Hyperthreading in 2023

[elmariachi]

Given the last few days, decided to test my CPU with Hyper-Threading disabled just to see the difference in GAMING as many with CPUs with high core count mentioned you can easily push another 100-200Mhz more because you lose alot of heat disabling Hyper-Threading.

My original CPU is a 13900KS which is overclocked to 5.7/4.5P/5.0R(HT ON) which I have been using for a while. I have 2 games which I predominantly play everyday which is Cyberpunk 2077 and MW2022. My CPU is also NOT delidded and is watercooled. GPU is also running same clocks on both tests (2,925Mhz Core and 12,000Mhz VRAM)

On my original CPU with 5.7/4.5P/5.0R(HT ON), this was my Cyberpunk 2077 Benchmark.

On my original CPU with 5.9/4.5P/5.0R(HT OFF), this was my Cyberpunk 2077 Benchmark.

While I may have only gained an average of 5FPS with HT off with higher clocks at 5.9Ghz, my minimum FPS jumped 20FPS more than with a lower clocks with HT ON. While average FPS and max FPS are around the same, minimum FPS is a big jump either because it is being compensated by higher clocks or with HT OFF or collectively both. I can run 5.9Ghz all core with HT OFF with around the same temps as my 5.7Ghz all core with HT ON.

I actually think CPUs today with more than enough cores do not rely much on HT for gaming. I understand older titles do make use of HT and some titles may perform worse with HT OFF. But so far with AAA titles at least with the ones I have tested, seems to give a good boost in minimum FPS. I also find frametimes also improved noticeably. I can sit between 2.9ms and 4.5ms of frametimes in games like MW2022 with Gsync and Vsync off.

Intel is also doing away with Hyper-Threading on 15th Gen (Meteor Lake) with rentable units instead. My guess is that with enough cores on CPUs today, Hyper-Threading seems to be less relevant for workloads. Would be interesting to see how that translates into performance overall given Intel is also going to implement 3D cache on their newer CPUs also to compete with AMD X3D Technology.
Only question, is to what extent will the new CPUs run cooler with better efficiency?

https://www.hardwaretimes.com/intel...able-units-why-hyper-threading-is-going-away/

https://www.pcgamer.com/stacked-3d-cache-is-coming-to-intel-cpus-and-gamers-should-be-excited/

https://www.extremetech.com/computing/intels-meteor-lake-cpus-to-offer-on-package-lpddr5x-memory

 

[fkchon]

Yep I tried the same thing with 7950X3D. 8 vcache cores SMT off was the fastest. All 16 cores and SMT on was the slowest.

Most games just aren't made to take advantage of the extra threads. So turning them off reduces latency jumping from thread to thread and also uses less power/thermal headroom for then you can get extra frequency boost.

7950X3D + RTX4090

7800X3D + RTX4090

Also just noticed 13900KS minimum fps is much higher than AMD CPUs.

 

[watzup_ken]

I don't really think games nowadays can fully utilize all the threads and this is good testing to validate this.

 

[Yongkit]

Yep I tried the same thing with 7950X3D. 8 vcache cores SMT off was the fastest. All 16 cores and SMT on was the slowest.

Most games just aren't made to take advantage of the extra threads. So turning them off reduces latency jumping from thread to thread and also uses less power/thermal headroom for then you can get extra frequency boost.

7950X3D + RTX4090

7800X3D + RTX4090

Also just noticed 13900KS minimum fps is much higher than AMD CPUs.

It seems like you are running windows borderless, perhaps try full screen to double check the performance again?

 

[elmariachi]

Yep I tried the same thing with 7950X3D. 8 vcache cores SMT off was the fastest. All 16 cores and SMT on was the slowest.

Most games just aren't made to take advantage of the extra threads. So turning them off reduces latency jumping from thread to thread and also uses less power/thermal headroom for then you can get extra frequency boost.

7950X3D + RTX4090

7800X3D + RTX4090

Also just noticed 13900KS minimum fps is much higher than AMD CPUs.

