brightcandle

He is running Arma 2 for the big coop games. Arma 3 is limited to much smaller games and if you look its not doing 30 fps reliably either.

Its not really sudden, each successive patch has reduced performance. That is pretty normal performance nowadays with an overclocked CPU. I wont go into the what and why but its a problem most of us have in even surprisingly small scenarios (10 players 20 AI). Play on Arma 2 maps and you'll get better FPS, for now I would say 70% of altis is too low performance to be worth playing.

I haven't seen anyone test Arma 3 on a Haswell and compare it to Sandy Bridge. But in a lot of other CPU driven applications the upgrade can be worth about 20% performance improvement at roughly the same clock speed, and in theory at least that should translate into Arma 3 frame rate. I suspect but can't prove that Arma 3 would disproportionately benefit from the increased ports count due to the way the game calculates heavily in certain parts of the frame. In addition extra cores also help because some of the rendering is done in parallel and that can be worth a boost in frame rate as well. But the game is mostly single threaded, hits heavily on memory bandwidth and on calculation performance. So in my mind the ideal CPU for Arma 3 is probably a 6 core 12 thread Haswell-E like the 5820k and then overclock it as far as it will go, ideally around the 4.4Ghz or more. A reasonable compromise is a 4 core 4 thread like the 4690k which is also overclocked as far as you can get it. You can probably get within 20% of the frame rate of the 4690k overclocked with a 2600k overclocked to 4.5Ghz, and some 2600k's will go higher. You will probably still be 30% off of what the 6 core CPU can do and even with the higher clock speed the 2600k might overclock to you wont close the entire gap. So you don't need to replace your CPU, all depends how much the potential 20-30% is worth to you and what its going to take to get it, but the basic rule for Arma 3 is overclock your CPU. The main issue is that by my measurements if you wanted to sustain 30 fps in all the game modes out there commonly played on multiplayer servers (like Altis life, wasteland etc) then you would need an 8Ghz haswell. That obviously isn't possible so alas getting 30 fps continuously is not a reality with this game as it stands. But you can get decent frame rates with overclocked Sandy Bridge and above CPUs if you stick to smaller player number less script heavy games.

Games designed and written on DX12 sure, those that weren't aren't likely to see much of a boost. Arma 3 wont see any useful boost at all when its ported due to the bottlenecks the profiler pointed out as being elsewhere.

DX12 isn't going to fix the simulation that will have to continue to run single threaded because that is how its written and how all the scripts are written. Since this can get to 30+ms on its own that right there is a limiter to 30 fps. Then about 50% of the frame goes to code around rendering, but as we have already determined Arma doesn't spend much time in the actual API calls, instead it spends most of it in its own code. The multithreading we see in the rendering part appears to be only marginally useful and BI certainly aren't using DX11 to anywhere near its peak, draw calls of 1500-2000 are more like 2003 levels not modern day games (its actually a little bit impressive the game looks as it does with that few calls, but perhaps its due to a lot of CPU calculation). So my theory, which I can't yet prove is that the rendr activity is doing pre API render work, which right now is not parallel and is one of the dominant factors in the time for the game. There is other stuff in there as well, surprisingly asset loading is non negligible as is client side AI and PhysX, all of which run in the main thread. But really if we are going to point at a big issue its the lack of multithreading in the Arma code, in its engine and everything else. This is mostly a single threaded game. So unless they fundamentally rewrite the simulation and the rendering for DX12 (extremely unlikely since they have touched neither appreciably in 2+ years) it would seem its unlikely to bring any real benefit because the draw call overhead is a relatively small part of Arma execution time per frame, the profiler and GPUView data tells us that for certain. There are a lot of ifs about how they implement things for DX12, which incidentally would also help DX11 as well by basically the same amount. We can hope but I am not hopeful because mostly the performance binary remains focussed on improving server side performance, like client side doesn't matter at all. Yet all the evidence points to client side performance being a major issue in very small (10 person with 30 AI) games and a massive issue in larger ones.

One thing the profile data showed me was that asset loading is done on each frame right at the end after its rendered, on the main thread. We all know about the ExThreads setting so it was a bit of a surprise to me to find a couple of milliseconds disappear into asset preloading on the main thread but there it was anyway. I suspect an SSD really helps with that particular part and since its right there in the main thread which is the bottleneck for the entire game its going to have an impact. As far as I can tell multiplayer seems to hit the FPS a bit in two distinct ways. The first is that the world sim time goes up. This is most likely a combination of script running time but its also likely the inclusion of other human players. The second thing that increases is rendering time and my theory there is its associated with the player models themselves, due to the high customisation of the models I suspect we get a lot of CPU time generating those models every frame from the kit available and that causes a reduction in performance. Oddly enough it doesn't seem to matter if you are looking at them or not, just whether they are within your object view distance.

