HyperThreading On or Off on your Raptor Lake CPU poll and discussion.

Betroz · May 12, 2024

viki_zaiko said:
People run 400+ Mhz on P cores with HT off, I do.

That is not normal, at least not with a i9 like the 14900K. Maybe a 13700K/14700K is different. Can you pass 10+ runs of CB15 Extreme with no WHEA error?

viki_zaiko · May 12, 2024

Betroz said:
That is not normal, at least not with a i9 like the 14900K. Maybe a 13700K/14700K is different. Can you pass 10+ runs of CB15 Extreme with no WHEA error?

Nah, i don't use such hammering tools, i only use for testing to see when i have the best single-core performance as it is what matters to me.
I passed every game i have on my PC for at least 2-3 hours, that's good enough for me.
I think I9 has more overheating problems, so it might be normal.

SebastianFM · May 12, 2024

deidian said:
Although your scenario is a no no: if you encode and run a game at the same time performance of both is going to drop, best thing you could do is manually resource limit the encoding process until you determine it doesn't affect the game.

It was only an example. For me it is possible to run a game only on P cores and something totally different on E cores.

deidian said:
Also said in the message than a 8p+16e CPU won't notice the difference generally because it already got plenty of physical cores.

You're right.

X909 · May 12, 2024

deidian said:
HT Off may allow you to run +100Mhz higher on CPU that already runs 5Ghz.

I can not confirm this. The max. clock in medium heavy or light workloads (Gaming) didn't change in my last HT off tests.
Hopefully I find the time to test this week a little more if a well tuned setup with Powerlimits and ICCmax can remove the Vcore penalty at full load for the HT on setup completely or to an amount that it doesn't matter anymore.

Carlyle2020 · May 12, 2024

Yeah, since i mostly run single core games i choose to run my 13900ks at 8x 6Ghz allcore without HT and e-cores.
220 watt was easy to cool even with an AIO.

I´m old but i still can feel it if i run only 5.5 instead of 6.0 Ghz just navigating windows.
Even the startup screen is different and shows different captions during the bios bootup.

My hands are past their esports qualification but my inernal refresh rate is still high

Cryptedvick · May 12, 2024

Neo_Morpheus said:
Nice testing Cryptedvick, even though it shows HT/off takes slightly longer to load, 1m7s compared to 1m14s, but that wouldn't matter.

That was my local time, it was ~ 1AM when I did my testing.

It was part of a bigger test on TLOU that included HAGS ON vs OFF (which showed that HAGS ON was better together with HT OFF). Game loads quite fast, its on RAID 0 NVME

fray_bentos · May 12, 2024

viki_zaiko said:
I use HT off, P-core offset +4, E-core offset +3, lite load on default 9, voltages on auto.
Max temp 82 degrees after 2 hours of gaming. I get the best single-core performance this way.

82 C... gaming... ?

fray_bentos · May 12, 2024

X909 said:
Only if you let the HT on system take additional power. When you set proper limits, that effect should not come into play.

I'm not sure about that. Let's say I set the power limit to the value it would hit without HT enabled. Next, I enable HT and hit said power limit - this will downclock the P cores. If I'm gaming that's bad. However, if running a multithreaded productivity task, having HT on might be slightly better than not.

The crux of it is that I do care if I skip frames in games, but I don't care if productivity tasks take a few more minutes/seconds to complete.

I also know that Spiderman Remastered runs a lot worse with HT enabled.

SebastianFM · May 12, 2024

fray_bentos said:
I'm not sure about that. Let's say I set the power limit to the value it would hit without HT enabled. Next, I enable HT and hit said power limit - this will downclock the P cores. If I'm gaming that's bad. However, if running a multithreaded productivity task, having HT on might be slightly better than not.

In such a case, the difference in performance between HT off and on is small, only a few percent. First of all, when P cores will downclock then E cores will downclock too. Secondly, when HT is off, Vcore voltage can be lowered even further.

i9-13900KF. I used default settings for this test, only set the power limit to 220 W.

HT off, Vcore offset -0.1 V.
Cinebench R23: 34352.
Cinebench 2024: 2020.

HT on, Vcore offset -0.05 V.
Cinebench R23: 37446.
Cinebench 2024: 2167.

HT off vs HT on.

Font Material property Rectangle Pattern Parallel

Font Rectangle Material property Screenshot Pattern

fray_bentos · May 13, 2024

SebastianFM said:
In such a case, the difference in performance between HT off and on is small, only a few percent. First of all, when P cores will downclock then E cores will downclock too. Secondly, when HT is off, Vcore voltage can be lowered even further.

i9-13900KF. I used default settings for this test, only set the power limit to 220 W.

HT off, Vcore offset -0.1 V.
Cinebench R23: 34352.
Cinebench 2024: 2020.

HT on, Vcore offset -0.05 V.
Cinebench R23: 37446.
Cinebench 2024: 2167.

