GeForce Titan: 2500K vs 4790K (66 games tested)

[BFG10K]

Introduction

I upgraded my 2500K to a 4790K and tested 66 different games on a Titan whose results I wanted to share today. As usual, the settings used are the actual settings I play these games at.

If you see lower than expected scores (especially in older games), it’s because most tests have been designed to show the worst-case scenarios; regular gameplay will often run much faster.


Setup

· Intel i5 2500K.
· Gigabyte P67A-UD3-B3.

· Intel i7 4790K.
· Asus Z97-K.

· GeForce Titan, driver 337.88.
· 16GB DDR3-1600.
· 128GB Samsung 830 (OS).
· 960GB Crucial M500 (games).
· 30" HP LP3065.
· X-Fi XtremeMusic.
· Seasonic X 560W.
· Fractal Arc R2.
· Windows 7 64 bit (SP1).

Everything was run at stock and hyper-threading was disabled on the 4790K. I disable HT because I have a large library of older titles that I frequently play, and it can cause issues with them.

The results portray average frames per second.


Results

Conclusion

Many games are clearly GPU bound at my settings so they experience little to no performance gain, but a few show significant performance improvements.

In particular, games based on the Serious 3.x engine (#12 and #47) show a higher than expected improvement. I checked repeatedly and the results were consistent, and I couldn’t see any error on my part. The engine gets frequent patching through Steam so if I had to guess, I’d suspect it supports AVX2 which the Sandy Bridge lacks. Either that or it’s extremely sensitive to L3 cache size.

 

[r3pshow]

Good job, very useful info...

 

[RussianSensation]

Thanks for the overview. I think it's fair to say that outside of special cases such as BF4 multi-player and Arma 3, strategy games and MMOs, PC gamers should spend $ on a faster GPU setup and then SSD before considering an upgrade from a modern CPU such as 2500K. Once overlcocked to 4.3Ghz+, it's going to be a wash really comparing i5 2500K vs. 4790K @ 4.8ghz. Most people upgrade now for features (M.2 / SATAe) or because they are bored

What would have been interesting to see is the comparison of minimum FPS over a period of time. Let's hope next generation PC games have more NPCs, more advanced AI and physics to give us reasons to upgrade from 1st and 2nd generation i5/i7s.

 

[Arkadrel]

Am I the only one that thinks the improvements are reaaaaally small?

And in the titles where the improvements are big, its usally a older game that runs pretty well before hand as well.

PC gamers should spend $ on a faster GPU setup and then SSD before considering an upgrade from a modern CPU

^ basically what I take away from this guys posting too.
I dont think id bother spending ~350$ for 6% avg performance, where most is from older games.

Most people upgrade now for features (M.2 / SATAe) or because they are bored

yeah.

 

[zebrax2]

The chart look miniscule on my end. Anybody else have this problem?

 

[BrightCandle]

Its even better than gamegpu.ru data that I have been gathering because its all on the same GPU and its a lot of games. That is incredibly useful information and shows just how much difference the CPU performance can make in certain games.

My gather from the gamegpu.ru data for the 4770k v 2600k is quite interesting as well, and done in a similar format (comes from https://dl.dropboxusercontent.com/u/3638175/GameGPU CPU performance.ods) looks like the following:

What I also find quite interesting is the 3960X verses the 4770k comparison because that 6 core v 4 core latest architecture is really the question a lot of people have in regards to gaming CPUs:

We can see in a lot of cases the 3970X is quite often the same. On average it comes out identically, but we can also see it can be 10% slower than the newer architecture. But we also see in just as many cases that it can be up to 12% faster. The bigger releases in the list (BF4, COD, thief watchdogs) all show gains with the 6 core whereas the ones where its slower tend to be smaller releases, although Diablo 3 is definitely an exception in that list. The comparison of the 3970X to the 2600k is much more telling (you can go make that yourself if you like from my spreadsheet) of the difference 6 cores makes but its interesting to see what that old 6 core is still doing.

Your data also suggests that the Haswell-E should bring pretty good improvements in some games, and combined with the 30% (15/50) games showing improvements of 6 core its potentially worth its extra outlay.

 

[BFG10K]

Am I the only one that thinks the improvements are reaaaaally small?

You're absolutely right, and that's the reality of the situation here: I'm mostly GPU limited, even on a Titan. The settings aren't random but are carefully selected because I'm a strong believer in taking the fullest advantage of a GPU.

