Sign in with
Sign up | Sign in
Your question

What is bottlnecking

Last response: in Graphics & Displays
Share
February 8, 2009 2:22:41 AM

I have a 4870 gfx and a E8400 cpu, the cpu is clocked at 4ghz but ive dropped it down to 3.85 and the 4870 is clocked at gpu 785 men 1000mhz, ive been playing Crysis and getting good fps of 30 at least i think its good, the game is set at full and the screen res is 1280 x1024.

I have noticed a slight lag every so often like firing a round but it doesnt fire til a split second later and slight freezing and such like, someone said it might be bottleneck between the cpu and gpu but i wunt know.
Can anyone help me sort this out ?

More about : bottlnecking

a b U Graphics card
February 8, 2009 3:00:35 AM

What you may want to look at is vsync. I know nvidia has that, check the settings for the 4870 and see if they have that. I have to use that on my lcd as well, because the screen can't refresh fast enough to completely keep up with the video output, what vsync does is caps the fps so that the monitor is able to keep up. could be what's going in your setup, especially at your resolution.
February 8, 2009 4:55:28 AM

Stiffex said:
I have a 4870 gfx and a E8400 cpu, the cpu is clocked at 4ghz but ive dropped it down to 3.85 and the 4870 is clocked at gpu 785 men 1000mhz, ive been playing Crysis and getting good fps of 30 at least i think its good, the game is set at full and the screen res is 1280 x1024.

I have noticed a slight lag every so often like firing a round but it doesnt fire til a split second later and slight freezing and such like, someone said it might be bottleneck between the cpu and gpu but i wunt know.
Can anyone help me sort this out ?


Does it only happen in Crysis, not other games? If so, then change the quality setting back one step.
Related resources
a c 271 U Graphics card
a b à CPUs
February 8, 2009 12:30:45 PM

If possible try upping your res to 1680 x 1050.
February 8, 2009 2:38:46 PM

Mousemonkey said:
If possible try upping your res to 1680 x 1050.


The screen is 5:4, but you want him to use 16:10?
a c 271 U Graphics card
a b à CPUs
February 8, 2009 7:35:01 PM

Ok, so a new monitor may be required, but yes I want him to use 16:10.
February 8, 2009 8:01:33 PM

Mousemonkey said:
Ok, so a new monitor may be required, but yes I want him to use 16:10.


How would having a higher resolution widescreen monitor help with the performance issues?
a c 130 U Graphics card
a b à CPUs
February 9, 2009 6:37:32 AM

Oh no here we go again,
When you game at a relativly low res 1280 x1024, sometimes it can cause the CPU to struggle to keep up with the GPU. What upping the res does is test this by giving the GPU more to do. This gives the CPU slightly more processing time before the next frame is required.

Mactronix
a b à CPUs
February 9, 2009 8:18:39 AM

Well said Mactronix,
But I'm wondering a little, will the fps increase or decrease? I mean like its more work...
a c 171 U Graphics card
a c 84 à CPUs
February 9, 2009 8:37:30 AM

Depends on how its bound. If it is CPU bound, there will be nearly no drop in frames at all, as the GPU was waiting to much. Going from 1280x1024 to 16x10 is only one jump in res, I doubt there will be much difference at all. In addition, I doubt it will fix his problem. I would look to newer drivers and patches to the game. If that doesn't fix it, look at power/heat issues.
February 9, 2009 11:03:34 AM

Probably just stuff loading into memory. Maybe try turning some settings down. I wouldn't be too worried about it.
a c 130 U Graphics card
a b à CPUs
February 9, 2009 3:14:51 PM

I agree with 4745454b that ideally you would want to go up a couple of resolutions to test it properly but feel that the differance (about 60%) is enough for a quick check.
I dont think it will be the main cause of the issue in this case though i dont think the set up should run into these types off issues. What San Pedro is saying would also be my first guess at what is going wrong.
@ Stiffex
What are the rest of the system specs ?

Mactronix
February 9, 2009 6:38:11 PM

mactronix said:
Oh no here we go again,
When you game at a relativly low res 1280 x1024, sometimes it can cause the CPU to struggle to keep up with the GPU. What upping the res does is test this by giving the GPU more to do. This gives the CPU slightly more processing time before the next frame is required.

Mactronix


He's only getting 30fps, that means he is GPU bound. If he were CPU bound he would be getting much higher FPS.

