Simulation_Player
Registered
AI running player physics would be gigantic waste of resources.....resources which could be way more useful for future player/car physics updates/features. players physics are one that matters anyways.
Released April 2022 Release Candidate Update
- Thread starter Paul Jeffrey
- Start date
Paul McC
Registered
Hmm, interesting but, im confused! ACC presents a much bigger load on CPU than RF2 even when only just running the player car, so for example: ACC @ Spa, Just player car= CPU about 40%. Same senario with player car and 49 A.I.= CPU about 70%
RF2 @ Spa, Just player car= CPU about 20%. Same senario with player car and 49 A.I.= CPU about 26%
So im a bit miffed as to why RF2 would max out a CPU if the A.I. used player physics? Also when I mentioned that A.I. running on player physics might sort out some A.I. issues I did'nt mean anything like strategy problems, pitting probs and such (sorry, probably should have clarified that earlier), I just meant the way the A.I. cars actually move, their grip levels out of corners (which is practically impossible to match in some cars), braking grip and also odd things like (and this usually happens with BMW CS & BMW M4 Class 1 for some reason, not GT3 though), catching the edge of a flat kerb and flipping right over (just take the BMW M4 class 1's out to Indy GP and you'll see what I mean).
EDIT: P.S. (& this is quite a big one) What happens to you if you bang doors on track with an A.I. is very different to what happens to them, they are much more anchored in a collision than player car, this is where ACC having all cars on same physics is good, as if you collide, you and A.I. have a pretty equal chance on loosing it or keeping it.
Im not sure about multithreading but am I right in thinking RF2 only utilises 1 core? Whereas ACC uses all.
Cheers!
Last edited:
memoNo1
Registered
Not everyone has the knowledge about the matter like you. I e.g.!
This is not an attack against you personally.
I would have been interested in your answer.
But OK.
Simulation_Player
Registered
I think Lazza's previous comment covered the topic already.
quote >>
AI with player physics is just plain stupid idea unless everyone have some NASA supercomputer.
even then it will be objectively useless since you don't need full blast physics to simply make AI "drive"
There are way way more crucial in players car physics left unexplored....it is not like rf2 has hit its full limits on vehicle dynamics and now only room for improvements is giving AI the player physics.....even then this topic is MIGHT be slightly worth arguing about.
infact i think giving AI player physics will actually brake them in different, more hard to fix way ....who knows.
quote >>
- AI running player-level physics is unrealistic for more than about 3 AI with current CPUs (multithreading would help, but still not big fields) and wouldn't help with any of the AI issues people are concerned about (actual AI physics rarely gets mentioned, from what I've seen)
AI with player physics is just plain stupid idea unless everyone have some NASA supercomputer.
even then it will be objectively useless since you don't need full blast physics to simply make AI "drive"
There are way way more crucial in players car physics left unexplored....it is not like rf2 has hit its full limits on vehicle dynamics and now only room for improvements is giving AI the player physics.....even then this topic is MIGHT be slightly worth arguing about.
infact i think giving AI player physics will actually brake them in different, more hard to fix way ....who knows.
Paul McC
Registered
Er, what? Did'nt know we were arguing mate, just discussing a topic. And if you don't understand what im saying then fine, maybe I could have explained it better. Anyway no worries, cheers!
Paul McC
Registered
Well no not really because as i've already pointed out
"ACC presents a much bigger load on CPU than RF2 even when only just running the player car, so for example: ACC @ Spa, Just player car= CPU about 40%. Same senario with player car and 49 A.I.= CPU about 70%
RF2 @ Spa, Just player car= CPU about 20%. Same senario with player car and 49 A.I.= CPU about 26%"
So ACC uses more CPU running just the player car than RF2 does running a whole field of of 49 A.I. plus player car, so I don't understand how it could tank a CPU anymore than ACC does, and thats only on a I79700F, so not the top tier of cpu's even.
ACC A.I. use same physics as player as far as I understand and im not running a super computer here.
I agree though that A.I. running same physics might cause a whole shedload of other problems though as you say, but I do like the way it works in ACC.
