How much does a fast CPU really help in games? pclab.pl 2012-2016 article

[lehtv]

There's been a thread or two over at VC&G subforum about pclab's collection of gaming benches, comparing graphics cards over 3-4 years... But I decided to look at that article from another perspective. I want to focus on the relevance of CPU overclocking and HyperThreading in current games using current graphics cards.

Typically, this question is answered with average framerates in average sitautions or even in GPU-heavy prerendered benchmarks. The typical conclusion I see is something along the lines of "overclocking is just a hobby, you don't really need it" and "hyperthreading is basically not useful unless you're buying it for productivity, you'll be fine with an i5". I constantly make conclusions like this myself as well, but how true is that really? Since gamers these days are pretty demanding when it comes to motion fluidity, visual immersion, and maintaining smooth 60 fps or even above 60 fps on a 144Hz monitor, I figured performance in worst case scenarios could be used to approximate the quality of the subjective gaming experience. If your framerate is dipping below subjectively acceptable thresholds too often using a midrange CPU, it won't really matter that your average framerate is three frames away from that of someone who uses the same GPU but with a top end overclocked CPU.

A section in the article (pages 11 and 12) compared AMD and NVIDIA CPU overheads. That comparison was done roughly so (google translation):

Tests done all graphics cards on a platform equipped with a very fast processor, Core i7-6700K, which was overclocked to 4.7 GHz. This is undoubtedly a good idea for testing the fastest GPU, but few combine this type of CPU card, say, middle class. [...]

For this reason, we performed additional tests, including Radeon R9 390 and GeForce GTX 970 (both cards are factory overclocked), using the weaker processors: Core i5-4690K @ 4.5 GHz (four cores and four threads), Core i5-4570 (four cores, four threads, 3.2 GHz + Turbo) and a unit in the form of AMD FX-9590 (four modules, eight threads, 4.7 GHz + Turbo). [...]

[W]e performed additional tests, including Radeon R9 Fury X and GeForce GTX-980 and Ti (the first one has been manually tweaked, the second is a model reference), using the weaker processors: Core i5-6600K @ 4.5 GHz and Core i5-4570 (3.2 GHz + Turbo) [...]

We chose four games, in our opinion the most important, where you can find places with fast-paced action, where many opponents, so the processor is of great importance, as well as the place where the player's interaction with the virtual world is limited to exploring the prepared scene.

Basically, what they did was deliberately pick the worst case situations where you would expect the most work for the CPU.

Since this is the CPUs subforum, let's not make it AMD vs NVIDIA. Let's just compare the overclocked Intel CPU's to the stock i5-4570. I collected all the average fps data into a single spreadsheet for a nice overview (you can check the minimum fps numbers in the article, they're not far below the averages):

Avg FPS @ 1080p


FPS numbers alone aren't that interesting, so let's convert the whole thing to:

Fractions of i5-4570 performance


In terms of raw CPU performance, i5 @ 4.5GHz is roughly 30-35% faster than i5-4570, and the potential benefit from HT can be up to 30-40% with heavily threaded apps. How much of this is actually visible in demanding gaming situations?

With GTX 970 the differences are small, not subjectively significant. With R9 390 and faster GPU's, you start to actually see the effects of an overclocked i5 and i7. But on average, the benefit is far less than you'd expect with pure CPU bottlenecking.

If we focus on the worst cases of these worst case scenarios (which I've chosen as any scenario where i7-6700K OC gives more than 5% better performance compared i5-4570, on a GTX 970), differences are better visible, with performance scaling basically linearly with i5-6600K clock speed when using Fury X or 980 Ti:

Fractions of i5-4570 performance - worst cases

Finally let's compare i7 to i5:

i7 OC vs i5 OC- worst case scenarios only


Benefit of i7 over i5 is 8-21% in worst case scenarios, depending on GPU. Some of that benefit is explained by the 200MHz higher overclock, and some by the generational advantage of Skylake over Haswell, which basically makes HyperThreading more or less not relevant even when using a Fury X.

I'll leave the conclusions to be drawn by you guys ... so, discuss.

 

[moonbogg]

The faster the better I say. i5's are kind of DANK these days for a decent gaming rig. Nice post. Thanks for sharing.

 

[Sweepr]

Great post. I'd say Core i5 remains a solid choice for most gaming rigs, of course if you're thinking 'future proofness' you might want to pick the latest-gen Core i7 or HEDT.

 

[TeknoBug]

What an awful waste of a CPU (and electricity) the FX 9590 is. Yup I'd call the i5 too, although I'm mostly happy with an i3 for now (and I came from highly OC'd AMDs for almost 2 decades and a couple i7's).