HT off vs HT on.
View attachment 2659333 View attachment 2659334

Thanks for providing the data confirming what I suspected. Scenario with HT off is better for gaming, lower voltage and lower power and 54-55 P-core ratio (vs 50-51 with HT on) and higher e-core ratio at 43 (vs 40-41 with HT on).

deidian · May 13, 2024

Thta's Cinebench not games. Games are latency sensitive and Cinebench is raw throughput. Games are still going to run 5.4-5.5Ghz on P-cores(even 5.6Ghz if configured) and not come close to 220W power except on loading. The results from Cinebench don't extrapolate to games at all: they wouldn't even running 720p unlimited framerate(which is impractical measurement: performance for games is relevant at the framerate and settings everyone plays)

X909 · May 13, 2024

fray_bentos said:
I'm not sure about that. Let's say I set the power limit to the value it would hit without HT enabled. Next, I enable HT and hit said power limit - this will downclock the P cores. If I'm gaming that's bad.

Why should it hit power limit in gaming with HT? HT only consumes more power when it also boosts performance (if we assume same VCore).

persizi · May 13, 2024

Тhe conclusion is:
No need to buy a Arrow Lake, just turning off the HT on the Raptor Lake to convert it on Arrow Lake.

X909 · May 13, 2024

SebastianFM said:
In such a case, the difference in performance between HT off and on is small, only a few percent. First of all, when P cores will downclock then E cores will downclock too. Secondly, when HT is off, Vcore voltage can be lowered even further.

i9-13900KF. I used default settings for this test, only set the power limit to 220 W.

HT off, Vcore offset -0.1 V.
Cinebench R23: 34352.
Cinebench 2024: 2020.

HT on, Vcore offset -0.05 V.
Cinebench R23: 37446.
Cinebench 2024: 2167.

HT off vs HT on.
View attachment 2659333 View attachment 2659334

Does the HT on system really need 50mV more with the power limit in place?

X909 · May 13, 2024

I prepared my test setup yesterday. Found the lowest vcore HT off setting (6100P core / 4400 ecore, 1.292V, 325W PL, 380A iicmax) for CB23. It scored 37000.
Today I will test how much more VCore the same setup needs for HT on. The first run with same VCore ran through but dropped WHEAS (score was way higher than with HT off so maybe I will limit power a little more), so I think it will be 10 or 20mV more for the HT on system. If thats the case, I would expect all single and medium workloads to run with the same performance and same power, regardless of the HT setting. At full load, HT on will throttle more but perform better. Lets see

X909 · May 13, 2024

persizi said:
Тhe conclusion is:
No need to buy a Arrow Lake, just turning off the HT on the Raptor Lake to convert it on Arrow Lake.

Unfortunately not... because you get all the thread isolation, security fixes and stuff needed for a HT CPU also when turning HT off. If we could get a microcode without all that, so with hard disabled HT and max. single thread performance - than we are closer to Arrow Lake (but still by far not there

).

persizi · May 13, 2024

I've been sarcastic.

SebastianFM · May 13, 2024

So is there any game that has better performance with HT enabled on a CPU with 8, 12 or 16 E cores? No one has shown such results yet.

acoustic · May 13, 2024

SebastianFM said:
So is there any game that has better performance with HT enabled on a CPU with 8, 12 or 16 E cores? No one has shown such results yet.

Yzonker did this testing a while ago. I believe the best performance was 8P/8E, no HT. I can't remember for sure if he had HT on or not, though -- paging @yzonker

It's not really surprising to me, though. Games typically are built for 16 thread utilization (except some big outliers like BF with the Frostbite engine).

There was also talks that the Windows Scheduler was built for the 12900K (8P/8E), and was part of the reason why Windows responds better to that config. I've been testing the "Heterogenous Policy In Effect" setting again -- I had used Policy 0 for the longest time, but testing Policy 1. This is with 8P/16E, HT disabled.

yzonker · May 13, 2024

acoustic said:
Yzonker did this testing a while ago. I believe the best performance was 8P/8E, no HT. I can't remember for sure if he had HT on or not, though -- paging @yzonker

It's not really surprising to me, though. Games typically are built for 16 thread utilization (except some big outliers like BF with the Frostbite engine).

There was also talks that the Windows Scheduler was built for the 12900K (8P/8E), and was part of the reason why Windows responds better to that config. I've been testing the "Heterogenous Policy In Effect" setting again -- I had used Policy 0 for the longest time, but testing Policy 1. This is with 8P/16E, HT disabled.

Of the games I tested that was the best config. Although admittedly it didn't make a big difference. I think the newest games handle the high thread count CPUs pretty well.

@digitalfrost posted this a few days ago in this thread.

(32) HyperThreading On or Off on your Raptor Lake CPU poll and discussion. | Overclock.net

This post was also interesting. I found HT didn't help at all for the TLoU shader comp which is backed up by what this user found.

(32) HyperThreading On or Off on your Raptor Lake CPU poll and discussion. | Page 3 | Overclock.net

I need play with this again as I think despite TLoU shader comp not taking advantage of HT, the CPU still runs cooler during the compile. That's what I remember anyway, but would like to confirm. I was really only interested in the total compile time, so I'm not 100% sure.

HyperThreading On or Off on your Raptor Lake CPU poll and discussion.

Do you have HyperThreading On or Off?

HT On

HT Off

I have 2 different profiles in BIOS for on vs off depending on usage

Betroz

viki_zaiko

SebastianFM

X909

Carlyle2020

Cryptedvick

fray_bentos

fray_bentos

SebastianFM

fray_bentos

deidian

X909

persizi

X909

X909

X909

persizi

SebastianFM

acoustic

yzonker

Top Contributors this Month

Recommended Communities