Intel haswell i7-4770K preview article-TMSHW

[boxleitnerb]

Turning down the graphics settings reduces the GPU load and your CPU gets more influential to your fps, that's nothing new.

I'm not going to play Crysis 3 at anything but max details with a 3770K + GTX Titan, that's my choice & scenario, and in my scenario we both know that what my chart shows is true and right.

Turning the game settings down will change things ( how much ? depends on the settings & the downgrading level ).
But you can't invalidate results with that argument.

Alt-tabbing out of the game isn't a good idea, sorry, most games do not like that and you'll get inconsistent results even at the very same system clocks & settings.
The 2600K stops scaling there simply due to its IPC, architecture in general and the GPU bottleneck.
As long as we don't have low level access to BF3's engine, we can't tell where the Ivy Bridge increased perf. increase comes from ( could be anything, more efficiency in the Physics calc. thread, etc etc ).

If you're an "eye candy gamer" with a high-end system, you usually game at max details ( in-game ) and AA/TrAA/AF usually, and in that case it is totally indisputable that regardless if you're gunning with a 4.5GHz 3770K or a 4.7GHz 4770K, you'll have the very same performance at high resolutions & maxed game settings ( *maybe some RPG/strategy/mmo games could be an exception, don't know, I said I can't play these kinds of games ).

You can see a plethora of games tested just like BF3 in the chart above here

Why play with max details if that doesn't give you enough fps? Ultimately it's about fps - turning down settings is not a crime
Imagine this case (at max details):

CPU A can do 60 fps avg
CPU B can do 45 fps avg
The GPU can do 30 fps avg. Conclusion -> CPU A and B are equally fast (30 fps). Which is wrong. And not everyone wants to play at 30 fps. Someone who would rather want 60 fps will not know which CPU is better.

Some people need 100+ fps in BF3 for competitive online play. Why should they play at 2560x1600 with maximum details and TrSSAA if their GPU(s) cannot get them those 100fps? That makes no sense. This applies to all games. I'm not saying people should reduce their settings because fps drop below a certain threshold sometimes. I'm saying, a benchmark should show that. Benchmarking CPUs in GPU bottlenecks is a waste of time since it doesn't give you any new information. Your BF3 benchmark may be fine for someone who only needs around 60fps, but for others who need more, it is useless.

Since games can be GPU or CPU bottlenecked depending on where you are in the game, what you're doing and what kind of fps you need, you always need two benchmarks. One that shows what your CPU can ultimately do and one separate benchmark for the GPU. One single benchmark will never give you all the information you need. Only by combining these two benchmarks you can really gauge the performance you can get from a certain CPU/GPU combo in all relevant cases.

If we take the example from above:

CPU benchmark (maximum settings, because they can influence CPU load, but 1280x720, no AA, no AF):
CPU A: 45 fps avg
CPU B: 60 fps avg
Need 60 fps -> get CPU B

GPU benchmark (maximum settings, 1080p, AA+AF):
GPU: 30 fps avg
Need 60 fps -> get a faster GPU or go SLI/CF or reduce some details.

That way you know exactly what components you need before you buy.

Alt-tabbing can lead to stability problems, that much is true. But it doesn't affect performance in most cases. It's so easy to test that - alt-tab out and in of the game 2 or 3 times and see if the fps stay the same (in a static scene of course where other players or events don't skew results even if you do nothing). If the results are consistent, just go ahead and bench. But I agree with you that usually you should have a certain test environment. Savegames are best for that.

The results you linked to are worthless imo. For one thing, in-game benchmarks were used, but these often show much higher fps than you can actually see in the game. That is because not always is AI, pathfinding etc. computed in those benchmarks, but they are rather movies instead of real gameplay. For instance the GTA 4 ingame benchmark shows at least twice the fps that are common in the game. Very misleading. And secondly, on top of that you make sure due to the GPU bottleneck that relevant performance differences are just "hidden". That might be irrelevant in the case of 2600K vs 3770K since the differences would be very small anyway. But the method in general is just wrong.