Gaming benchmarks: Core i7 6700K hyperthreading test

[Melvis]

Interesting results and thank you for all your hard work.

Its the total opposite when it comes to AMD though, 4core vs 8core (depending on the game) you can see up to almost double FPS in games over a quad core. (Ive done the testing)

 

[ViperXTR]

As an owner of an i3 2100 long ago, HT helps for this, specially in heavy/CPU hungry games back then like GTA4 (and even more so useful in emulators), disabling it causes lower framerate and noticeable stutters/lags. For i7, it's a different story

 

[Mussels]

As an owner of an i3 2100 long ago, HT helps for this, specially in heavy/CPU hungry games back then like GTA4, disabling it causes lower framerate and noticeable stutters/lags. For i7, it's a different story

i have an i3 here in one of these systems (forgot the model, 2xx0 something) and i'd never turn HT off on it. I wonder if its a hardware difference, or just that software is badly optimised for 4+ threads

 

[Dethroy]

i have an i3 here in one of these systems (forgot the model, 2xx0 something) and i'd never turn HT off on it. I wonder if its a hardware difference, or just that software is badly optimised for 4+ threads

Sadly it's the latter.

 

[Artas1984]

bumping a slightly older thread, as i've been doing testing myself with my new CPU with HT on/off in gaming (takes 10C off load temps, for no gaming loss) and i googled it leading me to here...


My question is: is FRAPS the problem here. It's basically recording the 3D and playing it back, and the CPU load may be entirely different to rendering it initially. Have you tested with in-game benchmarks where available to see differences there?

Not sure if i'm wording my concern correctly, anyone else able to expand on the idea?

Fraps is not recording any 3D stuff. While benchmarking games, it only calculates the frames those are past-rendered, not pre-rendered. I've noticed that FPS does not change when fraps benchmark is on or of.

 

[FordGT90Concept]

None of these are Vulkan/Direct3D 12. If Hyperthreading helps in a game, it would be leveraging those APIs.

 

[flapbunn]

Gaming on my 5960X in games like FC4 is significantly better with HT turned off

 

[natr0n]

Yay for real cores.

 

[Kanan]

Gaming on my 5960X in games like FC4 is significantly better with HT turned off

Yes because 8 threads are easily enough for anything and 16 just too much. It's like divided power at times.

 

[cdawall]

I have no issues with mine?

 

[TheoneandonlyMrK]

So this adequately shows why AMD can't at this moment beat Intel in IPC terms, because due to the advantage in thread dispatch that micro opps and it's cache provide, Intel utilises most if not all of a cores resources regardless of Ht , thread dispatch is their main bottleneck imho it's deciding a cores performance and regardless of Ht on or off it stays the same.
Just my opinion.

 

[David Fallaha]

Interesting, wonder what a 5820k or 5960x would get. Have a feeling that it will be less of a impact.

If it were not for the fact that i am away from my main computer this weekend i would check it my self.

jst reading this...have you checked?

if anything you need hyperthreading much less -i got big esp emulation boosts turning it, much smoother / less glitched framerates

 

[David Fallaha]

I thought it was known that HT doesn't really benefit in games with exception of i3.

But HT is a huge factor to me because HT makes around at least 20% difference in rendering time.

-just wondering, if it's a single-socket Xeon you have (and so potentially overclockable) why not turn HT off and ramp your clockspeed (this i did to great effect)

 

[Viruzz]

Thanks for benchmark.
But the issue with HT is rather strange, I always kept it off myself, 5820K user here, 6 core is more than enough for everything.
But as far as I understand there is difference of OS, ie Win7 is not good for HT, 8.1 is a bit better and Win10 is best available.

I found this guy on YouTube doing what everyone asked, he took 5820K and emulated every possible configuration: 2C/2T, 2C/4T, 4C/4T, 4C/8T, 6C/12T (except 6/6), he tested 3-4 games and they all got 1-2 fps increase in 6C/12T situation.


5820k @ 1.195v @ 4.2ghz (3.3ghz Cache ratio at 1.055v).
Asrock Fata1ity X99
Crucial DDR4 (4x4GB Sticks @ 2400mhz - 14-15-14-40 @ 1.21v)
GTX 970 @ 1600mhz / 1640mhz @ + 0.49v
256GB MX100 SSD
630 Watt X't Enermax PSU
Win 8.1 Professional

Now this guy tested 5830K @ 4.5Ghz with 2 cores disabled to simulated standard i7, he tested 4/4 and 4/8 situation and based on his benchmarks lots of games get lower FPS with HT on

This is Windows 8 test so again, no windows 10 Benchmarks yet

Full system specs
Intel i7 5930K overclocked to 4.5Ghz
MSI Gaming 7 X99A
Nvidia GTX 980Ti Inno 3D Reference
Nvidia GTX 980Ti Zotac Referecne
Drivers 361.43
32GB of Corsair Vengeance 2666Mhz DDR4
OCZ Agility 3 120GB SSD + 2TB Seagate Barracuda Storage
Corsair Graphite 780T Black
Corsair HX 1000i PSU
Corsair H110i GT cooler
Samsung 850 EVO 500GB
Samsung SM951 M.2 128GB PCIe 3.0 SSD
OCZ Agility 128GB SSD
Toshiba 3TB HDD

 

[Caring1]

Now this guy tested 5830K @ 4.5Ghz with 2 cores disabled to simulated standard i7, he tested 4/4 and 4/8 situation and based on his benchmarks lots of games get lower FPS with HT on...

