AI kills the performance completely (instructions how to test inside)

[aop 1]

Roughly 20 AI per side, so over 40 in total, as you would see if you read the post you just quoted. "Airfield" doesn't mean "runway". I spawned them positioned in groups around buildings at opposite ends of the airfield so the two forces would work their way towards each other across the base, rather than just have two ranks open fire on each other (that would be boring and pretty short).

Take a chill pill. Someone saying "I'm not sure this is right, but I could be wrong" is not some kind of personal attack. If you don't want other people to share their thoughts and experiences in an attempt to help then don't start a thread.

You are not helping anyone. You refuse to run the test described in the first post and yet you state that the AI isn't the problem.

What is causing so big disparity between ArmA 2 and ArmA 3 performance in same map when multiple characters are involved? Higher polygon models?

[Fuse 1]

You are not helping anyone. You refuse to run the test described in the first post and yet you state that the AI isn't the problem.

What is causing so big disparity between ArmA 2 and ArmA 3 performance in same map when multiple characters are involved? Higher polygon models?

Please actually read what is written, rather than skimming it and making assumptions. As I've said multiple times, I never said "AI isn't the problem." I said "Maybe AI isn't the problem, but I could be wrong", and explained why I thought that.

What is causing the disparity between running the same map and scenario in two different versions of the engine? Really? Just think about that for a moment.

I'm washing my hands of this. I came to contribute my experiences to the discussion, not fight because someone said I might be incorrect.

[aop 1]

Here is another quick test with FRAPS, 1 minute benchmark.

Watching Agia Marina from Southern hill, no AI:

Min fps: 58

Avg fps: 62.517

Max fps: 64

Exactly the same place, 3 OPFOR and 3 BLUFOR squads in town:

Min fps: 30

Avg fps: 33.600

Max fps: 38

Seems like AI is eating all my fps to me.

edit:

Mission setup:

http://cloud-2.steampowered.com/ugc/882976114553809006/3CFBA4DB4C83B2CD576067036DF13442E7B69A3F/

For first test I just had no AI.

Edited March 10, 2013 by aop

[Iroquois Pliskin 0]

AI & physics, aop. Without PhysX you'd probably take a minimal hit, comparable to ArmA II performance.

Run a test with 3 vs 3 squads but with 2 additional vehicles per side to account for the physics simulation, since AFAIK infantry ragdolls become active only during the moment of death.

[aop 1]

Same test with 2 HMG cars for both sides:

Min fps: 28

Avg fps: 32.367

Max fps: 40

[Iroquois Pliskin 0]

Three squads that's 15 inf, right? Try replacing the infantry with 7 APCs per side.

[aop 1]

Three squads that's 15 inf, right? Try replacing the infantry with 7 APCs per side.

Three squads is 30 infantry, each squad has 10.

---------- Post added at 10:53 PM ---------- Previous post was at 10:47 PM ----------

15 HMG cars on both sides (30 AI on both sides, 2 in each car):

Min fps: 24

Avg fps: 29.417

Max fps: 33

[Iroquois Pliskin 0]

15 HMG cars on both sides (30 AI on both sides, 2 in each car):

Min fps: 24

Avg fps: 29.417

Max fps: 33

Well, vehicles lower the FPS even further: max & avgs have gone down. Definitely need multi-core optimisations for AI+physics this time, otherwise Altis will be unplayable even on XM i7s.

[-Coulum- 35]

So is it pretty much confirmed that the arma 3 ai cause more of a slow down than arma 2 ai did?

[Iroquois Pliskin 0]

So is it pretty much confirmed that the arma 3 ai cause more of a slow down than arma 2 ai did?

Only due to physics calc that is generated by their activity. Spawn 50 empty cars in motion and you'll get the same FPS drop as 25 cars with AI in them.

[white 1]

How do you figure that it's AI that is unoptimised?

Haven't you people considered that the added strain of PhysX calculations on every tangible entity in your missions is the culprit, even if AI optimisation stays roughly the same as it was in ArmA II? (Which I suspect to be the case w/ AI)

Since multi-threading isn't exactly stellar in ArmA series, you will have to overclock your CPUs to 5 GHz, if you want to have anything more than a few squads running around.

