- User Since
- Mar 27 2013, 8:03 PM (489 w, 3 d)
May 10 2016
I would love to see this implemented as it would allow modders and others, myself included to do some really interesting stuff.
Fault time: 2013/12/04 11:34:54
Fault address: 68FB21CB 01:001511CB C:\my\Steam\SteamApps\common\Arma 3\PhysX3_x86.dll
May 9 2016
@GT500 Interesting....Whats your system specs? I am curious as to why you lost frames while most everyone else gained a considerable number.
@GT500 - I agree the game runs no where near were it should be. However there are things that we can do to increase the performance to playable levels.
For starters change that MaxFramesAhead to 0. Granted the difference from 0 to 3 is probably some where in the sub percental range it still something.
Set your launch options to "-cpuCount=120 -exThreads=120 -nointro -nosplash -nobenchmark"
Ensure your vsync is disabled and I would turn off AA. If you'd like you can add me on steam (look a few posts back) and I will help where I can to get you up to playable levels.
In regards to the CPU brand: I can see that, in my previous experiences the Intel chip sets, especially the newer generation of i7's, handle threading a lot better than their AMD counter part. However as I mentioned before the variations that you will see in performance doesn't have to do so much with brand as the chips actual performance, but like I said the chips internal handling of threads does play a weeee' little part in that.
@True Cruel - try this
-maxmem=6144 -maxvram=1280 -cpucount=120 -exthreads=120 -nosplash -skipintro -nobenchmark
edit: Fixed CPU count.
@TrueCruel - The reason you're seeing lesser use with higher settings is due the fact that threads are backing up and having to wait on slower threads. (Basically you end up with a false positive on the usage.) What is your launch options set to?
If you'd like add me on steam: Darkmere (Image is a Valknut) and I'll see what we cant do about getting those frames up.
@TrueCruel - Interesting discovery. Can you try that again, this time with Shadows set to 100?
@GT500 - As much as I love bashing AMD Im afraid this isn't the case as this is an issue with CPU's in general not just one brand versus the other. Have you attempted any of the 'fixes' mentioned above? If so what are your settings and system specs? What is your CPU & GPU use?
@rogerx - That is a known issue with the latest firmware for the GTX 5xx/6xx series. Revert to a previous version or try beta drives and you should be good.
What's your Steam Name?
@Johncage - You're comments are unwarrented and ill-desired. When you find yourself fighting the urge to speak, fight harder...
@rogerx - The Vsync was the increase GPU usage not CPU. For some reason leaving it enabled caused a bottleneck on the GPU, a heavy one at that. The same applies to the maxrenderahead setting.
As far as the CPU is concerned from one programmer to another; its an issue with tasking the worker threads. Of this I am 99% positive. When I wrote some of my first multithreaded applications I found myself running into similar issues.
The threads them selves are declared under a 'head' thread which manages them. The sub threads of the 'head' aren't set up as worker threads so the primary thread has to wait until the slave thread finishes before it can continue. This is why you notice a considerable boost in performance when we tell it that we have 120CPUs x 120Threads. We're basically informing it that we have a HUGE pool that it can make. We remove any queuing that was going on inside the 'head' thread so it alleviates the problem but does not resolve it.
Had this been 8-years ago we would have seen a huge degradation in performance when over allocated the system however now-a-days multicore CPU's are very efficient in handling multiple threads.
@ Rogerx - Good to hear.
Another note, rather a blast from the past. Go My Documents>Arma 3 Alpha and open the Arma3Alpha.cfg. Find the line that says "GPU_MaxFramesAhead=1000;" and replace with "GPU_MaxFramesAhead=1;"
If you havent already turn off vsync and you should notice a considerable boost in performance.
Your data is a bit outdated not to mention for a previous revision of an engine. So while I can see where you're coming from I believe your argument invalid.
Made an interesting discovery regarding cpuCount and exThread
Keep in mind I have all my settings maxed. 1920x1080 @ Ultra w/ max render distances on Scuba Showcase.
Bone Stock (no modifiers)
Avg CPU: 13%
FPS Max: 38
FPS Avg: 30
FPS Low: 22
-cpuCount=12 -exThreads=12 -nointro -nosplash -maxmem=16384 -maxVRAM=4095
Avg CPU: 16%
FPS Max: 42
FPS Avg: 34
FPS Low: 28
-cpuCount=120 -exThreads=120 -nointro -nosplash -maxmem=16384 -maxVRAM=4095
Avg CPU: 18%
FPS Max: 47
FPS Avg: 37
FPS Low: 32
Apparently when I had previously set my cpuCount and exThreads I must have accidentally hit a 0 in there. Results seem much better then I expected.
In regards to the memory usage this is something that would have to be done on their side. I don't see any swapping occurring however I do see a tiny (really small) reads occurring periodically from some of the game files. This suggests that their not putting everything to memory like they should/can be doing.
Do you have skype/steam? I wouldn't mind having someone to talk to whilst I do some testing. :-P
@mewle - I agree with your Rumsfeld quote. ;)
In regards to the engine itself I would imagine that a great deal of the issues we are seeing should be fixed or resolved once we hit beta so long as we make a big enough stink about it that the Devs see that it is indeed an issue.
I should also note I played around with some of the settings for the game, forcing threads and cpu count, etc. So far what I have found is that regardless of what I force upon it it will always use 90-103 threads. (this includes when I set cpu qty 1 and threads sub 90). This leads me to believe that their engine uses magic(static) numbers for initializing threads. I would be curious to see how they perform thread initialization and handle threads inside the engine.
On a side note I am working on the memory usage issue and will report back my findings later.
Same issue. More visible when running a server w/ AI. It seems that the AI take precedence over all other functions and due to the lack of multithreaded/core support it creates a bottle neck.
With AI (24 total) system usage looks like the following. CPU:21%, GPU:38%.
No AI. CPU:17%, GPU: 89%. (This is with modified files to tweak performance.)
I have played with everything that I can (Up to memory manipulation) to attempt to increase performance however nothing provides a stable usage. Even with the GPU at 89% @ ~36FPS I often will get drops in usage down to 0% @ 1-5FPS for a time varying from 1-6sec.
All my research points to the games engine. If I had to guess the engine has no thread queuing and a fixed number of available threads. The end result is unstable threads that can quickly form bottle necks as we have seen.
I should note that this is not to discourage others from attempting to tweak the game and its engine to try and get better performance. Simply noting my own discoveries and observations.
Im running two Vertex4 6GB/s in Raid0. (Huge throughput and I/O). Still have similar performance to my roommate and other friends who play with similar specs. Like Stk2008 pointed out that is not to say that there isn't an improvement by having an SSD simply its not that -big-. I can say that with the SSD's I load maps and scenarios significantly faster than my roommate.
I have also played around with CPU/GPU clocking to see if this would have affect on the FPS. Noticed a sizable increase in performance as I OC'd CPU up. 3.2Ghz -4Ghz yielded about 7FPS while 3.2Ghz to 5Ghz yielded almost 15.
With my system I can push the upward limit of 5Ghz without burning out but I would not recommend other attempt to without understanding what it is they're doing first.
If someone could though, could they perform some overclocking and get measurements from their system? I believe this will conclusively point towards an I/O bottleneck on the CPU.
Intel i7 3930k @ 5Ghz
16GB GSkill Ripjaw DDR3 Ram @ 1800MHZ
Asus Sabertooth x79
EVGA GTX 570GTX @ Core:990Mhz Memory:2050MHZ
(I have tried with and without SLI. No significant changes in performance.)
Creative xFi Titanium
Vertex 4 90GB x2 (raid0)
WD 1TB Green