Major Performance Issues

[gazzthompson]

Also get major FPS drops when looking at certain areas of maps regardless of what is actually on screen.

Q6600 @ 3.4ghz
4GB ram
GTX 560
W7

[maarit]

hmm,thats odd,now i can play all high,8xaa,v-sync and big resolution.
in MARLIN near that tv-tower FPS was sometimes 28.
all i do is that i cleaned my pc inside.
got to keep testing this.

[Prevtzer]

Does anybody without Win 7 have performance issues?

[tankninja1]

The razor game booster helps a bit I'm up to an average of 20fps (up from less than 15fps average).

[Boris]

Does anybody without Win 7 have performance issues?

Yeah, XP 64 here, though my machine is pretty damn old now so performance issues are not going to be so surprising as some others here.

Code: Select all

System Information
------------------
  Operating System: Microsoft(R) Windows(R) XP Professional x64 Edition (5.2.3790)
      Architecture: 64-bit
          Language: English (United Kingdom)
       Motherboard: http://www.abit.com.tw/ AN8 Series(NF-CK804)
         Processor: AMD Athlon 64 X2 4400+ Dual Core Processor
            Memory: 4.00 GB
         Page File: -260100096.00 B
    .NET Framework: 4.0

Display Information
-------------------
 Display Device(s): Plug and Play Monitor on NVIDIA GeForce 7800 GT
                    Plug and Play Monitor on NVIDIA GeForce 7800 GT
    Driver Version: 6.14.12.9610 (296.10)
    Display Memory: 256.00 MB
     Multisampling: 2, 4
               DPI: 96 (100%)

Audio Information
-----------------
  Primary Playback: kX Wave CT4620 10k1 [9c00] 0/1
 Primary Recording: kX Wave CT4620 10k1 [9c00] 0/1
               EAX: True
           EAX 1.0: False
           EAX 2.0: True
           EAX 3.0: True
           EAX 4.0: False
           EAX 5.0: False
             X-RAM: False

Disk Information
----------------
      Install Path: D:\games\bf2pr
                    Free: 67.09 GB, Total: 140.00 GB, NTFS
          Mod Path: D:\games\bf2pr\mods\pr
                    Free: 67.09 GB, Total: 140.00 GB, NTFS
     Profiles Path: C:\Documents and Settings\Administrator\My Documents\ProjectReality\Profiles
                    Free: 9.78 GB, Total: 25.00 GB, NTFS
    Update DL Path: D:\games\bf2pr\binaries
                    Free: 67.09 GB, Total: 140.00 GB, NTFS
   Update Log Path: D:\games\bf2pr\logs
                    Free: 67.09 GB, Total: 140.00 GB, NTFS

Game Information
----------------
      CD Key Valid: True
    Installed Mods: bf2, pr, xpack
       Current Mod: pr
       BF2 Version: BF2 1.5
        PR Version: 1.0.35.0
          Language: english
   Debug Available: False
     Large Address: True
Supported Hardware: None

Profile Information
-------------------
      Profile 0001: boris
              Type: Online
 Last Used Profile: True
        View Intro: False
        Fullscreen: True
      Display Mode: 1280x1024@60Hz
Display Mode Valid: True
   Graphics Scheme: Low
     Multisampling: 4x
             VSync: True
   Terrain Quality: Low
   Effects Quality: Low
  Geometry Quality: Low
   Texture Quality: Low
  Lighting Quality: Low
   Dynamic Shadows: Low
    Dynamic Lights: Low
 Texture Filtering: Low
    Audio Provider: Software
    Provider Valid: True
     Audio Quality: High
               EAX: False

The issue for me appears to be CPU-related. I ran some comparison tests on an empty, locally-run Al-Basrah server here looking into the city where frame rate appeared to take the biggest hit, and found that there was very little difference between the 3 preset video profiles in PRLauncher: Low, Medium & High. The results were as follows:

Code: Select all

Low    Profile, 2xAA, VDS100% = 38.0 fps, GPU Load = ~19%

Medium Profile, 2xAA, VDS100% = 37.7 fps, GPU Load = ~21%

High   Profile, 2xAA, VDS100% = 35.5 fps, GPU Load = ~23%

Readings were all taken from the exact same position. Used GPU-Z to log GPU usage stats.

