Beta isn't actually that far away

[almanzo 144]

Not necessarily. With my 4.6Ghz i7 I'm seeing diminished CPU usage and subsequently GPU usage in multiplayer. In single player my CPU usage is much higher and my GPU is able to be capped and pegged at 99% usage. I'm sincerely hoping that there's some bad bugs in the network and server code that need to be addressed still, because multiplayer performance is abysmal even on my high end machine. I am aware my GPU isn't exactly a GTX 680 but it still performs damn well and I throttle back my graphics settings to Standard to counter for this.

My fps in single player is usually 100+ at 1920x1200 and my game visually looks nearly as good as anyone elses. The problem is the online code and perhaps the hardware running the servers. It needs some serious tuning.

Remember one thing, linux dedicated stand alone server release will fix alot of this. Not everything, but server performance will gain alot from being stand alone

[DaRkL3AD3R 1]

Remember one thing, linux dedicated stand alone server release will fix alot of this. Not everything, but server performance will gain alot from being stand alone

What's the current server implementation? The dedicated servers running alpha I mean. I know that a command line based dedicated server will yield much greater results than say a listen server would.

I really hope you're right though. It makes me very optimistic of final product performance but I just don't know if I can believe it :(

[white 1]

Can I ask what your CPU is? And are any of your CPU clocks running at 100% while your GPU is at 45-55%?

---------- Post added at 20:15 ---------- Previous post was at 20:11 ----------

If you play on a fresh MP server with only a few players I am fairly sure you will get similar FPS to your SP performance. There is a major issue with crappy server hardware - exacerbated on an 'old' server in persistent game modes like Wasteland with a lot of players - you need a BEAST server to run this smooth. This has nothing to do with BIS its a simple law of computing - the server needs to calculate HEAPS!

Its on my sig, AMD Phenom II X6 (Thuban) @3750. first core is always higher, the sum of all core togheters get cpu usage to around 32-34% which is the same as 2 full cores in a 6 core. (thats when using all 6 cores obviously)

With only 2 cores avaiable it does spike both cores to above 97%, but of course it doenst stay that way and barely touches 100%. the game syncs all threads with the main thread, including those dozens of physics and clouds threads handled by the other cores, so even when using 6 cores the first core has to wait for those while at the same time trying to handle everything else alone at the first core, and for a very small moment it stops and wait. That can be seen using an api monitor, and was shown by a user named "Homer Simpson" to Dwarden with screenshots on a Steam discussion topic with a lot of information which Dwarden closed.

i rarely saw my gpu spike to 60%, even with everything on ultra and with AA on, and ive seen it go as low as 15% during multiplayer, no matter what launch settings, how many cores i let avaiable, etc. CPU bottleneck on the very first core is the reason that i can turn on AA with no fps impact whatsoever. on any other game turning on AA equates do losing a lot of fps since it takes a lot of gpu power.

So on our side if you already have spare memory, ssd and a good gpu, the onyl thing you can do is buy a better cpu or overclock as high as you can.

Multihtreading can be done properly, the issue is that this engine doesnt. The game support multicore but doesnt use more than 2 cores properly. you can find benchmarks with dualcores handling the game the same as 8 cores, and im not talking about intel vs amd, im talking amd vs amd.

About dayz. well i remember when rocket was still only working on the mod and said something like how the animations and how the AI sync with the server were the major performance issues that prevented the game from habving more zombies, and they were the things he couldnt fix in dayz mod since they were engine bound, or something to that effect. and the things they claimed to have changed are exactly that, and have stated that the game fares better now with 1000 zombies than it did with 200 on the mod. i might be a little off with the numebrs but you get the point.

i think its a lot easier to get an i7 overclocked to 5ghz as a server handling 1000 zombies than all 20 ppl playing with clients like that with all of them processing those same 1000 zombies and having to sync them all between themselves, in order for noone bog down the entire server with all the syncing that still happens in arma 3.

