What is bottlenecking?

[symbi0nt]

Quick question, when these reviews talk about GBU bottlenecking and CPU bottlenecking, what do they mean? What causes each of them?

Am I right in assuming high resolution gaming maxes on the GPU, and so the benchmark is more dependant on the CPU instead? If this is the case why don't reviewers use awful GPUs instead of nvidia 9 series?

Thanks.


And while i'm at it, I usually use 10x7 resolution. Is this abnormal? Why do I see loads of reviews using higher? What do you guys all use?..
Anyway, would my res mean shelling out extra case for a better cpu is a bit pointless?

 

[Robteam70]

Most people game at 1280 or 1600 because that is the native rez for most non-widescreen monitors. I use 1280 - you will generally get the best picture quality using the native resolution of an lcd monitor.

 

[symbi0nt]

A bottleneck is a part that causes performance loses. Games depends on the GPU, especially at higher setting. At higher settings, the CPU isn't a huge factor in performance.

For example, you have dual 7900GTs and 1gb of RAM, and try to play BF2 at max settings. 1gb of RAM will cause chippoy gameplay, since BF2 uses more than 1gb of RAM at higher settings.

So when looking for -good- conroe benchmarks, I should be looking for ones with a low resolution, even if they have high end gpus?
I would have thought the opposite, since the GPU would be able to run low settings very easily and max out the fps?

 

[waylander]

Bottlenecking is when one component of the entire system holds back the performance of the whole... kind of like the weak link in the chain.

For gaming usually the GPU is the bottleneck as the CPU and RAM are not taxed as much as the GPU. In some instances, like the first response, the RAM can be the bottleneck.

Even high resolution gaming is all about the GPU, the CPU is almost never at 100%. Plus if you were trying to find out whether the CPU was the bottleneck then you'd want the best GPU so that the GPU wasn't the bottleneck, make sense?

They also test at higher resolutions because this stresses the system even more, also, some GPU's perform better at higher resolution than others. At 1024 x 768 even mid level cards can pump out really good FPS, so the difference between a mid level card and a high level card is not so apparent. If you go to 1600 x 1200 then you really see how different a $300 card is from a $600 card.

 

[TechnologyCoordinator]

Am I right in assuming high resolution gaming maxes on the GPU, and so the benchmark is more dependant on the CPU instead? If this is the case why don't reviewers use awful GPUs instead of nvidia 9 series?

No, you are wrong.

If you run at a high resolution your GPU is maxed out and your CPU will not be used to its fullest capacity.

If you run at lower resolutions your CPU will be maxed out and your GPU will not be used to its fullest capacity.

That's why CPU benchmarks at lower resolutions show actual CPU performance better. For example, today's Conroe benchmarks showed it with a greater gain at lower resolutions.

No matter what you do you'll have a bottleneck in your system. Try to match your hardware accordingly and just adjust settings after that to get the best performance possible.

In my system (see sig for specs) the CPU is the bottleneck, I've got a slow-mo Athlon 64 3200+ paired up with a mid-range GeForce 7800 gt.

 

[MrCommunistGen]

A bottleneck is when one (or more) components in a system are slower than the rest and cause the performance to drop despite the fast component.

Edit: apparently a lot can be said in 12 minutes
-mcg

 

[symbi0nt]

Thanks guys. I think I understand a bit better now.

One more thing though;

That's why CPU benchmarks at lower resolutions show actual CPU performance better.

I came across this image as part of a 4-piece-set mostly displaying the same thing 4 times. Does this mean that although I should avoid a low end cpu for bottleneck reasons, the cpu really won't have a profound impact on games in the long run?

 

[ikjadoon]

I agree that playing at lower resolutions will get you a more precise measure of CPU performance, but nobody really plays at those resolutions. OK, more than "nobody", but a small percentage. So a benchmark run at 1280x1024 with all different processors, but a fast enough GPU will show that they all have almost the exact same frame rate. This article is case-and-point:

[H] Enthusiast

A computer is only as fast as the slowest part.