The primary issue I've been encountering in v1.0 is that it's dropping to 15.0 fps quite often (actually, that's just the lowest value BF2 will display, so it's likely even less than this), and then the player movement starts rubber-banding. I don't remember ever experiencing this kind of rubber-banding before in versions prior to v1.0 except for when there was an actual server issue that caused it, where typically everyone would experience it at the same time, or when my own net connection was acting laggy.

I don't know what's different now in v1.0, but if I were to throw a wild theory out there, I wonder if it could be something to do with how the PR binaries have been rigged to work in this version. Something like when CPU resources become exhausted, the net code aspect of BF2 takes a priority hit where it didn't before, maybe like PRLauncher holds priority over PRBF2.exe or something, causing a restriction under certain conditions.

Another theory is that it could be a consequence of forced shadows, where it's possible they're being calculated on the CPU by this engine, not GPU, and so throwing up an increased load onto the CPU I've not experienced before (as I used to run all low settings 0.98 prior).

Some may say it's the player count, but I don't think so necessarily, as I've been on a server with ~40-players and still ran into it. It does indeed do it more often on a busy 100-player server, but it also does it if there's just a lot of activity going on in general, as in a busy firefight, explosions, sounds, irrespective of player numbers. I've also gotten the feeling coop bot play isn't upto what it used to be either as I've played some 64-bot Muttrah and felt that it ran worse than I remember it (hell, I used to play 96-bot muttrah often here and it ran well enough). The overall online problem though just appears to be that the processor chokes, then the net code suffers. If the net code didn't croak under load then things might feel a lot better (because you all know just how hideous rubber-banding is to gameplay), but it still leaves the issue of why exactly the load appears to be so high in comparison to older versions.

Annoyingly, I only recently deleted the old 0.9 binaries here, else I could have done some proper A/B tests between the versions. I may even download/install them again sometime...

Ultimately though, A: I need to upgrade the machine here, and B: The BF2 engine is just rubbish by todays standards, and probably about reaching its limits here with this mod.

[Skitrel]

Another theory is that it could be a consequence of forced shadows, where it's possible they're being calculated on the CPU by this engine, not GPU, and so throwing up an increased load onto the CPU I've not experienced before (as I used to run all low settings 0.98 prior).

Some may say it's the player count, but I don't think so necessarily

Alternatively, it could be both.

How are shadows calculated in game? I've got a theory that shadows might be getting calculated by the CPU whether or not the actual object casting the shadow is getting rendered. So, when looking in the direction of lots of infantry, objects, etc, all of them are having their shadows drawn with absolutely zero render culling, while previously people using low shadows were not getting this issue because there were none rendering.

This would explain why, inexplicably, when looking towards populated areas or areas filled with many objects the fps takes a big hit whether or not you can actually see those objects.

If that's the case, shadows should be rolled back to the previously unforced state because render culling isn't helping lower end systems that were previously not experiencing this issue in those instances.

[Pyrohair]

Alternatively, it could be both.

How are shadows calculated in game? I've got a theory that shadows might be getting calculated by the CPU whether or not the actual object casting the shadow is getting rendered. So, when looking in the direction of lots of infantry, objects, etc, all of them are having their shadows drawn with absolutely zero render culling, while previously people using low shadows were not getting this issue because there were none rendering.

This would explain why, inexplicably, when looking towards populated areas or areas filled with many objects the fps takes a big hit whether or not you can actually see those objects.

If that's the case, shadows should be rolled back to the previously unforced state because render culling isn't helping lower end systems that were previously not experiencing this issue in those instances.

+1. Can it hurt to try?

[Mats391]

When i pop the smoke from an uparmored hmmwv my FPS drops from 60 to under 15 while looking at the smoke. This is the first time though that i got unplayable FPS.
Tested on: Local, Mestia STD

i know you did a lot to optimize the effects but it still seems to be the biggest issue at least for me.

[ExeTick]

When i pop the smoke from an uparmored hmmwv my FPS drops from 60 to under 15 while looking at the smoke. This is the first time though that i got unplayable FPS.
Tested on: Local, Mestia STD

i know you did a lot to optimize the effects but it still seems to be the biggest issue at least for me.

they could have missed to optimize something. thats why it is needed to post reasons when you get performance drop

[Boris]

+1. Can it hurt to try?