Paul McC
Registered
Ah, brilliant! I had no idea as i've never used Dev mode! I might have to look into that, thanks very much indeed mate! Cheers!I think there is a setting in dev mode that makes the AI drive with player physics. I guess we could try seeing how many AI we can add before the computer grinds to a halt
Simulation_Player
Registered
Both completely different softwares, you can't really compare them.
I don't understand why kunos decided to run AI at full player physics , maybe ACC is not that deep/complex physics wise so they have room to spare.
Looking at the stats you posted, it seems soo unnecessary to lose so much performance for ai cars physics....which u are not driving ,
so why bother with how they drive ?as long as you can tune their lap times, braking , grip level etc.
it is truly inefficient way.
I would much rather have more resources making player car physics as realistic as possible, simulate more elements real time etc, than make ai look realistic as possible for eyecandy.
On ACC you gained whopping 30% more CPU usage.....are you really gonna look at every 49 ai cars to see how they move about and stuff?
Some AI issues can't be solved by just giving them more complex physics, sometimes all it takes is a good AIW.
Some extra parameters to tune them better etc.
Paul McC
Registered
You might be right as of corse ACC does'nt use a full physical tire model like RF2 does, but then why does it use more CPU with just the player car on track than RF2 does with the same, I don't know, maybe its just the nature of the Unreal Engine, I guess it could just be that its not the indiviual cars soak up all the CPU, maybe its just the engine itself.
Because of the way they act if they come into contact with the player car for instance. Its not like they are not without their faults and RF2's A.I. does do some things better, but in terms of contact, alot of the time they are much more planted than you. Also they can grip out of corners (especially in the wet) in a way in which the player just plain cant. Also I do enjoy my offline racing so the A.I. behaviour is important to me.
Yes absoloutely! For example this is one area where ACC is not so good as the A.I. will ridgidly stick to their path if you are on the inside of a corner and half a car ahead they will end up pit manovering you on the exit as they will not yield, whereas RF2's will, but also if you are half a car behind they won't yeild, as it should be, so you can't bully them, but they will respect if its you're corner, so to speak. Doe's that make sense?
Paul McC
Registered
I know you like to do the odd bit of offline stuff mate! You can't beat online when its a server full of good racers, but its nice to have the A.I. as realistic as possible when you are offline. Whats your take on all this? Cheers!
Simulation_Player
Registered
ai being more planted on contact that is bit tricky to say if it is either realistic or not, unless u have similar situations IRL to cross check.
regarding wet surface grip.......maybe ai is too strong for the settings u choose. I would suggest not bothering with how ai behaves on perticular cases , just tune them for your overall laptimes, for me it works.
my only gripe is regarding the fastest path they choose, they take too much kerbs and spin out on some tracks. but this can be fixed by aiw.
regarding flags & rules ai , i turn that off completely because i like to do burnouts in pits and i hate slow pit limiters lol.
Your implication and logic has a little flaw. You don't know exactly how much of the cpu load is used for the physics or simply to run the game. UE4 is pretty well known for stressing the CPU even without running high end physics.
Paul McC
Registered
Yeah I already stated that "I don't know, maybe its just the nature of the Unreal Engine, I guess it could just be that its not the indiviual cars soak up all the CPU, maybe its just the engine itself."Your implication and logic has a little flaw. You don't know exactly how much of the cpu load is used for the physics or simply to run the game. UE4 is pretty well known for stressing the CPU even without running high end physics.
AlexHeuskat
Registered
With the Unreal Engine, it uses more the CPU cause all is about resolution scale, the game is only 13.8 gb, cause it uses the CPU for the resolution scale every time....that's why it's so beauty at a very high resolution (4k or more) and blurry at 1080p...But it pushes very hard the CPU, not only the GPU.
davehenrie
Registered
That explains why rF2 looks better than ACC at lower resolutions. I had to massively upgrade my old system to truly see ACC clearly.
memoNo1
Registered
To be honest, I don't know enough about the subject matter to have a say in it.