Can the CPU bottleneck the GPU ?

[dave_beast]

Whoever above said that an i3 would bottleneck hgih end GPUs. Let's see the evidence, because I can link to a review where they tested i3/i5/i7 and in most situations fps was the same accross all 3 cpus. When I say the same, literally often identical. It was quite an eye opening review.

The reviewer from the review you read is obviously justifying not spending money on a SB/IB setup. An I3 will bottleneck, the low fps will be lower and and the top end it will lose some FPS, maybe not enough to make it game changing, but it will still be bottlenecking the GPU.

There's 2 situations though,
1. Bottlenecking where there is minimal loss, but still bottlenecking
2. Bottlenecking where drastic amounts of FPS are lost on minimum and maximum FPS.

 

[andybird123]

here is one example of where a different CPU or at the very least massive overclocking would really help

http://lab501.ro/placi-video/gigabyte-radeon-hd-7970-oc-partea-iv-scalare-cu-procesorul/17

http://lab501.ro/placi-video/gigabyte-radeon-hd-7970-oc-partea-iv-scalare-cu-procesorul/18

I would note that the settings used to produce this effect are rather extreme (8xAA!)
by turning down AA you would very easily have playable settings with a 7970 on a much lower processor than an overclock 3980X!

but then equally, here's another case with extreme settings where no matter what CPU you use you get the same result;
http://lab501.ro/placi-video/gigabyte-radeon-hd-7970-oc-partea-iv-scalare-cu-procesorul/21

so as you can see, you can't take similar processors and similar graphics cards and similar games and extrapolate a clear answer - you pretty much have to take it on a case by case basis to determine in what situations a CPU will or won't be a bottleneck

 

[Robes]

here is one example of where a different CPU or at the very least massive overclocking would really help

http://lab501.ro/placi-video/gigabyte-radeon-hd-7970-oc-partea-iv-scalare-cu-procesorul/17

http://lab501.ro/placi-video/gigabyte-radeon-hd-7970-oc-partea-iv-scalare-cu-procesorul/18

I would note that the settings used to produce this effect are rather extreme (8xAA!)
by turning down AA you would very easily have playable settings with a 7970 on a much lower processor than an overclock 3980X!

but then equally, here's another case with extreme settings where no matter what CPU you use you get the same result;
http://lab501.ro/placi-video/gigabyte-radeon-hd-7970-oc-partea-iv-scalare-cu-procesorul/21

so as you can see, you can't take similar processors and similar graphics cards and similar games and extrapolate a clear answer - you pretty much have to take it on a case by case basis to determine in what situations a CPU will or won't be a bottleneck

If you look at those Arkham City ones though you'll notice even the i7 2500k struggles, I suspect that was probably a second or two of some major physics that caused that low min FPS, but that's the problem with benchmarking I guess it doesn't give you a feel of how it actually plays.

 

[andybird123]

If you look at those Arkham City ones though you'll notice even the i7 2500k struggles, I suspect that was probably a second or two of some major physics that caused that low min FPS, but that's the problem with benchmarking I guess it doesn't give you a feel of how it actually plays.

it's stock 2500k for a start and most people run them OC'd quite heavily

my point was, you can find a case to support either side of the argument very easily - the mistake people make is to find one case (or two or however many) and use that to apply crass generalisations, where as you need to take it on a case by case basis - what if the person asking "will I bottleneck" doesn't play batman - then the answer might be no (this is the one most people would give)

but if batman is the only game he wants to play all day every day on 8xAA then the answer might be yes! without more info people will jump to their preferred conclusion without thinking it through

 

[Robes]

it's stock 2500k for a start and most people run them OC'd quite heavily

my point was, you can find a case to support either side of the argument very easily - the mistake people make is to find one case (or two or however many) and use that to apply crass generalisations, where as you need to take it on a case by case basis - what if the person asking "will I bottleneck" doesn't play batman - then the answer might be no (this is the one most people would give)

but if batman is the only game he wants to play all day every day on 8xAA then the answer might be yes! without more info people will jump to their preferred conclusion without thinking it through

True, agreed - A stock i7 2500k obtained 5fps more than a stock i3 2100.

My point was that it may have been one seconds worth of frames that caused that and may not have actually made any difference to game play (between CPUs) but the results in this instance don't highlight that.

But yes I agree with what you're saying, every situation is different, all you can do is read as many reviews as possible and draw your own conclusion.

 

[andybird123]

+1 to that - or even just test it yourself
as you say, what a benchmark on the internet might show up as a potential problem might not be once you get everything together and just play the game

 

[jaybee]

Well as you're not freely linking to yours here's mine:

http://lab501.ro/placi-video/gigabyte-radeon-hd-7970-oc-partea-iv-scalare-cu-procesorul

In Romainian but you can see the graphs or it you're really interested you can translate in Chrome.

Also

http://www.guru3d.com/article/radeon-hd-7970-cpu-scaling-performance-review/1


Here's the sort of thing people will sometimes highlight as bottlenecking:

the 1280x1024 results show CPU scaling this means at that resolution the lower end CPUs are bottlenecking the GPU. But there are two flaws with saying this, firstly 70fps on the lower end CPU is more than adequate and secondly, who plays at 1280x1024 on a 7970!?

On the more realistic gaming resolutions you can see that the GPU has become the bottleneck as they're all pretty much the same.

In the graphs you showed, none had an i3 down on the test so... ?

Of the links you did provide, it clearly shows the i3 compared to an i7 is almost identical in the majority of tests. BF3 = identical. Other tests showed the i7 getting sometimes 8% more fps than the i3. That's not a bottleneck.

I can't find that article I was saying about. Someone linked to it on here. I will keep looking...

 

[dave_beast]

Other tests showed the i7 getting sometimes 8% more fps than the i3. That's not a bottleneck.

That is a bottleneck then!

 

[jaybee]

That is a bottleneck then!

I define a bottleneck as when a cpu is significantly preventing a graphics card from performing. An i7 getting 3 fps more than an i3...it's expected. It's not a bottleneck as I would describe one. A true bottleneck would be say an old single core CPU from way back performing at 100% constantly, and the gpu at 5% with much lower fps.

The term bottleneck as you are defining it is confusing for people, as you are basically saying an i5 could be a bottleneck just because an i7 could get 1fps more in any test. That's not helpful. It's an4l and a poor representation of the situation.

 

[andybird123]

+1... to me, the term bottleneck should be reserved for cases where one thing is clearly having a MAJOR effect on FPS and needs to be changed to allow good frame rates

I mean even the entomology of the word, look at a bottle, the neck is tiny and the contents of the bottle huge, meaning that the neck is preventing the contents from flowing freely... 3FPS or 8% or anything tiny like this is irrelevant - we should only be calling something a bottleneck if it's making the difference between being playable and not being playable - if changing 1 component allows a change from non-playable to playable then clearly that component is a bottleneck

if a CPU is a bottleneck in one particular game then you seriously have to question if that 1 game (or even few games) are worth a full system rebuild vs. a GPU upgrade that might get you good frame rates in the majority of games (it might be to some people but that's a judgement call based on cost/benefit)

 

[Kaapstad]

If I can run a game with everything maxed out and get good minimum frame rates I really dont worry if my cpu is err bottlenecking or not. For games I leave it on 4.0ghz for benchmarks I run it higher.