So how much does hyperthreading matter for gaming?

[SteveGrabowski]

---

I have been wondering how useful hyperthreading really is for gaming, and whether getting that i7/Xeon instead of an i5 makes much sense. So I turned to GameGPU, which is one of the few sites that benchmarks CPUs for each game. They always do their benchmarks at stock speeds, and their CPU benchmarks are always done with either extreme overkill GPUs (GTX 690, GTX 780Ti SLI, GTX 980 SLI) at 1080p or high end GPUs at 1050p or 720p to isolate the CPU bottleneck. For fourth generation chips they compare the i7-4770k and i5-4670k which differ by 100MHz stock clock speed, and for second generation chips they compare the i7-2660k and the i5-2500k which also differ by 100MHz clock speed. So with that said, here are the differences in a few games, computed as the gain in percentage of average framerate one gets going to the 100MHz faster i7 vs the i5 of the generation: 

 

Note all numbers are rounded to the nearest percent, negative numbers mean the i5 did better than the i7, and I try to use tests done with Nvidia GPUs instead of AMD ones when both are offered. On average the 100MHz clock bump would give a 3% gain for the i7 vs the comparable i5 if one can assume linear speedup (which is a hell of an assumption and kind of a best case, but a simple one we'll approximate as true).

 

 

 

The choice of games was just somewhat recent stuff I'm interested in playing. Anything less than a 3% gain should probably be interpreted as hyperthreading slowing a game down. So for the most part I think these results tend to agree with the gamer religion that an i7 is most of the time a waste of money over an i5, especially since the vast majority of us will run at higher resolutions with lower end graphics cards than GTX 980s and thus be GPU bound. Nevertheless, BF4 and DAI are some pretty huge outliers and are both pretty recent, so they may be pointing to a future where those virtual cores provided by Intel's hyperthreading really start to matter

 

For me, I guess I went the safe route and paid $50 more to go with a 4C/8T Xeon rather than a similarly clocked 4C/4T i5. What made this decision easier for me was there isn't a card $50 above the GTX 970, so going over for the Xeon didn't alter my GPU selection at all. It's a little disappointing that the HT does so little most of the time though.

[ZetZet]

4 cores HT - 8 threads does not improve the performance much.

 

HOWEVER, i3 with two cores and HT is great.

[Sauron]

It doesn't make any difference if we're talking i5s vs i7s, it makes a significant difference in i3s vs pentiums.

[samcool55]

4 cores HT - 8 threads does not improve the performance much.

 

HOWEVER, i3 with two cores and HT is great.

True, And even then, some games are so badly optimized they will run better on 2 fast cores compared to 4 slow cores. So speed does matter in some cases more than amount of cores.

[Godlygamer23]

Depends on the CPU, number of threads available in the game, and the optimization levels.

[C0LL0SS0S]

more than 4 Cores don't make sense game wise

[SteveGrabowski]

You guys don't read anything into the huge performance deltas for BF4 and DAI though? It'll be interesting to see if these are just outliers or the beginning of a trend for 2015.

[Megahurt]

New games make more use of more threads but Directx12 should reduce the load on the cpu, so I don't know if that trend will continue.

[zappian]

The hyperthreading on the i3 is used on games.

 

If someone told you it isnt they are wrong.

Thats why the i3 keeps up with pentium g3258@4.6+GHZ and in some games like far cry 4 beats it.

 

Otherwise the pentium g3258 would crap on it in every game and it just isnt the case.

[SteveGrabowski]

New games make more use of more threads but Directx12 should reduce the load on the cpu, so I don't know if that trend will continue.

 

Windows 10 still won't be out until the summer though. I wonder what kind of time delay we'd be looking at before games start being written for DX12. What was it like for DX11, released in 09? I don't know since I was a console gamer in those days.

[Megahurt]

Yes it will be a while until we get directx12, but developpers will get it before we do and might choose to focus their efforts on it instead of optimising for more threads.

[XiaoPineApple]

Yes it will be a while until we get directx12, but developpers will get it before we do and might choose to focus their efforts on it instead of optimising for more threads.

few more years?

till intel/amd start selling 4core+ only CPU

game developpers then start writing for dX12.

dualcore/ i3+HT is still average gamer using.

