ScardyBobScardyBobJoin Date: 2009-11-25Member: 69528Forum Admins, Forum Moderators, NS2 Playtester, Squad Five Blue, Reinforced - Shadow, WC 2013 - Shadow
If your 'waiting for GPU' value is greater than roughly 0-1ms, then lowering your resolution will increase your fps. If not, then your CPU and your resolution won't affect your fps.
In my case, I'm not usually waiting on GPU, but decreasing the resolution to ridiculously low levels does dramatically increase FPS (by about a hundred or so). Not sure what to think of that.
Interesting. For me the single biggest thing affecting performance is resolution, theres no way for me to get the game playable on my desktop resolution, but when I turn it down the game becomes playable.
I think one of the biggest impacts resolution has is on shader processing time. For most players, the CPU is the greatest bottleneck at the moment. Combine that with the fact that most players turn every single extraneous visual setting off, and you have a situation where the resolution isn't really having an impact on much at all.
Increasing my resolution past 1200 still drops my FPS by about 15-20 though.
In my case, I'm not usually waiting on GPU, but decreasing the resolution to ridiculously low levels does dramatically increase FPS (by about a hundred or so). Not sure what to think of that.
Yeah, its not an all encompassing rule. Its certainly still possible to get higher FPS by dropping your resolution even when CPU bottlenecked. However, if your reducing your resolution and not getting improved FPS then a CPU bottleneck is the likely culprit.
aeroripperJoin Date: 2005-02-25Member: 42471NS1 Playtester, Forum Moderators, Constellation
I got an older amd athlon x2 5600+ dual core and its definitely the bottleneck on my machine (gtx 460 1gb video). I set the priority on NS2 to "low" so it doesn't max out my cores at 100% when there's lots of action in the game, and cause fps hitches. Game works pretty well with every detail off and a low resolution.
I got an older amd athlon x2 5600+ dual core and its definitely the bottleneck on my machine (gtx 460 1gb video). I set the priority on NS2 to "low" so it doesn't max out my cores at 100% when there's lots of action in the game, and cause fps hitches. Game works pretty well with every detail off and a low resolution.
You say the game works better on low priority than normal or high?
matsoMaster of PatchesJoin Date: 2002-11-05Member: 7000Members, Forum Moderators, NS2 Developer, Constellation, NS2 Playtester, Squad Five Blue, Squad Five Silver, Squad Five Gold, Reinforced - Shadow, NS2 Community Developer
edited April 2013
There are three ways bottlenecks that may limit your fps:
- logic thread. CPU only, increases with number of entities (players, MACs, Crags etc ..). Changes the state of the game. Lua.
- render thread. CPU only, increases with graphics complexity (number of models, map geometry, lights, graphics options). Produces data for the graphics card (GPU). C++
- GPU - Graphics card. increase the same way as the render thread basically. Draws stuff on your screen.
You can see what you are bottle-necked by using r_stats:
"Waiting for GPU" != 0 -> The bottleneck is your graphics card. Lower graphics options
"Waiting for render thread" != 0 -> The bottleneck is the renderer thread. Lower graphics options
None of the above -> Your bottleneck is the logic thread ... nothing to do. You could probably increase your graphics options at no fps cost.
Now, the problematic part is that the bottleneck moves during a game. Early on, the render thread/GPU may be a bottleneck because there are few entities relevant to you, so the logic thread is not the bottleneck.
Late game combat sees more players fighting close to large number of alien crags/whips/shifts/infestation, meaning that the logic thread becomes the bottleneck.
Notice that as soon as the logic thread becomes the bottleneck, your graphics options ceases to matter. So your minimum FPS moments are not likely to be affected by your graphics settings (unless you are playing competitive games, where the number of players and entities are low compared to a public game).
This means that you should try to run r_stats at the worst times, when your fps is really low, and check where your bottleneck is. Granted, not the easiest of things... but if it turns out that your fps is limited by the logic thread, you might as well increase your graphics options, as that won't affect your worst-case fps anyhow...
"None of the above -> Your bottleneck is the logic thread ... nothing to do. You could probably increase your graphics options at no fps cost."
I believe AO and atmospherics(?) actually use some cpu, so increasing these graphics options might cut some fps. Then again, this might come under "render thread" which is supposedly 0 (could go up if you increase AO and atmos though). Someone else confirm?
aeroripperJoin Date: 2005-02-25Member: 42471NS1 Playtester, Forum Moderators, Constellation
You say the game works better on low priority than normal or high?