He's trying to play Crysis on full settings. Even with an HD4870 and only 1280x1024 that's not possible with a good,consistent frame rate.
a c 130 U Graphics card
a b à CPUs
February 9, 2009 8:06:52 PM

It doesn't work like that, the GPU is dependant on the CPU to feed it info so if you have a CPU restriction you end up with low FPS.
This is the whole point of adjusting the resolution. The CPU is basically responsible for working out the physics, where stuff is on screen etc, this stays pretty constant regardless of resolution.
The GPU on the other hand takes this info and colours /shades the scene, adds AA/AF etc. At a lower resolution and especially at lower quality settings this is much easier and so the GPU can fill its frame buffers and be waiting for the next info from the CPU much more quickly than it can at higher resolutions as obviously there are many more pixels to colour shade etc.
If you increase your res and the FPS go down then that shows that the GPU isnt up to the extra workload and indicates that it was the problem.
If the FPS stay the same but gameplay is smoother or if they go up that indicates the CPU was holding the GPU back.

@ Stiffex,
I still dont think this is whats happening but its worth testing anyway if you can. How much Ram do you have and also are you using V- sync ? as ohiou_grad_ says it can have a detrimental effect on your performance, if you have it on try turning it off.

Mactronix
February 10, 2009 4:59:23 AM

mactronix said:
It doesn't work like that, the GPU is dependant on the CPU to feed it info so if you have a CPU restriction you end up with low FPS.
This is the whole point of adjusting the resolution. The CPU is basically responsible for working out the physics, where stuff is on screen etc, this stays pretty constant regardless of resolution.
The GPU on the other hand takes this info and colours /shades the scene, adds AA/AF etc. At a lower resolution and especially at lower quality settings this is much easier and so the GPU can fill its frame buffers and be waiting for the next info from the CPU much more quickly than it can at higher resolutions as obviously there are many more pixels to colour shade etc.
If you increase your res and the FPS go down then that shows that the GPU isnt up to the extra workload and indicates that it was the problem.
If the FPS stay the same but gameplay is smoother or if they go up that indicates the CPU was holding the GPU back.

@ Stiffex,
I still dont think this is whats happening but its worth testing anyway if you can. How much Ram do you have and also are you using V- sync ? as ohiou_grad_ says it can have a detrimental effect on your performance, if you have it on try turning it off.

Mactronix


Here are some benchmarks showing CPU limited performance.
a c 130 U Graphics card
a b à CPUs
February 10, 2009 6:41:09 AM




No, thats a cherry picked review showing one sentance that mentions being CPU bound and even then they are not correct because, leaving aside the fact that its not a FPS chart but a demo run time chart; If it were cpu bound the times should have hit a wall, the graph should show all results the same or very similar, but it didnt.
All it showed was the natural increase or rather decrease in execution times you would expect when upgrading to a faster CPU.

Anyway we seem to be taking over Stiffex's thread trying to explain the concept to you so not to be funny but maybe we should take this to PM's if you want to discuss this further.
Mactronix :) 
February 10, 2009 3:15:00 PM

mactronix said:
No, thats a cherry picked review showing one sentance that mentions being CPU bound and even then they are not correct because, leaving aside the fact that its not a FPS chart but a demo run time chart; If it were cpu bound the times should have hit a wall, the graph should show all results the same or very similar, but it didnt.
All it showed was the natural increase or rather decrease in execution times you would expect when upgrading to a faster CPU.

Anyway we seem to be taking over Stiffex's thread trying to explain the concept to you so not to be funny but maybe we should take this to PM's if you want to discuss this further.
Mactronix :) 


You're the one who doesn't seem to understand. :) 

They specifically ran at a low resolution to be CPU limited, to show the differences between the CPUs. BTW, only one out of the 4 charts was run time, the rest were FPS.

Those were newer games run at maximum settings at 1024x768 on an 8800GTX, and the fastest CPU was only E6550 @ 2.33GHz. So according to you the CPU would be struggling, yet the FPS were much higher than 30FPS.

The OP has an E8400 around 4GHz. There is no way that he is CPU bound with 30FPS.
February 10, 2009 3:16:00 PM

Stiffex said:
I have a 4870 gfx and a E8400 cpu, the cpu is clocked at 4ghz but ive dropped it down to 3.85 and the 4870 is clocked at gpu 785 men 1000mhz, ive been playing Crysis and getting good fps of 30 at least i think its good, the game is set at full and the screen res is 1280 x1024.

I have noticed a slight lag every so often like firing a round but it doesnt fire til a split second later and slight freezing and such like, someone said it might be bottleneck between the cpu and gpu but i wunt know.
Can anyone help me sort this out ?


Sorry to have sidetracked your thread. Has your issue been resolved?
a c 130 U Graphics card
a b à CPUs
February 10, 2009 5:06:06 PM

theAnimal said:
You're the one who doesn't seem to understand. :) 

They specifically ran at a low resolution to be CPU limited, to show the differences between the CPUs. BTW, only one out of the 4 charts was run time, the rest were FPS.

Those were newer games run at maximum settings at 1024x768 on an 8800GTX, and the fastest CPU was only E6550 @ 2.33GHz. So according to you the CPU would be struggling, yet the FPS were much higher than 30FPS.