Watching the screen overlay it's clear with Hyperthreading on, the GPU has less load and a lower temperature. That to me shows the CPU is a bottleneck with no Hyperthreading enabled as the GPU works harder.

 

[Viruzz]

Watching the screen overlay it's clear with Hyperthreading on, the GPU has less load and a lower temperature. That to me shows the CPU is a bottleneck with no Hyperthreading enabled as the GPU works harder.

I just rewtahced the video and I cant say that I noticed it, maybe here and there but the opposite can be said too. What I did notice is 30% lower total CPU load 75% vs 50% although it might be fake because it calculates all 8 cores and they dont have so much load.

The optimal situation would of been a software that can Disable/Enable HT in windows, so we could disable it when we game or dont need to lower the temps.
In my case, I game in 4K, so I can keep it enabled it wont affect the FPS at all since its GPU bound even with 1080Ti (I benched my 5820K CPU/Cache overclocked to 4.5Ghz vs Default settings and had identical FPS in 4K, ZERO difference), so I bet enabling HT wont reduce any FPS too, but what I dont want is unnecessary temp increase, so software to enable disable HT would be ideal.

I did some research and there is a way to have HT on for windows and day to day tasks and have it disabled for Games and programs you dont need many cores for.
The Build in Windows permanent solution is to create a Shortcut using Affinity command, said command can also launch a program in any priority, so we can one time create shortcut for game to lunch it with X cores and X priority.

The Freeware software option called Prio it can save your Affinity and priority among other things.

The paid "Gaming" solution is the 15$ 'CPUCores' app sold on steam, not only that it can disable HT for any program added to its database but it can set priority and "optimize" windows for gaming, people with weaker systems report 10-20% improvements.
The app can "isolate" windows services and background programs to separate chosen one-two cores, temporally unload/disable non system essential services and background apps.
Im not paying for it. (Personally as Avast user I rather enjoy the free gaming mode that comes with Avast, its rather nice to pre-choose programs and games and set power mode to high or balanced during load, most "game optimizesrs" ignore this function but for me its much more effective than all that crap they do, because day to day I use Balanced Power mode that also limits my CPU to 3.3Ghz max to save power and avoid heating the room but when I game I switch to High performance mode that also locks the CPU to 4.5Ghz)

 

[EarthDog]

Why bother...really...

 

[trparky]

So here's a really dumb question. Say you have a CPU with Hyperthreading or SMT on AMD Ryzen and you lock the game to a specific number of cores using the Windows Task Manager (or Process Hacker), can you negate the loss of performance with SMT enabled?

I'm assuming that the loss of performance is due to the fact that SMT creates virtual cores out of the remaining processing capacity of a real core that's not 100% used. Or am I barking up the wrong tree here?

 

[Viruzz]

So here's a really dumb question. Say you have a CPU with Hyperthreading or SMT on AMD Ryzen and you lock the game to a specific number of cores using the Windows Task Manager (or Process Hacker), can you negate the loss of performance with SMT enabled?

I'm assuming that the loss of performance is due to the fact that SMT creates virtual cores out of the remaining processing capacity of a real core that's not 100% used. Or am I barking up the wrong tree here?

Yes, thats exactly how you Soft "Disable" HT. In windows task manager cores come first and treads come second, if you have 8C/16T than its 0-7 real and 8 to 15HT
If you lock your game to the actual cores, its like if HT was disabled.

In my previous post I found a program that can remember your settings, do them once for your games and programs and thats it, you can have HT enabled for 7zip and video encoding and play games on real cores without HT penalty.

 

[Frag_Maniac]

I have to think this is why Intel bumps stock clocks up on i7s. They know full well HT on their quad cores is not a sell at all for gamers.

 

[gupsterg]

Yes, thats exactly how you Soft "Disable" HT. In windows task manager cores come first and treads come second, if you have 8C/16T than its 0-7 real and 8 to 15HT
If you lock your game to the actual cores, its like if HT was disabled.

In my previous post I found a program that can remember your settings, do them once for your games and programs and thats it, you can have HT enabled for 7zip and video encoding and play games on real cores without HT penalty.

Dunno about Intel as not had a chip from them with HT, but on Ryzen the odd cores are SMT and even real, link.

 

[mouacyk]

Does anyone know of a benchmark that compares 6c/6t to 4c/8t in BF1? I'm interested in frametime variance (smoothness) around 120-144hz. I think the new Coffee Lake hexacores are going to be perfect for 144Hz(+) gaming, once you turn off HT since it is on a mean and lean platform unlike HEDT. This is why for quad-threaded games, the 7700K is still king - low cache/memory latency = more responsiveness.

HT generally benefits unoptimized/unoptimizable code where there are lots of branch mis-predictions or memory stalls (production workloads). Games in general, due to their real-time requirements, optimize away memory stalls and branching as much as possible already. With HT on, the thread scheduler may over-commit game code for execution (biting off more than one can chew), which results in execution stalls, which manifest as frequent frametime variance.

 

[FreedomEclipse]

If i disable HT, bf1 starts to stutter real bad

 

[mouacyk]

If i disable HT, bf1 starts to stutter real bad

Can you post the value you see for "Job Threads" when you type the following into the console:
Render.DrawScreenInfo 1