On my particular computer yes. I run NS2 with the resource monitor on the other monitor, and under normal priority the game can hit (both core) 100% cpu usage, which causes noticeable fps hitches in heavy combat. I've set it to "high" before which does seem to help until I start hitting medium-late game heavy combat and it uses more CPU. So if you got an old dual core like myself and notice it hits 100% cpu usage during game, it's worth a go. Now its around 90-95% usage at the heavy combat peaks and rarely ever hits 100%. Granted I only maintain ~15fps during the heaviest combat, but it does help. I got my windows 7 optimized to run a minimal of other programs with NS2 is running. I use a handy program called Prio to save the process priority as low so I only hafta do it once.
I haven't checked what matso wrote, but I wouldn't be surprised that the logic thread is my bottleneck.
matsoMaster of PatchesJoin Date: 2002-11-05Member: 7000Members, Forum Moderators, NS2 Developer, Constellation, NS2 Playtester, Squad Five Blue, Squad Five Silver, Squad Five Gold, Reinforced - Shadow, NS2 Community Developer
You say the game works better on low priority than normal or high?
I haven't checked what matso wrote, but I wouldn't be surprised that the logic thread is my bottleneck.
NS2 can use upto 2.5* cores in late-game combat, so if you are hitting 15 fps you are probably running out of cores to place threads on - you might call it being severly CPU limited. So in your case, you are probably bottlenecked by both the logic thread and the render thread. Lowering you graphics options will probably free up some CPU and give you some more fps.
Windows (or your graphics card driver) may well run an internal process to feed the graphics card data, which may explain why you get less hitching with NS2 at low prio; it would mean that the data NS2 produces are shifted to the graphics card as soon as it is produced.
You could try comparing NS2 at high and low prio and check the "Waiting for GPU"; if it goes up when NS2 runs at high prio, it would validate that theory.
* There are a couple of helper threads that offload processing from the logic thread - but this only helps when you have cores left over to run things on.
You say the game works better on low priority than normal or high?
I haven't checked what matso wrote, but I wouldn't be surprised that the logic thread is my bottleneck.
NS2 can use upto 2.5* cores in late-game combat, so if you are hitting 15 fps you are probably running out of cores to place threads on - you might call it being severly CPU limited. So in your case, you are probably bottlenecked by both the logic thread and the render thread. Lowering you graphics options will probably free up some CPU and give you some more fps.
I was wondering how you would get the situation where you would be bottlenecked by the render thread before the GPU. Though, I'd still love to see a screenshot of 0ms waiting for GPU and >0ms waiting for render thread.
You say the game works better on low priority than normal or high?
I haven't checked what matso wrote, but I wouldn't be surprised that the logic thread is my bottleneck.
NS2 can use upto 2.5* cores in late-game combat, so if you are hitting 15 fps you are probably running out of cores to place threads on - you might call it being severly CPU limited. So in your case, you are probably bottlenecked by both the logic thread and the render thread. Lowering you graphics options will probably free up some CPU and give you some more fps.
I was wondering how you would get the situation where you would be bottlenecked by the render thread before the GPU. Though, I'd still love to see a screenshot of 0ms waiting for GPU and >0ms waiting for render thread.
This is what my game looks like with a C2D and GTX 660. Wait for GPU is always 0 even on high details, and wait for render thread is 0+.
You say the game works better on low priority than normal or high?
I haven't checked what matso wrote, but I wouldn't be surprised that the logic thread is my bottleneck.
NS2 can use upto 2.5* cores in late-game combat, so if you are hitting 15 fps you are probably running out of cores to place threads on - you might call it being severly CPU limited. So in your case, you are probably bottlenecked by both the logic thread and the render thread. Lowering you graphics options will probably free up some CPU and give you some more fps.
I was wondering how you would get the situation where you would be bottlenecked by the render thread before the GPU. Though, I'd still love to see a screenshot of 0ms waiting for GPU and >0ms waiting for render thread.
This is what my game looks like with a C2D and GTX 660. Wait for GPU is always 0 even on high details, and wait for render thread is 0+.
Might posting a screenshot with r_stats and high detail? That's a pretty unusual CPU/GPU combo, but it gives me an idea of what to look for when diagnosing fps issues.
There are three ways bottlenecks that may limit your fps:
- logic thread. CPU only, increases with number of entities (players, MACs, Crags etc ..). Changes the state of the game. Lua.
- render thread. CPU only, increases with graphics complexity (number of models, map geometry, lights, graphics options). Produces data for the graphics card (GPU). C++
- GPU - Graphics card. increase the same way as the render thread basically. Draws stuff on your screen.
