Pyrophosphate

Particles are almost the only things that actually run faster on a GPU. That's why games that use GPU PhysX basically make a big particle mess and that's it. Any kind of actual physics intended to be used in the game requires that data travel both ways over the PCI lanes, and the incoming data then needs to be reintegrated into the game state. That all takes more time than simply doing the work on the CPU. Ballistics (especially), simple vehicle physics, and ragdolls are all faster on the CPU, even with the rest of the game running next to it.

Does Arma model any sort of armor angling?

I would direct you to read your own signature. I wonder why the devs (any devs, not necessarily Arma devs) would choose PhysX when they don't use the GPU-accelerated stuff. What's the benefit over any other physics engine?

The PhysX API is installed alongside the game, as it is with any other game on Steam that uses it. The difference between the API on Nvidia hardware and AMD hardware is that with Nvidia, certain effects are GPU accelerated. Not all effects, and probably not even most effects. It sounds like you've been getting your information from Nvidia promotional materials. PhysX doesn't do anything unique. At all. Except heavily and purposefully favor Nvidia hardware. PhysX is software. It is a physics API and nothing more. Bullet3D is also a physics API, one that also (magically) can use the GPU for physics. It also (more magically) does that through OpenCL, rather than CUDA, so it runs on all hardware. You're right, that's not true. Arma 3 uses PhysX for vehicle physics and ragdolls. Maybe particles, but I haven't seen any indication that they use PhysX. Water surfaces have no complex physics in this game. the clouds are the same clouds from TOH, before PhysX was implemented. Of those things that do or might use PhysX, only particle effects are worth bothering the GPU with, and since it has been confirmed by the devs that the GPU is not (and will not be) used for PhysX, I tend to think particles are not using PhysX.

I'm not sure you understood my post, we're on the same side here. I don't think anyone is calling for something other than an informal "run x benchmark on y settings", just for comparison. Nobody is trying to make the benchmark force settings on the user or anything like that. Just an informal standard, for this thread and others where people are talking about relative performance between systems. That said, with the low settings preset, I got ~8200. On ultra, I get ~2900. i7 2600k at 4.5ghz (hyper-threading on) 8gb 1600mhz RAM Radeon HD 7870 2gb at 1050/1200 Installed on Crucial M4 128gb SSD

Yes, if you have not done so already, overclock your CPU. That's the best way to improve performance in Arma. For Arma and pretty much any other game out right now, 8gb of RAM is functionally identical to 16gb of RAM, and it was shown several (dozen?) pages back that memory clock speed has virtually no effect on the game's performance. And how do you figure Touch Off's 670 is weaker than your 670? Also, with what settings did you manage 7000+? With an i7 2600k at 4.5ghz, 8gb 1600mhz RAM and a 7870 at 1050/1200, I get between ~2900 (ultra preset + vsync) and ~8200 (low preset).

Having standardized settings for use in this thread and for comparing rigs in no way diminishes someone's ability to use the benchmark to fine-tune their own settings.

The devs won't fix it? It's not some switch they need to flip and suddenly framerates double. Clearly it's a very complicated issue, and clearly they're aware of it. It was implied, if not stated outright, by Dwarden that they know what the actual problem is and intend to fix it. But, it takes time. The game engine is much bigger and more complex than any other popular game engine on the market. Also, you don't "update the utilization". That doesn't even mean anything. There's a threading issue going on here, not a utilization issue. You don't want it to be using more CPU time.

There's a lot of the same stuff kinda just being repeated in this thread. CPU utilization is low on pretty much any modern processor, and GPU usage isn't much better, most of the time. We get that. It appears (to me) that this game spends too much time waiting around for data from memory, probably because there are so many things that have to be updated each frame. If memory bottlenecking is the problem, then I don't think there will be much improvement in performance over the next couple years as computers get more powerful, for the same reason Arma 2 performance didn't grow in line with hardware performance. If anybody has the resources, test the game with the same hardware, but switch the memory out for stuff with different clock speeds and/or CAS latencies. See what happens.

Nevermind, delete please.

http://www.bistudio.com/english/company/developers-blog/91-real-virtuality-going-multicore I don't know if you've read that. It talks (a bit) about the threading capability of the rendering process.