Based on the actual profiling data I have presented I am surprised people still buy the garbage of fixing performance when we know this just wont help. Its not like we don't know what the issue is, we know on a grand level where all the CPU time goes, why the GPU usage is low. We can't fix it but we can see that DX12 will have only a minor effect on this particular game unless they drastically change the way they render. The amount of times we have been lied to about performance fixes and people still believe these guys when they tell you that the next fix will bring what you want, even though you have information, genuine actual science showing its not the case. I will never understand.

At a normal desktop viewing distance of a few feet the current estimate is that we will exceed the eyes ability to distinguish the screen from reality (in resolution terms anyway) around 12,000 vertical pixels (where current 4k is 2160p). It may be somewhere up to about double that or it might end up a little less but that is how good the human eye is. We can currently tell we are looking at a screen basically and will be able to for some time. More pressing is probably the very limited colours screens can produce from the visible spectrum as well as how accurately they are represented, but all these aspects need improvement. 4K and 1440p are hard to drive in Arma 3 and one of the reasons for it is not really GPU horse power, its that the CPU is involved and is impacted by the resolution. Most games don't produce much in the very of graphics using the CPU but Arma 3 appears to do so. Its likely responsible for much of the problem people currently have with the game and its definitely an issue scaling the game up in terms of pixel density. I have played at 1440p and done so with 2x 970's and basically I will always have a lower frame rate than someone who is running at 1080p regardless of settings because of the way the game is written.

They did? Where did they say that? I see a lot of talk about the server binary performance improvements but not client performance.

I have answered it as well as can be answered from outside the dev team: http://www.reddit.com/r/arma/comments/303xr4/understanding_arma_3_performance_problems/

I use a pair of Sennheiser HD 598s on a Soundblaster ZX soundcard. The key point however is the setup of the card and how it simulates 5.1 surround sound. So I also have SBX Pro on and thus in the game (and other games) they detect or are set to 5.1 speakers and the headphones place the sounds virtually. Gives me reasonable perception of distance and direction, something that DX basic headphone mode doesn't give me very well. So if the question is whether a soundcard is worth it for me at least I consider the surround sound algorithms to be the main selling point of the soundblaster and Xonar, not really the DAC capabilities.

Unfortunately your theory is completely and totally wrong, the facts and data couldn't disagree more.

While the answer is just no (mostly) the reason for it is due to the way Arma is designed and runs and its not so much complex as potentially counter intuitive. I can say as much as I like that the profile and GPUView data I produced some months ago shows exactly why all this is the case but a lot of people even with my explanations don't understand the diagrams because they don't regularly work with profilers and debuggers because they aren't programmers. I don't personally think the answer is just no, its no most of the time and its no when frame rate is at its worst. But its not always no because there are scenes in Arma that are GPU limited.

The GPU is busy rendering the frame in parallel with the CPU. There is quite a sizeable period where the CPU is talking to the GPU (rendering) and while most of this is Arma code there is some time in DirectX and some time in the drivers. Presumably a small amount of this time must be talking to the GPU and maybe if its overclocked that is quicker but in my actual practical experience it makes literally 0 fps difference, overclocking didn't get me 1 measly fps.

Arma doesn't run its CPU activities in parallel with the GPU engine ones, they are serial. Once the commands are passed off to the GPU it then renders and the game continues on with its CPU processing. As far as I know overclocking the CPU does not improve the CPU time that talking to the GPU takes and since the GPU is always under utilised it will finish the work faster (less latency) but because the GPU is idle waiting on the CPU to produce the next frame the frame rate will not increase.

What we do is carry a single AT4 in a fireteam and we also carry one in our special forces team as well. We don't use Javelins as we don't do dedicated heavy weapons teams. You will find the milsim groups based on specific units will copy that units layout, so they will have dedicated Javelin teams. Our layout is based on these real units without being specific so we can do things like have AT4s in all fireteams where a specific unit wouldn't carry that many. Neitherless its basically 1 throw away launcher for 4-6 men more or less and Javelins are dedicated anti tank teams only.

There are places in the game where you can get limited by the GPU, I don't want anyone thinking that isn't the case. But you have to go somewhere where there is a lot of vegetation and have high AA settings and then you will max out the GPU. Performance will be in the region of 60 fps on a decent GPU, and you could by really pushing the settings up get it down into the 45 fps range. Absolutely doable. Of course just changing the settings will immediately fix the situation and return you to a height of more like 70 fps. The issue is the cases where the game drops down to 25 fps or even 15 fps like I had on one altis life game with 20 players in it. Those instances are not GPU limited and no amount of changing the settings can fix the situation. The major problem with performance is CPU related, it doesn't mean that at times the game can't become GPU limited, it absolutely can, its just the GPU limited areas aren't the problem, they can be fixed with reducing the settings and they are no where near as severe as the problems with the CPU where the game will become unplayable and no changing of settings can ever fix it.

BI continues to pursue performance patches around improving the amount of players a dedicated server can handle. They long since abandoned even talking about improving client side performance, they haven't done anything but make it worse since retail. Its ironic that we are hear because there was a point in the middle of alpha where performance was genuinely improved and it felt really good and I remember playing quite a lot of PvP as a result, we had a very popular mission at that time running on a public server. But then they broke it again and its been this way ever since. I have explained what is actually happening technically many times but I think the politics of what is going on needs to keep being discussed otherwise the complaints about performance will fade away and they wont ever start to address the client side performance issues apparent to everyone that plays this game. There is simply no evidence BI is working on this problem, I can find no improvements in the performance builds because it all seems to be focussed on network performance, something we don't have an issue with.