You can see what you are bottle-necked by using r_stats:
"Waiting for GPU" != 0 -> The bottleneck is your graphics card. Lower graphics options
"Waiting for render thread" != 0 -> The bottleneck is the renderer thread. Lower graphics options
None of the above -> Your bottleneck is the logic thread ... nothing to do. You could probably increase your graphics options at no fps cost.
Now, the problematic part is that the bottleneck moves during a game. Early on, the render thread/GPU may be a bottleneck because there are few entities relevant to you, so the logic thread is not the bottleneck.
Late game combat sees more players fighting close to large number of alien crags/whips/shifts/infestation, meaning that the logic thread becomes the bottleneck.
Notice that as soon as the logic thread becomes the bottleneck, your graphics options ceases to matter. So your minimum FPS moments are not likely to be affected by your graphics settings (unless you are playing competitive games, where the number of players and entities are low compared to a public game).
This means that you should try to run r_stats at the worst times, when your fps is really low, and check where your bottleneck is. Granted, not the easiest of things... but if it turns out that your fps is limited by the logic thread, you might as well increase your graphics options, as that won't affect your worst-case fps anyhow...
I have GPU at 0, but render thread frequently rises above 0, even though I have an Ivy Bridge 3570k @4.5GHz...
IronHorseDeveloper, QA Manager, Technical Support & contributorJoin Date: 2010-05-08Member: 71669Members, Super Administrators, Forum Admins, Forum Moderators, NS2 Developer, NS2 Playtester, Squad Five Blue, Subnautica Playtester, Subnautica PT Lead, Pistachionauts
@shonan you have a huge myriad of issues with ns2... i'd really consider troubleshooting each one step by step, and see if they point to a universal cause like a CPU setting, overheating, etc. (have you disabled speedstep? aka EIST?)
@shonan you have a huge myriad of issues with ns2... i'd really consider troubleshooting each one step by step, and see if they point to a universal cause like a CPU setting, overheating, etc. (have you disabled speedstep? aka EIST?)
Yes, indeed, but I guess I am running an unusual setup with my graphics card etc.
The thing is, I have looked at all things and my system couldnt be running better, its perfectly stable, I have correct BIOS/UEFI settings, core parking disabled etc etc...
I just recently further fine tuned my overclock and its Prime95 stable, temps are fine, clocks are properly at 4.5 when playing etc. I just guess its my setup, it is unusual I have to admit with my Radeon HD 4870 X2, and has had some problems with certain games like NS2 which are developed by a small team with limited testing.
Actually now I've gotten the game quite playable, although with a lower than my desktop resolution, it has gone better in the latest patches (and 0.3GHz further overclock. Still if I could crank up the resolution and maybe some settings as well I couldnt be happier
IronHorseDeveloper, QA Manager, Technical Support & contributorJoin Date: 2010-05-08Member: 71669Members, Super Administrators, Forum Admins, Forum Moderators, NS2 Developer, NS2 Playtester, Squad Five Blue, Subnautica Playtester, Subnautica PT Lead, Pistachionauts
@shonan i would definitely attempt swapping in another card if you get the chance to, for testing purposes.
Or simply disabling the crossfire to start with. Even CEOs of graphic card companies (nvidia) have stated that MGPU setups will never be as problem free or latency free as single GPU setups.
After years of owning multiple SLI setups and going through nightmares, i eventually caved for a single GPU and have had zero issues.
Long story short: eliminate factors one by one until you see a change.
@shonan i would definitely attempt swapping in another card if you get the chance to, for testing purposes.
Or simply disabling the crossfire to start with. Even CEOs of graphic card companies (nvidia) have stated that MGPU setups will never be as problem free or latency free as single GPU setups.
After years of owning multiple SLI setups and going through nightmares, i eventually caved for a single GPU and have had zero issues.
Long story short: eliminate factors one by one until you see a change.
I have tried disabling crossfire, that resulted in a practically exact 50% drop in FPS. I know all that about multiGPU setups and the next card I am getting will be a single one, but still this card is powerful enough to run new games so I havent felt the need to throw dosh on a new one just yet. The only game performing sub-optimally for me right now is Natural Selection 2.
@shonan i would definitely attempt swapping in another card if you get the chance to, for testing purposes.
Or simply disabling the crossfire to start with. Even CEOs of graphic card companies (nvidia) have stated that MGPU setups will never be as problem free or latency free as single GPU setups.
After years of owning multiple SLI setups and going through nightmares, i eventually caved for a single GPU and have had zero issues.
Long story short: eliminate factors one by one until you see a change.