[kachel94]

May i ask you too include ArmA 2 Operation Arrowhead? 

You will see that the game cannot get enough cores and loves over clocked CPUs.

[Megahurt]

few more years?

till intel/amd start selling 4core+ only CPU

game developpers then start writing for dX12.

dualcore/ i3+HT is still average gamer using.

No clue what you're trying to say, DX12 should put less burden on the cpu, making more threads/more powerfull cpus less important. I also expect some game to make use of it as soon as windows 10 comes out, and other games to release patches to support it, so probably less than 1 year from now.

[RockiLocky]

No clue what you're trying to say, DX12 should put less burden on the cpu, making more threads/more powerfull cpus less important. I also expect some game to make use of it as soon as windows 10 comes out, and other games to release patches to support it, so probably less than 1 year from now.

DX12 does not take away the burden of the CPU. What it really does is just give evry core thr same mount of burdennto even it out instead of having all the load on the #1 Core as we see now.

[SteveGrabowski]

May i ask you too include ArmA 2 Operation Arrowhead? 

You will see that the game cannot get enough cores and loves over clocked CPUs.

 

Crap, I don't see a CPU test for that game from them. Only GPU.

 

http://gamegpu.ru/action-/-fps-/-tps/arma-2-operation-arrowhead-test-gpu.html

[Megahurt]

DX12 does not take away the burden of the CPU. What it really does is just give evry core thr same mount of burdennto even it out instead of having all the load on the #1 Core as we see now.

The way I understand it is that DX12 will allow software to speak direcly to the gpu without having to go through the cpu, and every dx12 demo show significantly lower cpu % usage while gpu usage remain roughly the same.

[RockiLocky]

The way I understand it is that DX12 will allow software to speak direcly to the gpu without having to go through the cpu, and every dx12 demo show significantly lower cpu % usage while gpu usage remain roughly the same.

Yes but if you look exactly you see that the load gets even out on all the cores thats why you fo not see alot of load in total. Take it like that for example if you got a cake and you need to eat it alone you take along time, but if you are 4 people in total you guys are 4times so fast. Basicly what I wanna say is that the communication over the CPU gets faster on DX12 cause it evens out the load on every core. Look st the demo

[Megahurt]

Yes but if you look exactly you see that the load gets even out on all the cores thats why you fo not see alot of load in total. Take it like that for example if you got a cake and you need to eat it alone you take along time, but if you are 4 people in total you guys are 4times so fast. Basicly what I wanna say is that the communication over the CPU gets faster on DX12 cause it evens out the load on every core. Look st the demo

That doesn't even make sense, if you have 1 core running at 80% or 4 running at 20%, you'll get the same overall cpu usage. If the multiple cores complete the same task in less time, you'll end up with a higher cpu usage.

[arniks8]

That doesn't even make sense, if you have 1 core running at 80% or 4 running at 20%, you'll get the same overall cpu usage. If the multiple cores complete the same task in less time, you'll end up with a higher cpu usage.

If you balance workload it can be done more efficiently.( maybe I'm wrong but that's what I think)

[Godlygamer23]

That doesn't even make sense, if you have 1 core running at 80% or 4 running at 20%, you'll get the same overall cpu usage. If the multiple cores complete the same task in less time, you'll end up with a higher cpu usage.

Depends on the number of cores present and the CPU architecture.

[Megahurt]

If you balance workload it can be done more efficiently.( maybe I'm wrong but that's what I think)

More power efficient I would agree, more cores working lightly would use less power than 1 at full blast, but it would still represent the same % of your max computing power.

[OCing Noob]

may be 2 or 3 % if the game is very demanding . otherwise useless. games even don't use full 4 cores of i5 no a days

[BLCT]

I'm pretty sure you don't even use Hyper-threading in games, you'll only end up using the on heavy workloads.

[SteveGrabowski]

I'm pretty sure you don't even use Hyper-threading in games, you'll only end up using the on heavy workloads.

Did you read the results in the original post? Clearly Battlefield 4 and Dragon Age Inquisitions make great use of hyper threading, unless you want to tell me a 3% bump in clock speed should give 20% better frame rates.