 

[Blue_Max]

Great info! Looks like in the nVidia gaming world, a faster i5 beats a stock i7. Helps MY purchasing decision at least.

 

[Yuriman]

Wonderful post, thank you. My conclusions:

The FX beats the Haswell non-K i5 exactly zero times in this set of benchmarks. i5 OC really does show advantages, and hyperthreading also shows major advantages. I'd have been interested in an i7-stock vs i5-oc comparison, since these setups probably cost about the same, but from the numbers we can infer that on AMD GPUs, a stock i7 would be about the same as an OC'd i5, with perhaps a slightly advantage to the i7, while with NV GPUs, the OC'd i5 would be a better choice in terms of raw performance.

I know this is not the GPU forum, but AMD's drivers don't do any favors for any CPU. In terms of worst-case scenarios, a slower-on-average NV GPU probably provides a better experience in most cases.

 

[Blue_Max]

https://www.youtube.com/watch?v=2ZxZiksWtRQ

The amazing DigitalFoundry compared the 2500k vs 3770k... [will update after I finish watching]

 

[JimKiler]

What an awful waste of a CPU (and electricity) the FX 9590 is. Yup I'd call the i5 too, although I'm mostly happy with an i3 for now (and I came from highly OC'd AMDs for almost 2 decades and a couple i7's).

I still had a Phenom II 955 until this time last year and although i never played brand new games I was doing just fine with an R9 270 paired with it. But if I wanted to play games on Zero Day i probably would have been frustrated and upgraded prior to my CPU dying.

 

[Mushkins]

What this discussion seems to leave out for some reason is that what *kind* of game (as well as the underlying code behind the game) you are playing greatly influences how involved the CPU is. As we all know, all games are not created equal.

All four of the games in that chart are primarily single-player games. Games with large-scale multiplayer components and MMOs especially are *much* more heavily reliant on the CPU due to all the netcode involved.

In those games an i5 is still more than adequate when paired with a solid GPU, but the performance gap between, say, an i3 and an i5 is one that's large enough to sway purchase decisions and price/performance metrics.

 

[sm625]

There is still a lot of i7-920 class hardware out there and it should be plenty fast for games. The choice between upgrading an old i7 vs putting all available money into a video card is a no brainer. An i7-920 with a 980ti SC should outperform a skylake i7 + GTX970, and cost less.

 

[monkeydelmagico]

Seems to confirm the premise I've been using for around 10 years: 4 cores is the minimum then spend the rest on the GPU.

 

[poofyhairguy]

Since this is the CPUs subforum, let's not make it AMD vs NVIDIA.

Too bad, that is the most obvious observation. Hmmm....I guess it all comes down to phrasing. Let me put it this way:

If I had a AMD CPU I would much rather own a Nvidia GPU. AMD+AMD is a bad combo obviously.

 

[know of fence]

To be unfair. Taking the 3.2/3.6 GHz i5-4570 as as base line is setting the bar pretty low. Consequently, the lower you set the bar, the bigger the "gains" will be, especially comparing across architectures and doubly so across DDR speeds.

As far as hyper-threading is concerned, Overclocking the CPUs basically eases or removes the bottleneck, so it's not surprising that we see little difference there.
Also, PClab.pl do their own i5 to i7 comparisons over a 14 games and I come to conclude after picking out the numbers that even at 4.5 GHz the difference is still considerable 6% avg improvement.
FPS taken from the Overclocking part of their take on "Should you upgrade from Sandy B. to Skylake" article using a 980 ti.

Anyway, complements on the presentation, I like the fact that one can see the reference 1.00, something that is usually a little ambiguous with just straight percentages.

 

[lehtv]

Thanks for all the replies. :thumbsup:

There is still a lot of i7-920 class hardware out there and it should be plenty fast for games. The choice between upgrading an old i7 vs putting all available money into a video card is a no brainer. An i7-920 with a 980ti SC should outperform a skylake i7 + GTX970, and cost less.

In terms of average framerates, you're probably right. But I can't really subscribe to the idea that one setup outperforming another in average framerates automatically makes that the better setup. I upgraded from i7-920 to i7-3770K three or so years ago because the i7-920 was bottlenecking a few games situationally. I had upgraded my graphics card from 560 Ti to 7950, and while it was much faster in terms of average framerates, I found that the poorly threaded games I was playing at that particular time were still suffering from the same low situational framerates. Frustration with that lack of consistency and fluidity was all it took for me to spend 300+ euros on a new setup, and it was worth it. That same CPU and motherboard carried over to my next build with a GTX 970, and the newest Haswell CPUs (that I would've had to upgrade to had I not bought the 3770K before) weren't really any better than the 3770K.

To be unfair. Taking the 3.2/3.6 GHz i5-4570 as as base line is setting the bar pretty low.