End of story, until/if devs post their thoughts on this.

some guy on steam discussion (here http://steamcommunity.com/app/107410/discussions/0/864961721676462825/ ) used an api monitor to show how most of arma 3 including its AI runs in 1 thread in 1 core, while some stuff like clouds, and physx spread across other cores, but the first core bottlenecks also waiting for them. this game does not uses more than 2 cores total. thats been shown by disabling extra cores and seeing no changes whatsoever in fps, but they get 90%+ usage. and also, people said that the effective way would be each AI to create a new cpu thread that would be spread across avaiable cores but that would also need some core changes in arma code.

if thats all true, i hope BIS make some extra effort and do them no matter what. but the thing is that even on games without AI the game gets unplayable fps because of the 1 core/thread limitation.

and you are right, the one thing that increases performance is overclocking as high as you can those 2 cores, thus being way better on intel which has superior per core performance.

Edited March 10, 2013 by white

[Sam75 0]

So I guess you better turn off HT.

[Iroquois Pliskin 0]

some guy on steam discussion (here http://steamcommunity.com/app/107410/discussions/0/864961721676462825/ )

Great to see 44 pages on this topic on Steam forums - awareness is good.

[EBass 0]

I think this has to be regarded as the single biggest issue at the moment. We (Almost) all play Arma to a large degree because of the massive epic battles it can support accross large ground. At current build A3 just can't do even really medium size battles. When Arma 2 was released a decent PC (not even top of the line) could play a battle with 1500 AI, hopefully this will be possible or almost possible when A3 comes out!

[Touch Off 10]

I really hope and believe BIS will manage to focus on this and know where they're going wrong.

I have trust in them. The same guys who bought flashy to me 12 or so years ago won't want to see this game fail. I could be cynical and believe that they released this game as alpha to make money early on, knowing that they've got a big problem with the code they've used for this new engine but, I really don't believe they've got it that wrong. I hope that they truly want the support of the community to resolve issues and create a game, somewhat in partnership with their fans, that is the best they've ever accomplished.

It already looks very good, feels nice at times and when fighting with a few AI after sorting out an adequate setting of eye candy that still feels good but offers stable/high fps on my particular PC, you can tell there are some good if not great improvements.

Are there enough? It's difficult to say, obviously but, patience is a virtue and without the frustration of AI walking through walls and the much lamented poor performance with MP (which I haven't tested) or SP then it is difficult to see the bigger picture, if such a picture exists, of course. I really do hope it isn't just a better looking game with the same problems as ever.

I think it's unfortunate that there are those who played early on who possibly started instantly thinking 'wow, this is good enough for a release candidate' as it obviously has muddied the waters with regards to the overall consensus and in my opinion is based on how the game looked without the awareness of the series' history.

Being somewhat pessimistic as to whether BIS have sorted out some annoying aspects of their previous titles allowed me to just question what it was they've done, at times a little too despondently.

When it comes down to it, the frustrating 'it's alpha' response, although tiring to read and pointless when discussing possible issues with the game based on previous experience, is still valid. I don't believe anyone here forgets that.

It is, of course, not finished. BIS, I'm sure would agree, would also say it's nothing up to the standards they'd want to us to believe is their finished product...yet.

Like Eagle Dynamics, and although the Black Shark1/2 element was a dodgy part of their business model, they still show us that with every update there's progress and every issue that's discussed, they are aware of and look to resolve. I'm sure this takes time to do well and having patience for their work is a necessity but, not a hopeless endeavour!

I do believe they (BIS) value our feedback and do have hope that they not only want to make money and be successful as a business but also achieve something with integrity. I don't think 'take the money run' applies.

That said, however, I knew I would only spend £20 on this game instead of the more expensive options because I for one do not want to see the same issues that held ARMA2 back and ultimately saw me venturing elsewhere to enjoy gaming/simming on the PC.

If these issues are still there upon release, I'm moving on. If BIS manage to get this bug sorted out soon, by Beta release or soon thereafter, then I'll be a very happy customer because the game has so much potential. I just hope BIS and the team involved want to see their plan to create a cutting edge military simulator come to fruition and want to show us what they're capable of. Their market has grown and they must know that their customers aren't all building the latest systems to play their game. With that in mind I hope that the optimisation, if and when it comes, is evidence of a refined and polished title that is accessible to many people whose PCs are within the specifications of the game but also shows how things have progressed with BIS's vision of their ARMA series venture.

Good luck! I hope you manage it!