmm ... Thanks for the feedback "TheSlamma"

I think a lot of the gain on lower spec'd PCs with this mod is because the AI doesnt bunch units in its base.

Oh I wouldn't say I'm hugely frustrated by the AI sim speed in general now I know what slows things down. Sim speed is fine using fewer cheating AIs and non-water maps. Even some water maps are okay. Like that one with the big round lake in the middle I've forgotten the name of.

The frustration was when we didn't know what caused it and Setons is something of a favorite. Oh and I suppose we'd all like to have naval AI back too.

My processor is an E6850 overclocked to 4GHz, 4GB RAM, 8800GTX graphics. I can see the max sim speed of guys in the same game and they're way below me.

Echoing TheSlamma, it's clear the changes you've made have improved performance there's no mistake. So please don't take it as a denial of the good work you've done so far when I pointed out Setons still bogged down quickly. Things have definately improved over stock!

Alright some Setons tests. First of all I teamed with a standard rush AI on one side and put two adaptives on the other side. Taking the left and right positions on each side respectively. No player on rear or forwards points.

Sim speed hit -6 at the 20 minute point where I gave up.

Alright, that was a lot of AIs. What about just one? So I did that. I took the forward position and put an a single adaptive on the other side in the forward position. Better start. Game stayed playable for some time. Then it rapidly started to go down hill past the 20 min mark. By 30 mins it was -9 and I gave up.

I'm literally just running Duncanes 0.8 oh and TotalVeterancy 1.5.

Hmm, it might be interesting to implement a simple test AI, and see if these slowdowns occur still.

The test ai would be completely simple. Eg.
- Build mexes, PGens and Land Facs.
- Build tanks if you have spare unit limit
- Send tanks to random points on the map.

We could then see if its really the AI itself thats causing the problems, or some other thing.

Is anyone experiencing the slowdowns NOT using the total veterancy mod?

I found deactivating it prevented the slowdown this morning.

Good lord really? Gah, you'll force me to retest with it off now.

Duncane, did you add something in the last update that allows the UEF to build percies from factories (not just the Fatboy)? I just noticed the AI building them. Sorian said he found a fix for it, but hasn't released a new version yet.

Yes I added a fix for percivals (its in the list of fixes on the first page with a ! next to it which indicates changed in last release). Sorian mentioned it was a typo so I found it and corrected it.

AdmiralZeech, the idea of plain AI is a good one, although my concern is that even a plain AI will still use the basic builder, attack and pathing functions and its probably one of these thats causing the slowdown. Its something to keep in mind though.

Lurkah, Interesting about the total veterancy mod. Let me know if turning it off does have an effect.

Im going to run some tests today on setons so I will let you know if I find any issues.

Tested TotalVeterancy and I think I can discount that as being a factor.

oops... I always download from the front page, and never read it.

I've been using total veterancy. I haven't noticed a performance issue.

D'oh!

well anyway thats about what my Ai is at the moment build tanks and throw them where it hurts. i could cut out all the fluff designed around and make it total random as well as standalone. but i am afraid the problem comes from core functions we dont reallywant to rewrite like pathfinder or build managers or worse C++ methods we dont have access to. ithink it would be good that GPG cleant a little their Ai lua and C++ code, there was typos in lua functions, there is still probably there is a truckload of lua code and i doubt any of us modders already had a chance to read it all, let alone take a close look at it.

maybee we could start a thread on support forum and add what we think is really plaguing the Ai code and that we fail to fix. taking diffrent design direction and recoding is modding fixing the original code so it works as intended should not be left to modders. As far as i am concerned a skirmish Ai should be a good filler for a FFA game or a good partner to enjoy 1v1, at the very least a skirmish AI should be a trainer to learn the basic game mechanics and cliches of multiplayer.

To date someone who steamrolls adaptative in 10 minutes still doesn't have any guarantee that he has learned the basics to take off from 1500 rating in ranked. The simple fact that stock AI is unable to force people to play under pressure early game but at the same time fails to defend itself and fails at teching too goes a long way telling how much the Ai GPG retails with the last patch sucks.

Sorian has started a thread about the performance issue so I will post further tests, comments etc over there.

Ajunta, I like to think with my AI fixes that someone couldnt "steamroller" the AI (particularly the AIx types) and not be at least a reasonable player

On the plus side I have an idea whats causing the AI to use team mates engineers. I hope to have a fix posted in the next few days.

i havent got the time to test 0.8 yet, sorian 0.92 neither, the fact is i am not a decent player, but stock Ai those released by GPG wont survive 8 minutes of a 1600 rated player, we are far from what was claimed about FA Ais.