~Ibrahim~

 

[TechnologyCoordinator]

I agree that playing at lower resolutions will get you a more precise measure of CPU performance, but nobody really plays at those resolutions. OK, more than "nobody", but a small percentage. So a benchmark run at 1280x1024 with all different processors, but a fast enough GPU will show that they all have almost the exact same frame rate. This article is case-and-point:

[H] Enthusiast

A computer is only as fast as the slowest part.

~Ibrahim~

I agree.

It's still important to have a fast CPU so that when that Geforce 8X00 comes out you'll be ready to take advantage of it if you want.

 

[symbi0nt]

I think I get it.

When the res/detail is low, the GPU is going slow and most CPUs even the old ones can keep up, thus making the game play at the GPU's standards. When detail/res is up high, the GPU is flat out but sometimes the CPU cannot 'keep up' with the GPU, and so the better the CPU, the more the CPU can 'keep up' with the flat out GPU?

 

[RichPLS]

I agree that playing at lower resolutions will get you a more precise measure of CPU performance, but nobody really plays at those resolutions. OK, more than "nobody", but a small percentage. So a benchmark run at 1280x1024 with all different processors, but a fast enough GPU will show that they all have almost the exact same frame rate. This article is case-and-point:

[H] Enthusiast

A computer is only as fast as the slowest part.

~Ibrahim~

Are you suggesting then that we do not need the kind of extreme performance Conroe offers, and should be willing to pay more money for less performance? What about the next generation of video cards supporting DX10? As video card performance increases eliminating them as the bottleneck, higher gaming resolutions will then be dependant on faster CPU's...

 

[ikjadoon]

No, I am all for Conroe and these price wars. I'm a consumer, just like you. I'm just saying in pre-conroe times, so to speak, moving from a 3000+ to a 4000+ would not have done as much good as going from a 7300 to a 7900. Do you get what I mean?

There will always be bottlenecks, I believe. When we get DX10 cards, we will see DX10 games which will stress the cards, I'm sure.

~Ibrahim~

 

[ikjadoon]

I think that is backwards, but I'm not sure. I think it is like this:

Most games are dependent on the GPU versus the CPU. So when a company shows a game trying to reflect CPU performance, they have to lower the resolution so that the card has little work to do, and then the CPU differences are there. When we bumb the res. the game will, of course, run slower, but is it the GPU slowing it down or is the CPU?

We are eliminating any factor's that the GPU plays so to find what the "real" power of the CPU is. I don't find this very realisitic because, as I mentioned earlier, few people play at that resolution and gaming benchmarks should ALWAYS be run at the resolutions that people are playing them at today. That is why gaming benchmarks don't always tell the full story of a CPU. Case-and-point is Conroe. It has about the same performance in games as do the AMD's, but in other things it is bounds and leaps ahead.

~Ibrahim~

 

[symbi0nt]

So what resolutions do you guys use? Do people actually game at 1600x1200!? I'm using 1280x960 now and that's the highest my monitor will go. I'm not sure I even want to go higher anyway.

At this sort of resolution, is there any need to switch, for example a E6300 for a E6600? Bearing in mind that gaming is pretty much what I do most.

 

[ikjadoon]

Well, most people play at their monitors highest resolution, if the GPU allows. That limits most people to around 1024x768 and 1280x1024. After that, the most common gaming resolution is 1600x1200. That gets into WideScreen gaming, which I don't prefer for FPS games. I like the regular 5:4/4:3.

Remeber, you need a faster GPU to run at higher reslutions. The bump between the 6300 to the 6600 is somewhat large, and it will give you some gain in FPS, but with a faster GPU and the upgrade the difference will be much larger. Just getting a faster GPU can help you, but then you don't want to bottleneck the processor, because the GPU can go faster, but the CPU can't keep up with it. To exaggerate this is to put a 7900GTX inside a computer with a Sempron 2100+. The 7900GTX can push some amazing frames, but the Sempron is holding it back.

In gaming I think the relationship between CPU/GPU is something like GPU=65% and CPU=35%.

~Ibrahim~

 

[xsandman]

I think I get it.

When the res/detail is low, the GPU is going slow and most CPUs even the old ones can keep up, thus making the game play at the GPU's standards. When detail/res is up high, the GPU is flat out but sometimes the CPU cannot 'keep up' with the GPU, and so the better the CPU, the more the CPU can 'keep up' with the flat out GPU?