Egos, probably.

It would be nice to come up with some ways to test this. First problem that comes to mind though is how to get around PRLauncher forcing the higher settings. This is the problem with that software that I really don't like about it, that few people seem to see - that we are having our abilities stripped away. I don't think it even allows true low settings on offline single player? (edit: indeed it doesn't - testing shows shadows are still rendered)

I tried playing Pavlovsk Bay last night during Drake's event thingy but it was nigh impossible at times. It would run fine up the north end of the map, or running through the forests as we headed south, but as soon as the bay area came close the fps turned red 15.0/66.7 and became impossible - unplayable low-fps & connection rubber-banding. But, I used to play this map without issue on versions prior to v1.0? Sure, there will have been a performance hit when looking towards the bay, but nothing like this. I had to quit the game.

What was causing it? I don't know. The 'scenario' being played out during the event had pretty-much the entire other team down there in the bay area as far as I know, so maybe that was the primary cause. Or maybe it was just a consequence of building shadows (correction: actually, not building shadows, as if I remember rightly these are pre-generated for statics during compile). Or maybe the sum of both. I just don't know at this point. I could maybe simulate it by editing the map to have 48 bots spawn down in that area to try to mimic the effects, but I'd still need some way to be able to disable shadows.

[Spook]

I just took a look at the CPU and GPU usage ingame. It never goes over 30% CPU and 40% GPU. Still my FPS drops to 30 sometimes and is generally very unstable.



Gainward Geforce GTX 680 Phantom 4GB
intel i5 2500k (I assigned 2 cores to BF2, so the 30% actually come from the 2 cores and not from 4 cores, where 2 cannot be used anyways)
Gigabyte P67A-D3-B3
Creative SB X-Fi Titanium
8gb DDR3 1333
Corsair HX750 750W
Win7 Pro 64bit

[Skitrel]

I just took a look at the CPU and GPU usage ingame. It never goes over 30% CPU and 40% GPU. Still my FPS drops to 30 sometimes and is generally very unstable.



Gainward Geforce GTX 680 Phantom 4GB
intel i5 2500k (I assigned 2 cores to BF2, so the 30% actually come from the 2 cores and not from 4 cores, where 2 cannot be used anyways)
Gigabyte P67A-D3-B3
Creative SB X-Fi Titanium
8gb DDR3 1333
Corsair HX750 750W
Win7 Pro 64bit

The BF2 engine isn't multi threaded, it can't use more than a single core at any one time anyway. CPU usage for actual gameplay will max out at 25% for a core on an i5 2500k, as it does on mine. You should consider this running at the absolute maximum your cpu can do for it. That extra 5% will be from the rest of the system, mumble, the prlauncher, etc.

If it's running 25% of cpu, it's over stressing itself.

[Spook]

Thats why I said that I already limited CPu usage to 2 cores with BF2. The engine uses 2 cores, and still is seems that it uses only 30% of those 2 cores. Or the displaying might be wrong and it actually still means 30% of 4 cores instead of 2.

I even assigned PRlaucnher and mumble to core 3-4 and BF2 to 1-2. No change at all. When recording gameplay its even worse that way. I assigned the recording tool to core 3-4 aswell and for some reason FPS drops about 10 frames more than with the standard setting were windows assigns the cores.

I mean i already have OCed to 4,5 GHZ ....how much does PR need to run smooth? 2x 6,0 GHZ? Thats ridiculous.

[Skitrel]

Thats why I said that I already limited CPu usage to 2 cores with BF2. The engine uses 2 cores, and still is seems that it uses only 30% of those 2 cores. Or the displaying might be wrong and it actually still means 30% of 4 cores instead of 2.

I even assigned PRlaucnher and mumble to core 3-4 and BF2 to 1-2. No change at all. When recording gameplay its even worse that way. I assigned the recording tool to core 3-4 aswell and for some reason FPS drops about 10 frames more than with the standard setting were windows assigns the cores.

I mean i already have OCed to 4,5 GHZ ....how much does PR need to run smooth? 2x 6,0 GHZ? Thats ridiculous.