I think the 13900K has the advantage of being able to run higher speed memory more effectively with the architecture. Some games show the gains of memory speeds and scales with it. Like I'm only running 7800C34 pretty tight with TRFC and TREFI.

But it's interesting to note even AMD with SMT turned off yields same benefits.

 

[elmariachi]

I don't really think games nowadays can fully utilize all the threads and this is good testing to validate this.

Yeah probably why even why hyperthreading is going away starting from 15th Gen. But this will also mean a massive change in how windows scheduler works as well.

 

[Encrypted11]

Does this mean CPUs stop being built around running Cinebench and E-Core count will stop doubling?

 

[hyperfuse]

Yep I tried the same thing with 7950X3D. 8 vcache cores SMT off was the fastest. All 16 cores and SMT on was the slowest.

Most games just aren't made to take advantage of the extra threads. So turning them off reduces latency jumping from thread to thread and also uses less power/thermal headroom for then you can get extra frequency boost.

7950X3D + RTX4090

7800X3D + RTX4090

Also just noticed 13900KS minimum fps is much higher than AMD CPUs.

i am using the 7800x3d.

so does it mean i should also turn off Hyper threading? the only game i play is CS2

 

[ragnarok95]

i am using the 7800x3d.

so does it mean i should also turn off Hyper threading? the only game i play is CS2

You should test with HT on and off yourself to get your answer.

 

[robolee]

i am using the 7800x3d.

so does it mean i should also turn off Hyper threading? the only game i play is CS2

if all you do is game, it's safe to say off smt. smt is very good at application stuff other than gaming

 

[chiu2000]

I don't really think games nowadays can fully utilize all the threads and this is good testing to validate this.

Game Devs say No

 

[elmariachi]

You should test with HT on and off yourself to get your answer.

Yeah you gotta test it. At least for Intel, disabling hyperthreading gets me +200mhz more for clock frequency than with HT on so it makes an added difference. Some game titles like it some don't. Enabling or Disabling really depends on use case scenario.

 

[elmariachi]

Game Devs say No

My e-cores are practically shaking legs most of the time even in game. Disabling HT I do see some added load on e-cores but by and large, the P cores does all the heavy lifting. Game Devs may say no but windows scheduler has the final say on what cores get used or not. Really depends on game title and your setup architecture.

 

[fkchon]

I think the 13900K has the advantage of being able to run higher speed memory more effectively with the architecture. Some games show the gains of memory speeds and scales with it. Like I'm only running 7800C34 pretty tight with TRFC and TREFI.

But it's interesting to note even AMD with SMT turned off yields same benefits.

The 7800X3D was running 8000C36 and 7950X3D was running 6000C30.

It seems like you are running windows borderless, perhaps try full screen to double check the performance again?

Yeah I do that with every game I don't think that will change the numbers that much.

 

[elmariachi]

The 7800X3D was running 8000C36 and 7950X3D was running 6000C30.

Yeah I do that with every game I don't think that will change the numbers that much.

It boils down to architecture. Intel running higher speed ram and AMD running higher speed ram scales very differently. But in short, I think disabling smt or hyperthreading seems to improve quite a fair bit for most AAA titles on both AMD and Intel.

 

[Yongkit]

Yeah I do that with every game I don't think that will change the numbers that much.

i guess zen 4 and 4000 series behalf differently.

below is my example

Full windows



Borderless windows

 

[matique]

On intel, HT off has almost always been the way to set up your cpu for gaming.

 

[Koenig168]

i guess zen 4 and 4000 series behalf differently.

below is my example

Full windows

View attachment 42731

Borderless windows

View attachment 42732

The minimum FPS differs markedly. Rather strange since I don't think it should make that much difference.

 

[Yongkit]

The minimum FPS differs markedly. Rather strange since I don't think it should make that much difference.

It sometimes depends on games, for mw2 if I use borderless my main fps will lost average 20fps

 

[Rogue]

FYI, E-cores pwned a lot of people due to UE 5 since it's DX12 only. Crash galore.

With solution being downing the CPU multiplier and/or disabling or de-prioritizing the e-cores via utilities like Process Lasso. lol.