That's simply not the case for most people. Under certain (fairly common) circumstances, almost everybody seems to get roughly the same framerate, regardless of hardware or settings. Many people around here, myself included, have recently built computers mostly for Arma 3, and yet still has the same problem. That said, reapplying thermal paste every once in a while couldn't hurt anything, but that's definitely not the solution to the large scale performance issues people have with this game.

There is no reason to use GPU PhysX. The only physics work involved is simple vehicle dynamics, ragdolls, and maybe some particles. The ballistic physics system is almost certainly the same system BIS has spent a decade developing. Vehicles and ragdolls run perfectly well on the CPU. GPU physics wouldn't actually help much, if at all.

I don't think you understand computer architecture. There's more to memory than how much you have. No memory in the world can keep up with your CPU. No SSD in the world can even almost keep up with your CPU. When your CPU requests some piece of data from memory, if it doesn't immediately have something else already there to do, it has to wait ~15-30 CPU cycles to actually get that memory. If the game tries to interface with the memory too much, that will both slow it down and reduce the apparent CPU usage, which happens to be what people are saying is the problem.

CPU usage as a percent doesn't mean what most people think it means, and as reported by Windows, doesn't mean much of anything at all. The game detects how many cores it has to work with and balances the workload accordingly. Windows moving an entire running thread from one core to another is not Windows doing its job. That is a performance issue. It slows things down. Name any complex application or game that reliably uses multiple cores at anywhere near 100% usage. Prime95 is not a complex application, neither is Bitcoin mining. Neither is any kind of benchmark. If something is using multiple cores at near 100% usage, it's a program designed specifically to do that.

Those stats are from a program called MSI Afterburner, not from the game itself. It's worth noting that there is no totally reliable way to measure CPU usage, at least partly because the CPU itself is doing the measuring. It's also worth noting that there are potential bottlenecks other than the CPU and GPU, even though that's what most people focus on. In any non-trivial program (or a program like Prime95 that is specifically designed to get as much work out of the CPU as realistically possible), the processor spends a lot of time just waiting around for things like thread locks, timers, data from RAM (as mentioned), data to and from the GPU, hard drive, etc. They don't have to wait long (anywhere from 10-20 clock cycles to a few milliseconds), but that time adds up. That can't easily be avoided because the CPU is so much faster than other parts of the system. So if everybody is hitting a 25-30 fps limit in some area with different CPU and GPU combinations, try looking at memory speeds, see if that makes a difference. Trying to pinpoint exactly where different bottlenecks lie could be helpful to the developers for knowing what it is they're trying to optimize. Also, not saying anyone in particular around here thinks this way, but it's sort of implied a lot (especially on the steam forum once DayZ got popular). There is no magical "optimize" button. The developers do not just sit down and "optimize" the game, and the game does not hit these bottlenecks because they, for whatever reason, chose not to sit down and optimize it. There are certain kinds of optimizations that are relatively obvious and painless to implement, like not drawing things that are behind the player, but those are few, and any game you've ever heard of already uses them, and the Arma engine has been using them for over a decade, I can almost guarantee that. Basically, saying the game "isn't optimized" doesn't help anybody with anything.

Two hotfixes actually.:) With hyper-threading, Windows tries to load up all of your physical cores before it fills up the extra logical cores. The hotfixes just do the same thing for AMD processors, which have 8 physical cores, but they're in pairs that share L2 cache and floating-point units, if I'm remembering that correctly. I could see how A, not doing that could choke an otherwise well running game, and B, Intel might have paid up some cash to keep those patches from being distributed as updates, but that's a cynical view, and this isn't the place. Summary: it could happen. More on the point, if Arma 3 can make better use of multiple threads, perhaps we i7 users could also get a noticeable boost over our i5 brethren.

Indeed I have. (Arma 2 is shown at 7:14)The folks who made that video applied some hotfixes made by Microsoft but never released as an official update which supposedly makes task scheduling for AMD processors as smart as it is for Intel CPUs with hyper-threading. That's what they think makes up the difference in performance between their video and a lot of the other benchmarks out there.

I've seen benchmarks where the FX 8350 absolutely crushes the 3570k at Arma 2; just about double the framerate.

I think the problem people have is that there are so many variables involved in the performance. You get 50-60 fps, but on what settings/view distance? How many AIs running around? Which map, and where?