It is more like the opposite. THink about having 2 highways, a graphics highway and a cpu highway. Graphics have to take the graphics highway and the speed limit is the performance of the card (which is much higher than the CPU highway because ONLY graphics are allowed on the highway and the highway is designed just for graphics). The physics has to take the cpu highway, the speed limit is the performance of the cpu (which is unknown). Think of the resolution as the amount of traffic. Now you have a large party of both graphics and physics that are trying to get from point A to point B. The party can not start until everyone is there. Since the speed limit is much higher on the CPU highway, if there is no traffic (640x480), then the physics take the longest because they can only go as fast as the speed limit on the CPU highway. (The unknown speed limit can be determined by dividing the distance by the time it took to start the party). Now if you start adding traffic to the graphics highway (1280x1040), then it will make it harder for the graphics to beat the physics to point B. If the graphics can't beat the physics to the party, then the party will have to start later, but it is no longer the fault of the CPU.

 

[the_guru]

I agree that playing at lower resolutions will get you a more precise measure of CPU performance, but nobody really plays at those resolutions. OK, more than "nobody", but a small percentage. So a benchmark run at 1280x1024 with all different processors, but a fast enough GPU will show that they all have almost the exact same frame rate. This article is case-and-point:

[H] Enthusiast

A computer is only as fast as the slowest part.

~Ibrahim~

Are you suggesting then that we do not need the kind of extreme performance Conroe offers, and should be willing to pay more money for less performance? What about the next generation of video cards supporting DX10? As video card performance increases eliminating them as the bottleneck, higher gaming resolutions will then be dependant on faster CPU's...
During the last years the CPU has become less and less important and the GPU more and more important when it comes to gaming. This trend will most likely continue which means that when the DX10 cards are released the games will be even more GPU-demaning and no more CPU-demanding. Unless of course the developers choose to put more advanced calculations on the CPU... But we know nothing of that so if the trend continues there will be no use for better CPUs in gaming.

 

[monkey999]

would increasing cpu ferquency lessen a bottle neck caused by the cpu

 

[xsandman]

would increasing cpu ferquency lessen a bottle neck caused by the cpu

correct, if the CPU is the bottleneck. But in games, the GPU is the bottleneck. So if you increase the frequency of the GPU, it will lessen the bottleneck in games.

 

[ikjadoon]

Yup. If you already have a good CPU, i.e. 3200+ and above or 3.2+ P4, then overclocking won't help too much, but every little bit helps. I would first OC the GPU, and if the performance isn't still what you were looking for, a CPU OC would be next. Both can yield higher FPS, but the GPU OC'ing is usually more FPS gain per Mhz.

~Ibrahim~

 

[symbi0nt]

What's the point in getting a good cpu for a gaming build then?

Edit - Well, I mean a E6600 rather than a E6300 or 3800X2 for example.

 

[ikjadoon]

Not very much. If you look at this Gaming Conroe review, you will notice exactly why:

[H]ard OCP

An OC'ed X2 performs on par or so with a Conroe build.

~Ibrahim~

 

[symbi0nt]

Don't high res benchmarks focus more on the GPU though?

Using the example of the traffic that was explained to me earlier in this thread - the CPU would get there first and we waiting for the GPU - which would arrive at the same time regardless of CPU?

 

[RichPLS]

Yes, if HOCP would have used another xfire video card, which the had, then Conroe would be leading by a large margin again... that review is biased, and they even lowered setting on the AMD rig to compete otherwise the AMD rig would not complete.
Also they did not point out how fast boots and load times were with Conroe, it was more to bolster die hard fanboi's bruised ego's, and could have been done in a much more scientific way...

 

[ikjadoon]

Well, what do you mean by Hi Res?

If you want to find out CPU performance, run a low res. game.

If you want to find out REAL performance, run at the resolution everyone is playing at, which is around 1280x1024 or 1024x768.

Right. You are exactly right. The CPU would be done with it's few little tasks and then you are waiting for the GPU to finish what it needs to do. But you've got to remeber, those low. res. benchmarks should NOT be read as what you will get if you choose the Conroe over a X2. Read the benchmarks at what resolution you will be playing at.

~Ibrahim~