By what argument is i5-4570 a low standard to compare to? Stock clocked Haswell i5 is probably one of the most common CPUs of gamers with a GTX 970 class graphics card. Not everyone's an overclocker. It's also demonstrably fast enough for a GTX 970 class card so if anything it's setting the bar high rather than low.

As far as hyper-threading is concerned, Overclocking the CPUs basically eases or removes the bottleneck, so it's not surprising that we see little difference there.

Yep. Overclocking can compensate for lack of HT, and vice versa. Although we'd also see bigger gains for HT with even faster GPU setups, e.g. 980 Ti SLI.

Anyway, complements on the presentation, I like the fact that one can see the reference 1.00, something that is usually a little ambiguous with just straight percentages.

Glad you like it

 

[escrow4]

Or just buy a 5820K and overclock it to 4GHz or above and you'll be set period. An i5 is borderline in 2016, an i7 is it and a real i7 is what you really want.

 

[coercitiv]

An i5 is borderline in 2016, an i7 is it and a real i7 is what you really want.

One always gets the i7 when budget is not the issue. If not for the increase in average FPS, then at least for smoothing out the min fps dips. However, when budget is limited, it is important to start looking at several build paths with similar costs.

Example:

• i7 + H chipset + base ram + high end gpu
• i5 + Z chipset /w overclock + fast ram + high end gpu

These two configurations can end up costing the same but will show different strengths and weaknesses when used in gaming. In both cases you maximize GPU power but are forced to make concessions with the rest of the system.

The only math I've done so far when it comes to K series CPUs is that under certain conditions (mb I/O requirements, deals on components etc.) it is worth buying the overclockable setup because it brings about a better than linear increase in performance/price. Again, this varies from case to case.

 

[CorsairDemon]

I've been recently building my friends Xeon X3440 Machines on intel boards. Toss a 130 graphics card and you got a great gaming machine.

A lot of the older hardware is so valid, it's almost a steal on the web now. Another example. Dell Precision T3500. 1366 X58 Systems for 150 on the bay. Running a media server with one (and Steam In-Home Streaming)

 

[lehtv]

https://www.youtube.com/watch?v=2ZxZiksWtRQ

The amazing DigitalFoundry compared the 2500k vs 3770k... [will update after I finish watching]

Will you though :sneaky:

 

[Blue_Max]

Will you though :sneaky:

Heh. I wish I had the definitive answer as to "best"... I'm furiously debating whether to buy a used i7-4770 based machine or spend a little more and get a shiny new i5-6600k combo. Overclocking vs. hyperthreading. The lack of definitive answer is driving me nuts.

 

[coercitiv]

Heh. I wish I had the definitive answer as to "best"... I'm furiously debating whether to buy a used i7-4770 based machine or spend a little more and get a shiny new i5-6600k combo. Overclocking vs. hyperthreading. The lack of definitive answer is driving me nuts.

There's no furious debate in your case: using a GTX 970 class GPU will have you GPU bound with either i7-4770 or i5 6600K oc. What's driving you nuts is you wish you could justify the new 6600K acquisition, yet can find no numbers to back it up.

Maybe you should focus on the new gen GPUs instead, you'll need to go that route as well if you truly want a big jump in performance.

 

[lehtv]

IMO the 6600K is a safer bet since you can pretty easily sell it and upgrade to 6700K or Kaby Lake i7 whenever it starts to look like there's substantial gains to be had from both HT and overclocking. With i7-4770, you're basically locking yourself out of the overclocking option, unless you also upgrade the motherboard

 

[tweakboy]

At the end you need a video card now to choose from.

 

[Leyawiin]

Too bad, that is the most obvious observation. Hmmm....I guess it all comes down to phrasing. Let me put it this way:

If I had a AMD CPU I would much rather own a Nvidia GPU. AMD+AMD is a bad combo obviously.

I've known that for a while and don't recommend an AMD GPU unless one has a fast Intel CPU, but its so glaringly obvious with the FX-9590 paired with the upper-midrange cards. Almost every game goes from sub-30 FPS slide show to actually playable with the GTX 970 (instead of the R9 390). AMD really needs to do something about their driver overhead.

 

[UaVaj]

for today gaming. i5 sandy bridge is all you need.

 

[lehtv]

I've known that for a while and don't recommend an AMD GPU unless one has a fast Intel CPU, but its so glaringly obvious with the FX-9590 paired with the upper-midrange cards. Almost every game goes from sub-30 FPS slide show to actually playable with the GTX 970 (instead of the R9 390). AMD really needs to do something about their driver overhead.

Somehow I doubt AMD is going to do jack about DX11 driver overheads, but will focus instead on getting even with NVIDIA in DX12