Edited April 17, 2013 by white

[sproyd 2]

Saying you have 32% CPU utilisation when one of your cores is at 100% isn't strictly correct. In the ideal world we'd just have one core that was 12.0GHz instead of 4 @ 3.0GHz but that's not the case and they don't equate to each other, we have multiple cores due to the laws of physics.

I accept that BIS could do a lot to improve multi-core performance.

However I don't accept everyone saying "OPTYMYIYIYZZZEEEEEEEEEEEEEEEEE NOOOWWW" when BIS have already done such a stellar job on the opty from A2 to A3. If the price we pay for such an awesome game is poor multicore optimisation then we'll have to adapt and get beefier clock speeds.

[MavericK96 0]

Saying you have 32% CPU utilisation when one of your cores is at 100% isn't strictly correct. In the ideal world we'd just have one core that was 12.0GHz instead of 4 @ 3.0GHz but that's not the case and they don't equate to each other, we have multiple cores due to the laws of physics.

I accept that BIS could do a lot to improve multi-core performance.

However I don't accept everyone saying "OPTYMYIYIYZZZEEEEEEEEEEEEEEEEE NOOOWWW" when BIS have already done such a stellar job on the opty from A2 to A3. If the price we pay for such an awesome game is poor multicore optimisation then we'll have to adapt and get beefier clock speeds.

...But as you say, you can only push speeds so far.

Multi-core, multi-threading is the reality of the future. ArmA 3 needs to utilize this fully.

[white 1]

Saying you have 32% CPU utilisation when one of your cores is at 100% isn't strictly correct. In the ideal world we'd just have one core that was 12.0GHz instead of 4 @ 3.0GHz but that's not the case and they don't equate to each other, we have multiple cores due to the laws of physics.

I accept that BIS could do a lot to improve multi-core performance.

However I don't accept everyone saying "OPTYMYIYIYZZZEEEEEEEEEEEEEEEEE NOOOWWW" when BIS have already done such a stellar job on the opty from A2 to A3. If the price we pay for such an awesome game is poor multicore optimisation then we'll have to adapt and get beefier clock speeds.

that doesnt make sense and serves badly as a poor excuse. 2 cores is 33% of a 6 core. spreading the game throughout those 6 cores and getting 33% usage in this case, because of how its coded and how the threads are separated (or not) wields the exact same performance of a dual core with high usage.

my 3d rendering software scales evenly across 6 cores using 100% of them 100% of the time. it works 6 times as fast as 1 core. and i can time it, well render software usually has an embedded timer to begin with.

the problem lies in how games are coded, most games are badly coded for multithreading, but those same games get a pass because they dont need to be multithreaded, they work fine with high fps on 1 ou 2 cores, or whatever. the shit hits the fan when a game needs more processing power to even get to minimum acceptable fps, and for that some games do multithreading beautifully, or at least as much as they need so the cpu wont bottleneck the gpu, and some, like arma, are horrible at it.

the excuse is how arma handles syncing, its high level scripting language and a choice into making AI and other stuff to use only 1 thread, because separating and syncing with the first core might not wield sufficient gains to make up for the effort of recoding the game that way. its a game design choice made on the first game and when the engine was being developed. but the industry is already working at full throttle on new engines and heavily multithreaded games, why? both consoles will be 8 cores with low frequency processors, they have to be heavily optimized for multicores in order to use those processors, and those games are already being made to launch later this year alongside bf4 and everything else, including ArmA 3.

Edited April 17, 2013 by white

[Insanatrix 0]

I have to agree with White, You're not going to see CPU's ever break the 5ghz barrier anytime soon if ever. What you are going to see are lower frequency multicore CPU's in the 3-4ghz range with ever increasing core counts. Really, we have been past the point of 32bit and single threaded applications for quite awhile, developers have just been extremely slow to catch up with technology because it would actually require work instead of reusing the same engine over and over with minor iterations here and there.