I also completely agree on spending the most you can on a powerful single GPU because most games are GPU limited at reasonable settings. This is what I've been saying for years.

With that said, I'm not disappointed in any way. In fact it's better than I expected as more games showed an improvement than I thought. So in the corner cases where I'm CPU limited, I can expect some good gains. Also after 3 years it's nice to refresh my CPU/platform.

The chart look miniscule on my end. Anybody else have this problem?

Click on the yellow line at the top of the chart to view normal size. It's big even on my 2560x1600. :awe:

 

[zebrax2]

No such thing on my end (Firefox and IE)

 

[BFG10K]

I'm on IE11 and it's there. This forum has always handled images this way.

Here's the direct link: http://s17.postimg.org/3m6w1vn7z/Chart.png

 

[zebrax2]

This is what i see

Edit: Even the direct link leads to an image that size

 

[guskline]

EXCELLENT work BFG10K. Really shows the staying power of the 2500k!

 

[tidy]

.

 

[n0x1ous]

thanks BFG - very interesting!

 

[guskline]

BFG10K: If HT was enabled for the 4790k what games of the ones you tested would not run? I'm curious as to what effect turning off HT affects fps.

 

[tidy]

.

 

[Sable]

Gives me a sense of smug satisfaction, thanks BFG.

 

[Zardnok]

This is what i see

Edit: Even the direct link leads to an image that size

Hit "Ctrl +" a few times to increase the size. You can "Ctrl -" back to normal text size when done.

 

[BrightCandle]

Gamegpu.ru has data for the 4670k and the 2500k which we can compare to the 2600k and the 4770k. There are other differences there obviously, its not just HT that separates these CPUs but its also real CPUs that real people buy and not some theoretical comparison on the 4770k with HT turned off. I doubt many people buy a 4770k over the 4670k with the intention of turning off HT to get to the clock speed and other differences, they determine HT isn't worth it and buy the cheaper 4670k instead.

So first up Sandy bridge:

Haswell:

So in both cases we can see only a few games with negative scaling, and its relatively minor at 5%. On the other hand the upside is worth above 20% gains, and the average gain is a modest 4% in both cases, but that doesn't tell the whole story because just like 6 cores about 7-8% of them gain more than 10%. Its not a massive difference in the average case but in a few games it really does help quite a bit. Just like the 6 core they all seem to be big release names as well like BF4.

Edit: I think these wide game reviews show that unless you review a CPU with a lot of games its really hard to determine there are differences and what they really make at the usual played settings. These review sites that test 4 games aren't possibly going to show the wide differences and lack of differences when compared to the 126 games worth of data we have on some sites.

 

[BrightCandle]

The chart look miniscule on my end. Anybody else have this problem?

In Chrome I get the option to resize it just above the image. But I find if I right click and open the image in a new tab it comes down full size as well, so you might want to try saving it locally or just opening the image in a tab of its own. The main image is certainly readable its just anandtech's forums shrink them down with css.

 

[spinejam]

Thanks for the info guys --- much appreciated!

 

[Dribble]

World of Tanks is a good one as that is completely cpu bound if your gpu is decent - check out the differences in BrightCandles tables. Very popular game that basically is still using a single threaded engine...

 

[dangerman1337]

Shouldn't you have games like Total War or MMOs or larger scale FPSs such as Planetside 2 or Battlefield 3/4?

 

[Sohaltang]

The chart look miniscule on my end. Anybody else have this problem?

Click the bar that says "click bar to increase size"

 

[guskline]

BrightCandle: Thank you for your postings. Cleared up my question.

 

[Attic]

You da man BFG10K.

I just upgraded to a 4690k yesterday from i3-4330. My main concern was to go i7 for HT or not. Given it's roughly $100 I think with your look here we can say for a majority of folks buying a CPU for gaming that the i7 is not worth it, but definately get an i5 over any i3.

That being said, i7 4790k is still worth it for someone who just wants to drop in a 4ghz proc and not worry for a long time and some games like Crysis 3 do benefit a decent amount from i7 HT.

For 60fps mins, where uncompromising gaming begins you still gotta get the fastest CPU you can afford and I think the 4790k is pretty good bang for the buck in that arena as there's really not much that will do 60fps mins 99.9% of the time on modern games apart from it.