The OP has an E8400 around 4GHz. There is no way that he is CPU bound with 30FPS.



The review is a CPU round up NOT a review on being CPU restricted. The only chart that even mentioned being CPU bound wqas the runtime chart and even then it was "somewhat" CPU bound which is differant to a having a bottleneck.

I have been at pains all the way through to point out that i dont think that the OP is CPU bound.

You are the one who asked how increasing the resolution would help and i tried to explain it to you. The fact that you asked proves you dont know what you are talking about.
Now you are trying to tell me i dont understand it :pfff: 
I tried explaining it in simple terms and i tried going into some detail but its clear the concept escapes you.

Mactronix
February 10, 2009 7:00:14 PM

mactronix said:

I have been at pains all the way through to point out that i dont think that the OP is CPU bound.

Since he's not CPU bound, then how on earth will higher resolution increase FPS?
a b U Graphics card
a b à CPUs
February 10, 2009 7:58:43 PM

I'll take my example from my i7/AM3 discussion. All numbers are made up to prove a point.

The CPU needs to feed the GPU, which does all the rendering. Lets say a CPU can send enough graphical data to draw 45 FPS, and sends that to the GFX card. Now, if the GFX card can render all the sent data at the rate of 55 frames a second, you have a CPU bottleneck, as the GPU is rendering faster than the CPU can send the GPU data.

The reverse is also true. Using the above example, now the GPU can only render 30 FPS. Thats a GPU bottleneck; the GPU is rendering less frames then the CPU is sending to it.


The easiest way to find where the bottleneck is would be to reduce resolution (not settings initially) one notch, and look for any increase in performance. If not, its likely the CPU is the bottleneck. Note, that increasing the resolution is not a valid test.

Most AM3/i7 "gaming" benchmarks are invalid, as any GPU setup used is an inherent bottleneck on the system, and most modern CPU's will result in the same FPS for the given game (hence, all the 2580x1600 comparisions where every setup give 23 FPS across the board; thats a GPU bottleneck).



My advice: Turn ONLY resolution down one notch, and compare FPS and look for any changes in how smooth the gameplay is. If FPS jumps by more than a few FPS, its probable you have a GPU bottleneck (with Crysis, that quite easy to accomplish). If FPS is more or less the same (<3 FPS increase), you probably have a CPU bottleneck.

I'm putting money on a GPU bottleneck. This one seems clear to me.
February 11, 2009 3:28:29 AM

gamerk316 said:
I'll take my example from my i7/AM3 discussion. All numbers are made up to prove a point.

The CPU needs to feed the GPU, which does all the rendering. Lets say a CPU can send enough graphical data to draw 45 FPS, and sends that to the GFX card. Now, if the GFX card can render all the sent data at the rate of 55 frames a second, you have a CPU bottleneck, as the GPU is rendering faster than the CPU can send the GPU data.

The reverse is also true. Using the above example, now the GPU can only render 30 FPS. Thats a GPU bottleneck; the GPU is rendering less frames then the CPU is sending to it.


The easiest way to find where the bottleneck is would be to reduce resolution (not settings initially) one notch, and look for any increase in performance. If not, its likely the CPU is the bottleneck. Note, that increasing the resolution is not a valid test.

Most AM3/i7 "gaming" benchmarks are invalid, as any GPU setup used is an inherent bottleneck on the system, and most modern CPU's will result in the same FPS for the given game (hence, all the 2580x1600 comparisions where every setup give 23 FPS across the board; thats a GPU bottleneck).



My advice: Turn ONLY resolution down one notch, and compare FPS and look for any changes in how smooth the gameplay is. If FPS jumps by more than a few FPS, its probable you have a GPU bottleneck (with Crysis, that quite easy to accomplish). If FPS is more or less the same (<3 FPS increase), you probably have a CPU bottleneck.

I'm putting money on a GPU bottleneck. This one seems clear to me.


Very well said. :) 
a b U Graphics card
a b à CPUs
February 11, 2009 11:23:31 AM

Thanks. I'm just tired of seeing all these "Why doesn't gaming imrpove when I upgrade the CPU" and "Where is the bottleneck" discussions.

Changing the resolution does more to tax the GPU then the CPU, which is why its such an effective way to locate a bottleneck. If decreasing the res one notch (in your case, to 1024x768) improves FPS (which it will), you have a GPU bottleneck. If not, then the CPU isn't strong enough to feed that card, and you have a CPU bottleneck.

Crysis, besides being badly coded to begin with, taxes your GPU more than any other card. I still feel that no setup should attempt that game above 1600x1200, as no GPU setup can keep up.