I have tried disabling crossfire, that resulted in a practically exact 50% drop in FPS. I know all that about multiGPU setups and the next card I am getting will be a single one, but still this card is powerful enough to run new games so I havent felt the need to throw dosh on a new one just yet. The only game performing sub-optimally for me right now is Natural Selection 2.
Ironically, even though you technically get double the framerate in XF, in practice it may still only be like getting the single GPU performance. It appears that ATI/AMD XF hardware suffers from the runt/dropped frame phenomenom described here. For example, FRAPs and/or the in-game FPS may be displaying something like this for your XF setup:
However, in practice, your actual displayed FPS is more like this:
Where your XF 4870s actually give you no better performance than a single 4870 setup.
@shonan i would definitely attempt swapping in another card if you get the chance to, for testing purposes.
Or simply disabling the crossfire to start with. Even CEOs of graphic card companies (nvidia) have stated that MGPU setups will never be as problem free or latency free as single GPU setups.
After years of owning multiple SLI setups and going through nightmares, i eventually caved for a single GPU and have had zero issues.
Long story short: eliminate factors one by one until you see a change.
I have tried disabling crossfire, that resulted in a practically exact 50% drop in FPS. I know all that about multiGPU setups and the next card I am getting will be a single one, but still this card is powerful enough to run new games so I havent felt the need to throw dosh on a new one just yet. The only game performing sub-optimally for me right now is Natural Selection 2.
Ironically, even though you technically get double the framerate in XF, in practice it may still only be like getting the single GPU performance. It appears that ATI/AMD XF hardware suffers from the runt/dropped frame phenomenom described here. For example, FRAPs and/or the in-game FPS may be displaying something like this for your XF setup:
However, in practice, your actual displayed FPS is more like this:
Where your XF 4870s actually give you no better performance than a single 4870 setup.
Yes, I am aware of this, I can do some further testing with and without Crossfire to see if it makes any difference regarding these things.
About the spike on changing weapons, it has disappeared for me, apparently due to the latest patch, cannot reproduce either with crossfire off nor on, except on very very rare occasions (once in 20 times). If it is of any interest heres two "spikes" if you even can call them that, which were result of changing weapons now: http://imgur.com/a/uTgET
And heres shots from the same situation as present on the other spikes, now with crossfire disabled. Now they dont alternate like that, but the lag is the same or even worse: http://imgur.com/a/ptC3A
About the performance generally, without crossfire it is a lot worse. I much rather play with Crossfire than without.
About the spike on changing weapons, it has disappeared for me, apparently due to the latest patch, cannot reproduce either with crossfire off nor on, except on very very rare occasions (once in 20 times).
Those spikes come and go for me as the patches go. Sometimes it is when I change weapons (especially the build tool and the welder), sometimes when I spawn as a marine... I have never been able to find the origin of the problem.
Comments
Increasing my resolution past 1200 still drops my FPS by about 15-20 though.
You say the game works better on low priority than normal or high?
- logic thread. CPU only, increases with number of entities (players, MACs, Crags etc ..). Changes the state of the game. Lua.
- render thread. CPU only, increases with graphics complexity (number of models, map geometry, lights, graphics options). Produces data for the graphics card (GPU). C++
- GPU - Graphics card. increase the same way as the render thread basically. Draws stuff on your screen.
You can see what you are bottle-necked by using r_stats:
"Waiting for GPU" != 0 -> The bottleneck is your graphics card. Lower graphics options
"Waiting for render thread" != 0 -> The bottleneck is the renderer thread. Lower graphics options
None of the above -> Your bottleneck is the logic thread ... nothing to do. You could probably increase your graphics options at no fps cost.
Now, the problematic part is that the bottleneck moves during a game. Early on, the render thread/GPU may be a bottleneck because there are few entities relevant to you, so the logic thread is not the bottleneck.
Late game combat sees more players fighting close to large number of alien crags/whips/shifts/infestation, meaning that the logic thread becomes the bottleneck.
Notice that as soon as the logic thread becomes the bottleneck, your graphics options ceases to matter. So your minimum FPS moments are not likely to be affected by your graphics settings (unless you are playing competitive games, where the number of players and entities are low compared to a public game).
This means that you should try to run r_stats at the worst times, when your fps is really low, and check where your bottleneck is. Granted, not the easiest of things... but if it turns out that your fps is limited by the logic thread, you might as well increase your graphics options, as that won't affect your worst-case fps anyhow...
"None of the above -> Your bottleneck is the logic thread ... nothing to do. You could probably increase your graphics options at no fps cost."
I believe AO and atmospherics(?) actually use some cpu, so increasing these graphics options might cut some fps. Then again, this might come under "render thread" which is supposedly 0 (could go up if you increase AO and atmos though). Someone else confirm?
