CPU for Gamers : 30 CPUs compared

[DGLee]

I tested & measured gaming performances of "all" properly-existing CPUs in the market. Criterion for determining 'properly-existing' is that, a product shouldn't be being replaced by latter SKU at the same price level(ex: IVB is replaced by Haswell). Hope that could help those who are considering buying or upgrading one's gaming machine. Here we start.

Test system is composed as follow:
(+ VGA : GF 780Ti)

image redacted


Since the host provider (of my blog) consider sudden traffic increment as an abusing, I'm going to upload the summary part of all results only.
Here's the performance summary at 1680 x 1050 resolution. The former one emphasizes just a sum of framerates of all games while the latter one implies average relative performance of each CPU. Of course the latter one is more important.

image redacted


Here's for 1920 x 1080 resolution.

image redacted

Here's for 2560 x 1600 resolution.

image redacted

You have been told not to show links to your own site. Next time is permaban
Markfw900
Anandtech Moderator


Well, the article is over. Thanks for reading this.
Have a nice day!

 

[NTMBK]

You only measured average frame rate! You really need to be looking at frametime latency, so you can measure smoothness as well. Take a look at this to get an idea why: http://techreport.com/review/21516/inside-the-second-a-new-look-at-game-benchmarking

 

[psolord]

great

thanks

I'd like to see some performance comparison with Tridef.

It can be a real burden on cpus.

 

[bononos]

You only measured average frame rate! You really need to be looking at frametime latency, so you can measure smoothness as well. Take a look at this to get an idea why: http://techreport.com/review/21516/inside-the-second-a-new-look-at-game-benchmarking

There actually was an article on cpu impact on framerate latency. The testbed used CC 12.3 + 7950. Piledriver was an improvement but still lagging behind Intel. Trinity was a disappointment. It would be interesting to see newer measurements with updated drivers + mantle and with a Nvidia graphics card.

http://techreport.com/review/23750/amd-fx-8350-processor-reviewed/5

 

[DrMrLordX]

You only measured average frame rate! You really need to be looking at frametime latency, so you can measure smoothness as well. Take a look at this to get an idea why: http://techreport.com/review/21516/inside-the-second-a-new-look-at-game-benchmarking

Minimum frame rates would also have been nice, though in some ways, that'd be redundant to what you requested.

 

[you2]

Still it is amusing to see how well the G3220 (G3240) (a < $100 part) compared to an I7. Also it seems like I5 is the sweet spot.

 

[Techhog]

Minimum frame rates would also have been nice, though in some ways, that'd be redundant to what you requested.

Minimum frame rate would be more CPU-limited though, wouldn't it? That would work out.

 

[NTMBK]

The problem with minimum framerate is that it can overstate the problem of a single high latency frame, and not show up repeated stuttering.

 

[Yuriman]

Thanks for the charts! Very informative.

 

[superrockie]

How did you get haswell refresh? I thought the releasedate was 11 may?

 

[SlowSpyder]

Nice job with the benches. Interesting to see how things stack up.

 

[crashtech]

Thanks for the interesting info, OP.

Remarkable to see how well the as yet unreleased i3s perform, though as many will be quick to point out, running the game is about all they will manage, so don't have any other tasks going.

 

[DrMrLordX]

Minimum frame rate would be more CPU-limited though, wouldn't it? That would work out.

Not . . . necessarily. I must admit that the complex interplay between CPU, platform, GPU(s), and driver still mystifies me at times. But, minimum frames could come from some kind of an edge-case scenario or driver-related hiccup that only pops up once every minute or so (see article linked by NTMBK), and that may have nothing to do with the CPU.

CPU-bound situations are usually at low resolutions with low quality settings. What happens to minimum frames under those circumstances versus average frames is something I simply can not quantify or qualify at this time.

That being said, much has been made of minimum frame rates in discussions related to gaming CPUs. I believe the implication is that, if a video card is going to fail to produce acceptable framerates, its failure will show up in min and avg fps. If a CPU is going to cause hiccups somewhere, you're more likely to see it in the min fps.

The problem with minimum framerate is that it can overstate the problem of a single high latency frame, and not show up repeated stuttering.

Well, right. But it will give a possible indication of stuttering (isolated or repeated), especially if the reviewer is nice enough to report such phenomena in the associated article. Even if you can't "eyeball" latency accurately, you can usually tell whether or not isolated instances of high frame latency exist just through observation.

 

[Ken g6]

Veery Interestink! I like it.

How did you get haswell refresh? I thought the releasedate was 11 may?

I was wondering this too.

I also have a suggestion: Could you set up a site where a user could check only the games and resolution they plan to play? And then the chart would arrange itself accordingly?

 

[Phynaz]

Thanks, this may be good stuff.

You only measured average frame rate! You really need to be looking at frametime latency, so you can measure smoothness as well. Take a look at this to get an idea why: http://techreport.com/review/21516/inside-the-second-a-new-look-at-game-benchmarking

Agreed, FPS is a rather outdated method of measurement.

How did you get haswell refresh? I thought the releasedate was 11 may?

This is why I say this may be good stuff. It could also be completely made up.

 

[Phynaz]

Could you set up a site where a user could check only the games and resolution they plan to play? And then the chart would arrange itself accordingly?

Linking to that site would be against the posting rules.

 

[Ken g6]

Linking to that site would be against the posting rules.

Well, what are we to do when AnandTech isn't getting it done?

Seriously, they don't even have the i3-4130 in Bench! They need to buy all these CPUs - or at least the highest-clocked ones with matching cache sizes and downclock - and test them against a lot more games than they currently do.

Or we need to go elsewhere for our benchmarks.

 

[Phynaz]

I agree totally, Anandtech's review quality and quanity has been dropping drastically lately, unless you are interested in phones. As a rule though, I'm betting you will not find anything in bench that isn't given to Anandtech for review.

 

[crashtech]

Yeah, the Bench is pretty outdated at this point. But I guess if we don't like it, we can always go post somewhere else.

 

[Gloomy]

Since you guys mentioned Mantle, I may as well post this here:

This is Golmund Railway conquest. The Mantle run was in a packed server, the DX11 run was in a ~30-40 man server, but that shouldn't affect the performance that much (I'll run another match in DX11 if there's interest in making it even). I can also provide the raw data if you folks want it.

CPU: 4770K @ 4.5GHz
GPU: R9 290 @ stock

I think the most telling part, and what confirms the subjective "smoothness" I feel with Mantle over DX11 the most is that DX11 had 686 frames slower than 16.7ms, while Mantle only had 3 over the entire match.

So if you were wondering if there was an actual, perceivable difference between Mantle and DX11 on a high end CPU, then the answer is an emphatic yes... even though the final average only differs by 8-10%.

Pray for a DX12 future. D:

 

[AtenRa]

@ Gloomy,

Could you pls give us a link for that application ?

 

[Gloomy]

@ Gloomy,

Could you pls give us a link for that application ?

http://webwalkers.cz/Windows/FLAcalculator/FLAcalculator.aspx

 

[TeknoBug]

Makes me giggle when I see a couple i3's sitting up there with the FX 8350.

 

[SlowSpyder]

Makes me giggle when I see a couple i3's sitting up there with the FX 8350.

Not that big of a deal... To be fair, at 1080P, the i3 4360 is 90%+ as fast as the i7 4820k, too.

 

[crashtech]

Not that big of a deal... To be fair, at 1080P, the i3 4360 is 90%+ as fast as the i7 4820k, too.

You don't want to start that, next thing you know, someone will say that the Pentium G3450 is 90% as fast as the FX-9590.