Granted, HD speed and RAM can play a part, but CPU/GPU is where the main holdups happen (at least until the Shared Memory Bus bottleneck happens in a few years...and we have to re-design computers as we know them...)
a c 130 U Graphics card
a b à CPUs
February 11, 2009 6:46:37 PM

theAnimal said:
Since he's not CPU bound, then how on earth will higher resolution increase FPS?


I never said it would, mousemonkey sugested trying it. Then you asked how it would help and i tried to explain it to you.
Its not my fault you dont get the princaples involved.

@ gamerk316,
Sorry but you are wrong with your statement. How the hell can reducing the resolution be a "valid" test but increasing the resolution not be a "valid" test :lol: 
Your talking crap it the same thing just in reverse.

The OP is at a low resolution already, Mousemonkey suggested trying to up the res to see what happened. the Animal dosent understand the concept and it would seem nether do you.
Both upping and reducing the res are valid tests, depending on the situation. I would go as far as to say that to get a true picture of where you are you need to do both anyway.

Mactronix
a b U Graphics card
a b à CPUs
February 11, 2009 7:19:06 PM

mactronix said:
I never said it would, mousemonkey sugested trying it. Then you asked how it would help and i tried to explain it to you.
Its not my fault you dont get the princaples involved.

@ gamerk316,
Sorry but you are wrong with your statement. How the hell can reducing the resolution be a "valid" test but increasing the resolution not be a "valid" test :lol: 
Your talking crap it the same thing just in reverse.

The OP is at a low resolution already, Mousemonkey suggested trying to up the res to see what happened. the Animal dosent understand the concept and it would seem nether do you.
Both upping and reducing the res are valid tests, depending on the situation. I would go as far as to say that to get a true picture of where you are you need to do both anyway.

Mactronix


My test is determined to find if any bottleneck exists at a specific resolution. Going up a resolution is not a valid test, as the extra workload put on the GPU combined with the fact most games are GPU limited make it impossible to compare FPS and determine if a CPU bottleneck exists. If frames drop, you know its probably a GPU bottleneck, but what if FPS stays the same (or near the same)? Is that the most the card can render (GPU bound), or is the CPU simply unable to keep up to the card (CPU bound)?

By reducing the resolution, you reduce the work the GPU has to do, while only slightly lessing the work of the CPU. As a result, you can tell by the FPS between two resolutions weather or not you are CPU/GPU bound at that resolution. (A big FPS gain of indicates a GPU bottleneck, no FPS gain indicates a CPU bottleneck).
a b U Graphics card
February 11, 2009 7:46:05 PM

I dont think theirs any "real" bottleneck. Did you try to reinstall drivers? what psu do you have?
February 11, 2009 10:24:53 PM

mactronix said:
I never said it would, mousemonkey sugested trying it. Then you asked how it would help and i tried to explain it to you.
Its not my fault you dont get the princaples involved.


I get the principles of bottlenecking and provided a link to back it up which you dismissed because it didn't fit your theory.

My initial question was essentially a rhetorical one.
a c 130 U Graphics card
a b à CPUs
February 12, 2009 6:12:03 PM

theAnimal said:
I get the principles of bottlenecking and provided a link to back it up which you dismissed because it didn't fit your theory.

My initial question was essentially a rhetorical one.


Thats what i thought all along actually you were just being pedantic for the sake of argument :non: 
a c 130 U Graphics card
a b à CPUs
February 12, 2009 6:42:53 PM

gamerk316 said:
My test is determined to find if any bottleneck exists at a specific resolution. Going up a resolution is not a valid test, as the extra workload put on the GPU combined with the fact most games are GPU limited make it impossible to compare FPS and determine if a CPU bottleneck exists. If frames drop, you know its probably a GPU bottleneck, but what if FPS stays the same (or near the same)? Is that the most the card can render (GPU bound), or is the CPU simply unable to keep up to the card (CPU bound)?

By reducing the resolution, you reduce the work the GPU has to do, while only slightly lessing the work of the CPU. As a result, you can tell by the FPS between two resolutions weather or not you are CPU/GPU bound at that resolution. (A big FPS gain of indicates a GPU bottleneck, no FPS gain indicates a CPU bottleneck).


Look, we are both coming at the same issue but from different ends is all.
If you increase the resolution and still get similar FPS then obviously you couldn't have been GPU restricted at the lower resolution as if you were the FPS would have dropped at the higher resolution. No ?
Then if the fps increase then that shows you were CPU bound slightly, and if they stay relatively the same that shows that you are CPU bound and upping the resolution doesn't help. And upping the resolution is the next step along when you have tested it your way around anyway so you are just saving time is all.
Ideally as has been said already you want to be using a couple of different resolutions for testing purposes which ever way around you do it.
To be honest i wish i hadn't bothered i only posted because i thought theAnimal was genuinely curious and wanted to know why upping the resolution was suggested. I though it was a bit facetious when the posts came back nit picking about the ratio.

Mactronix :) 
!