As all the others have mentioned the answer is no it wont. Its simply not the limit in the game. You want a faster CPU and ideally if you want to play at 30 fps on pretty much everything out there you need an Intel CPU at about 8Ghz. So even liquid Nitrogen cooling isn't enough to run the game to the minimum standard, you would need a world record overclock to get 30 fps in most public servers.

They are aware of the performance problem, but they haven't taken any steps to address it. All this time they have been addressing the dedicated server performance issues with network traffic and they haven't actually even started addressing the very real problem that the games CPU usage on the client is most of the problem. Its not a quick fix, its going to take a good long while to repair this but unless they actually start we can't count it down. There is no evidence they have started so I think its right that people should keep on about it because unless we express our opinions as their customers that the current situation isn't good enough they wont do anything about it. We have to constructive, we have to show them profile results and other performance tests of course but letting up on the issue is not the right thing to do, because right now they seem focussed on fixing something that isn't a problem for us.

Performance has been steadily dropping since Alpha. The issue is entirely in the client, I can find no evidence the problem is with network traffic or network performance. So part of the problem is while BI fix the areas you don't have an issue (and most of us don't as its not a pressing concern) the main game performance just sucks. You can get performance below 20 fps just walking around an empty map, you don't need AI or other players even, its trivial to make the game perform badly on Altis.

The data basically says that players don't consider mods worth buying, they certainly don't donate very often. The situation where you could buy a mod and then a day later not have it working and have no recourse (this is the situation right now for my group, we have key mods that don't work on 1.42 and we are stuck on 1.40 until a replacement comes out) means that you wont find me buying into that proposition. The basic problem of versioinng in this can't be solved, the companies are never going to take responsibility for every mod and its future compatibility and the modders are never going to be able to reimplement it on every patch instantly. The end result of which means real money exchanging hands combined with consumer protection laws makes it untenable. They thus have to remain unpaid for because no one can comply with consumer protection laws, and this situation otherwise does involve the customer getting ripped off. I have already explained the situation we find ourselves in with Dwarden, they intend to release 1.44 imminently and remove the 1.40 branch in full knowledge that will harm our community. If I was paying for that situation I would be really angry, but because the mods are free and they don't work and we have to fix our own bugs and such its not so bad. I will still be angry with BI, they ultimately know the harm they are causing and don't care but at least its not also a mod maker in that mix. I get the community wants to get paid for the effort, I just don't see how it can work given the current attitude of developers towards mods and the limited resources they and the modding teams have. Its a core fundamental problem of service and its more or less impossible to fix. Linux distros release every 6 months or so instead of continuously despite all the projects they depend on releasing continuously. The reason is that the interdependencies between projects and updates have be given time to filter through all the software, bugs to be fixed etc. This is why distros are regularly out of date on one particular piece of software by half a year or maybe more, because either a new version of a library it depends on isn't ready in time or stable enough for other purposes. Mods are in the exact same situation of being dependent on a black box project getting updates when the developers feel like it with no advance knowledge or access to early code and completely unable to see what will break and how until its released. So regardless of what happens the mod is going to be broken at launch of the new game patch. This is the same in Arma despite the development branch because the developers don't show us the final release version, they go away and add more stuff before we get it. Its a bad development practice to do it but BI has always worked this way since the development branch was introduced. You develop against what is in development at your own risk and its quite often a big risk as things change sometimes quite a lot between initial introduction and final patch. This is just not easy to fix. The reason its free is because actually making this a process that produces mods that always work for the customer is impossible today and it looks unlikely the development process from BI (or any other developer) is likely to change to avoid that situation. I am no even sure its possible without constant access to the full source code and the mods becoming something that forms part of the codebase. These aren't possible so it can't be fixed hence modders can't reasonably charge for a product that tomorrow might never work again.

The main issue with interclan games has always been mods. Every group plays with a different set and version, many have customised them further to fix problems or incompatibilities. Thus the end result is that between clan operations are extremely high effort. Its not just the maps, missions its also slot up and everything that goes into it. They are not easy to setup and I can't say having organised quite a few I ever want to do them again, its just too much effort.

A reduction in the view distance setting is reduces the CPU processing per frame substantially. It doesn't appear from my testing that the constant culling time is an impact. I suspect the reason is the substantial CPU overhead per object rendered within the engine dominates over the culling algorithm. Despite the volume of entities culling is still cheaper than rendering them, purely on a CPU basis. Right now that culling process is not a substantial impact (although I doubt its negligible, its just constant) but increase the object count very well make it a bigger problem.

In my experience with games from 8 to 80 people you see issues with literally just one player. The game may very well be server limited at 50+ players, but the problem starts at just a couple of players in a small mission, its completely client side.