The engine doesn't properly use 2 cores. This was visible back in 2005 when the arguments about hyper threading cropped up when the game was first released. Performance difference between dual and single core machines was shown to be next to nothing back then when the 100 odd page threads of arguments came to an end with benchmarked evidence. Fun times. It's funny to look back on how much people with "high end" machines were egotistical about their performance gains now that they're low end machines. E-penis length fights are the most entertaining when looking back in retrospect.

[dysin]

question: do the people with fps drops in 1.0 have the issue where certain helicopters will stutter your game? i've flown since i began pr, and this issue is new to 1.0 and was present in the beta as well.
the specific conditions are: brand new helicopter (not a second spawn), and only observed while piloting (not present as a passenger or on the ground). it isn't true for all helicopters. it isn't present in every match.

i wonder if these two issues aren't linked by a common cause, but where the helicopter observation is a much more virulent example of frame hits by that cause.
https://www.realitymod.com/forum/f26-pr ... l-lag.html
https://www.realitymod.com/forum/f26-pr ... a-lag.html
https://www.realitymod.com/forum/f26-pr ... n-cas.html
https://www.realitymod.com/forum/f26-pr ... helis.html
https://www.realitymod.com/forum/f26-pr ... ugged.html
https://www.realitymod.com/forum/f26-pr ... craft.html
https://www.realitymod.com/forum/f26-pr ... t-lag.html

[AFsoccer]

This is all very helpful. Please continue to post specifics so we can work on fixes. The hardest part is trying to figure out the culprit(s).

[=]H[=OPOSSUM]

This is all very helpful. Please continue to post specifics so we can work on fixes. The hardest part is trying to figure out the culprit(s).

Have you considered deactivating the forced settings in the launcher until further optimizing has been done? It could serve as hotfix for now.

[dysin]

there's something killing it for me. 15 stable fps on fallujah, never had an issue before, definitely not this bad. wouldn't that narrow it down to the effects? maybe it is the vehicle wrecks?

[Boris]

OK, so I just come to realise that it is actually possible to turn shadows off. I've been running the "Low" profile in PRLauncher without realising that all "Low" shown in video settings is not the lowest settings possible, but that it's possible to manually configure shadows and dynamic lighting as "Off" completely.

So, I set those off and did some testing. I found no real difference between them though on a 96 bot offline coop server with all the team stood still at the Barracks flag on Pavlovsk. FPS tests with and without seemed to produce about the same result.

Also played Muttrah online, ~98 player, but it was still rubber-banding a lot. Noticed every time CAS fired off rounds in the distance things went downhill a bit, but maybe that's to be expected though as the CPU's already running too close to 100% all the time. Still, I don't know why moving through the city is still causing problems in general, even when there's not many players nearby, nor any local fire fights going on.

Damn though, I was kinda hoping shadows were the cause of it.

[Skitrel]

OK, so I just come to realise that it is actually possible to turn shadows off. I've been running the "Low" profile in PRLauncher without realising that all "Low" shown in video settings is not the lowest settings possible, but that it's possible to manually configure shadows and dynamic lighting as "Off" completely.

So, I set those off and did some testing. I found no real difference between them though on a 96 bot offline coop server with all the team stood still at the Barracks flag on Pavlovsk. FPS tests with and without seemed to produce about the same result.

Also played Muttrah online, ~98 player, but it was still rubber-banding a lot. Noticed every time CAS fired off rounds in the distance things went downhill a bit, but maybe that's to be expected though as the CPU's already running too close to 100% all the time. Still, I don't know why moving through the city is still causing problems in general, even when there's not many players nearby, nor any local fire fights going on.

Damn though, I was kinda hoping shadows were the cause of it.

Dynamic lighting and Shadows are forced settings I believe. Changing them below medium does absolutely nothing, but some (including myself) have spoken about thinking we get slightly better framerates by having them set to medium rather than off or low. Whether that's in heads or not is questionable though.

Either way, you see no difference when lowering these settings because lowering them isn't supposed to do anything.

Devs should make a change to the forced settings without adding it to the changelog, just label it "testing for performance". Then see what the reactions are in the community without people knowing specifically what has changed. This might bring up a whole host of new theories and ideas.

I'd suggest doing only 1 change at a time if doing something like that though, that way the area could be effectively narrowed down to "latest change improve things enormously" or "latest change made things massively worse again" with each update.