On my particular computer yes. I run NS2 with the resource monitor on the other monitor, and under normal priority the game can hit (both core) 100% cpu usage, which causes noticeable fps hitches in heavy combat. I've set it to "high" before which does seem to help until I start hitting medium-late game heavy combat and it uses more CPU. So if you got an old dual core like myself and notice it hits 100% cpu usage during game, it's worth a go. Now its around 90-95% usage at the heavy combat peaks and rarely ever hits 100%. Granted I only maintain ~15fps during the heaviest combat, but it does help. I got my windows 7 optimized to run a minimal of other programs with NS2 is running. I use a handy program called Prio to save the process priority as low so I only hafta do it once.
I haven't checked what matso wrote, but I wouldn't be surprised that the logic thread is my bottleneck.
NS2 can use upto 2.5* cores in late-game combat, so if you are hitting 15 fps you are probably running out of cores to place threads on - you might call it being severly CPU limited. So in your case, you are probably bottlenecked by both the logic thread and the render thread. Lowering you graphics options will probably free up some CPU and give you some more fps.
Windows (or your graphics card driver) may well run an internal process to feed the graphics card data, which may explain why you get less hitching with NS2 at low prio; it would mean that the data NS2 produces are shifted to the graphics card as soon as it is produced.
You could try comparing NS2 at high and low prio and check the "Waiting for GPU"; if it goes up when NS2 runs at high prio, it would validate that theory.
* There are a couple of helper threads that offload processing from the logic thread - but this only helps when you have cores left over to run things on.
This is what my game looks like with a C2D and GTX 660. Wait for GPU is always 0 even on high details, and wait for render thread is 0+.
I have GPU at 0, but render thread frequently rises above 0, even though I have an Ivy Bridge 3570k @4.5GHz...
Well, that was when I had my athlon x2 lol. Then I got a core i3 3220 b/c this game uses at most 2 cores, and BAM... 70fps, 40 in combat.
Yes, indeed, but I guess I am running an unusual setup with my graphics card etc.
The thing is, I have looked at all things and my system couldnt be running better, its perfectly stable, I have correct BIOS/UEFI settings, core parking disabled etc etc...
I just recently further fine tuned my overclock and its Prime95 stable, temps are fine, clocks are properly at 4.5 when playing etc. I just guess its my setup, it is unusual I have to admit with my Radeon HD 4870 X2, and has had some problems with certain games like NS2 which are developed by a small team with limited testing.
Actually now I've gotten the game quite playable, although with a lower than my desktop resolution, it has gone better in the latest patches (and 0.3GHz further overclock. Still if I could crank up the resolution and maybe some settings as well I couldnt be happier
Or simply disabling the crossfire to start with. Even CEOs of graphic card companies (nvidia) have stated that MGPU setups will never be as problem free or latency free as single GPU setups.
After years of owning multiple SLI setups and going through nightmares, i eventually caved for a single GPU and have had zero issues.
Long story short: eliminate factors one by one until you see a change.
(5 days ago: http://www.techpowerup.com/183129/AMD-Makes-a-Breakthrough-in-Improving-Frame-Latency.html)
However, in practice, your actual displayed FPS is more like this:
Where your XF 4870s actually give you no better performance than a single 4870 setup.
What spikes exactly do you mean? Do you mean these? http://forums.unknownworlds.com/discussion/comment/2111881/#Comment_2111881
And if its those do you mean the ones happening when changing weapons or the ones you can see here? http://i.imgur.com/9Bsa1Nz.jpg
I can recreate both situations without Crossfire to test if it makes any difference.
Yes, I am aware of this, I can do some further testing with and without Crossfire to see if it makes any difference regarding these things.
About the spike on changing weapons, it has disappeared for me, apparently due to the latest patch, cannot reproduce either with crossfire off nor on, except on very very rare occasions (once in 20 times). If it is of any interest heres two "spikes" if you even can call them that, which were result of changing weapons now: http://imgur.com/a/uTgET
And heres shots from the same situation as present on the other spikes, now with crossfire disabled. Now they dont alternate like that, but the lag is the same or even worse: http://imgur.com/a/ptC3A
About the performance generally, without crossfire it is a lot worse. I much rather play with Crossfire than without.
Notice how the threads above are all random numbers?
Its just a byproduct of how profile works and it happens to everyone.
Glad the latest patch fixed it though.... I bet it had something to do with shaders
False positives? After the latest patch the top numbers are like that all the time, forgot to mention it in the post, apparently its a bug.
However the results are completely valid as everything else shown on the profiler is the same as before the patch.