With the new console's, I hate to say it but it's going to mark a shift in how games are developed, and it's actually a good thing. The focus is going to be heavily on parallel processing and multi-threading.

@sproyd, By saying that we need beefier clock speeds, are you ignorant of Moore's Law? You're statement is like saying "I want to fly without wings or any kind of lift, so I'll just have to adapt and defy gravity". Fact of the matter is that we are adapting, but we don't adapt around what we want, we have to adapt around what is and isn't possible with our current technology. Our technology is moving towards parallel processed multi-threaded tasks. I guess though by your assumption of reality, we should all have liquid N2 canisters feeding our CPU's.... That would be a great selling point for BI :rolleyes:.

Also no one is saying ZOMMMGGGG NAOOOO!!!!!!!! Most are just asking why this wasn't a more important issue, and why there wasn't more thought put into solving the issue's before this point. Maybe it can't be solved with the current engine, maybe it would take so much work that BI doesn't think that it's worth doing, They haven't exactly been forthcoming about it other than they don't want to go 64bit because it would possibly alienate someone with a Sempron from 2005, one of the last known 32 bit processors I know of, and the engine is multithreaded already. As tests have shown though, both in A2 and A3, it's poorly implemented and ends up bottlenecking itself into pretty much being a single threaded process because all children threads end up waiting on the parent thread to finish.

[dnk 13]

@DNK even though your comment is favourable towards the A3 engine you have to remember that OA is Takistan and Takistan is a VERY basic terrain with minimal objects. Stratis is much more detailed in my opinion and when I run Takistan on A3 (using allinarma mod) it runs MUCH faster than OA.

Yeah, I'm going to be testing that out with AiA today probably. I will expect a much more "optimized" experience.

@ white, Insanatrix

This is not how it's running on my CPU throughout many tests. All 4 of my cores usually get an equal load. Although there is almost certainly some "waiting" issue, I clearly am not seeing it as being an issue with "1 thread ties up 1 core" since none of my cores regularly goes much above 50% and I still get clear CPU-lag with high AI counts doing heavy pathfinding.

[white 1]

Yeah, I'm going to be testing that out with AiA today probably. I will expect a much more "optimized" experience.

@ white, Insanatrix

This is not how it's running on my CPU throughout many tests. All 4 of my cores usually get an equal load. Although there is almost certainly some "waiting" issue, I clearly am not seeing it as being an issue with "1 thread ties up 1 core" since none of my cores regularly goes much above 50% and I still get clear CPU-lag with high AI counts doing heavy pathfinding.

its pretty clear on amd, im not sure why it shows differently on intel. check out this revew/benchmark: (translate to english if needed)

http://gamegpu.ru/action-/-fps-/-tps/arma-iii-alpha-test-gpu/testovaya-chast.html

and now look at the cpu usage in crysis 3:

Edited April 18, 2013 by white

[Insanatrix 0]

Yeah, I'm going to be testing that out with AiA today probably. I will expect a much more "optimized" experience.

@ white, Insanatrix

This is not how it's running on my CPU throughout many tests. All 4 of my cores usually get an equal load. Although there is almost certainly some "waiting" issue, I clearly am not seeing it as being an issue with "1 thread ties up 1 core" since none of my cores regularly goes much above 50% and I still get clear CPU-lag with high AI counts doing heavy pathfinding.

I get a very lopsided load, like 60-70% on core 0 and then 40-30-30 on core 1/2/3.

Also I get lag with high AI counts, but if I set waypoints and then have a trigger that disables All AI parameters and turn them all off, I don't experience an increase in performance between AI on/off, it's a fluctuation of 2-3 fps at most. I use http://community.bistudio.com/wiki/disableAI . It's a larger issue with the resources the AI take up than the pathfinding itself.

[DaRkL3AD3R 1]

I get a very lopsided load, like 60-70% on core 0 and then 40-30-30 on core 1/2/3.

Also I get lag with high AI counts, but if I set waypoints and then have a trigger that disables All AI parameters and turn them all off, I don't experience an increase in performance between AI on/off, it's a fluctuation of 2-3 fps at most. I use http://community.bistudio.com/wiki/disableAI . It's a larger issue with the resources the AI take up than the pathfinding itself.

Well think about why increasing object view distance hurts your framerate. You're adding extra geometry and drawcalls to the scene for the CPU to prepare to render. When you have say 5 AI units in front of you, think about how many more polygons and draw calls there are in one infantry unit vs say a low detail tree sprite off in the distance?

I don't think it really is an AI calculation issue, at least not the major significant issue in having NPC's in a scene, but rather the cost to render them.

[Sneakers O'Toole 2]

I get a very lopsided load, like 60-70% on core 0 and then 40-30-30 on core 1/2/3.

Same here, i5 2500k (stock).

[dnk 13]

Also I get lag with high AI counts, but if I set waypoints and then have a trigger that disables All AI parameters and turn them all off, I don't experience an increase in performance between AI on/off, it's a fluctuation of 2-3 fps at most. I use http://community.bistudio.com/wiki/disableAI . It's a larger issue with the resources the AI take up than the pathfinding itself.

You're right, but you weren't. What I mean is that after doing some testing today with the newest build, I don't have this pathfinding issue anymore... I did have it with an older build, though. Same setup now yields drastically different results in terms of FPS.

The SPOTREP only mentions some minor AI changes... I'm lost for why this change happened. I will stop going on about pathfinding now, at least.

Could be due to:

""Geometry fixing (AI collisions with buildings tweaked)""

My "pathfinding" issue was when I placed the AI in cities and forced them to march through. It used to result in a halving of FPS with 80 AI - now it has maybe a 5% reduction.

Could also be:

""Fixed: AI no longer fires on targets it does not see (but which are reported by other group members)""

But I took away guns for both tests, so I'm not sure how that figures in (though this is a very nice fix anyway).

Edited April 18, 2013 by DNK

[kamaradski 10]

How did this thread got hijacked so badly ?

[comc 1]

When you go back to Arma 2 and look at Utes, Stratis is a god send. It's huge, runs great, looks great, and makes for some awesome engagements. Not to mention as a Wasteland map it's pretty darn cool. I like it.

I'm curious what Altis is going to look like though? I realize it won't be anything like Chernarus, as we're now in the Mediterranean instead of like a Russian influence, but is it going to look like Stratis only bigger?

The cool thing about BI's level design is; just look at Google Maps, and you'll have your answer!

[ric 1]

How did this thread got hijacked so badly ?

because the performance issue with this game is a MAJOR one and everyone is concerned.

---------- Post added at 00:15 ---------- Previous post was at 00:14 ----------

I get a very lopsided load, like 60-70% on core 0 and then 40-30-30 on core 1/2/3.

ditto on my 2500k and qx9650 on my server

[Insanatrix 0]

You're right, but you weren't. What I mean is that after doing some testing today with the newest build, I don't have this pathfinding issue anymore... I did have it with an older build, though. Same setup now yields drastically different results in terms of FPS.

The SPOTREP only mentions some minor AI changes... I'm lost for why this change happened. I will stop going on about pathfinding now, at least.

Could be due to:

""Geometry fixing (AI collisions with buildings tweaked)""

My "pathfinding" issue was when I placed the AI in cities and forced them to march through. It used to result in a halving of FPS with 80 AI - now it has maybe a 5% reduction.

Could also be:

""Fixed: AI no longer fires on targets it does not see (but which are reported by other group members)""

But I took away guns for both tests, so I'm not sure how that figures in (though this is a very nice fix anyway).

I've been using the dev builds since the alpha came out, so I've had those fixes for awhile I believe.

[Guest]

Locked as requested by the topic starter cause so many of you feel the need to derail a thread.