13900K - Hyperthreading on or off?

[Seth2021 291]

when you have a Hungry Beast like 13900k, there is no reason to tweak or turn off any core. I remember when i had 8700k and i was always on tweaking mode ( turning HT off/on, delid with liquid metal for better temps etc) .

New CPU technology in my opinion don't need any extra tweaking. 13900k is already running at 5.5ghz in all cores with out extra overclocking  and combined with 4090 you have the best combo for MSFS. 

 

[odourboy 1,290]

So I did a test. Just one, so I really need to ensure this is repeatable. I created a scenario where I was main thread limited then looked at performance with and without hyperthreading. To my surprise, the hyperthreading enabled yielded better FPS performance by about 8%. I also saw something more unexpected... the W11 scheduler had half the P-cores parked with HT enabled. With it off, it seemed to shift tasks running on E-cores to P-cores, and all the P-cores were being used.

[Bob Scott 7,745]

I've been doing a lot of testing with varying core configurations on a 13900KS in the last week here.  Turning off HT and some or all of the E-cores reduces CPU package power, as I'd expect, and lowers temps.  In conjunction with Intel's TVB (thermal velocity boost) it means I was able to sustain higher temp-modulated dynamic overclocks, because with nonessential circuits disabled, the temps didn't rise enough to drive TVB to reduce the clocks.  With good water cooling, HT off, several blocks of e-cores disabled and +2 TVB acceleration, I've been able to run with two P-cores at 6.0 GHz and the remainder of the P-cores at 5.8 GHz...if you're CPU-limited an extra 9% on the core running the main thread can make a difference.

If you are running an application and you see e-core utilization at low levels, then your usage of HT is going to be even less, as the HT virtual processors are the last on the OS priority list.  In other words, you don't really need it, and any HT-assigned threads are reducing the respective P-core's throughput and splitting the available cache memory.  Turning HT off in the BIOS costs you nothing more than 30 seconds in the config menu and a reboot, and it's just as easily reversed if you find or suspect an adverse effect.

Does that make an appreciable difference with a particular application?  Only experimentation can answer that, as there are far too many variables in software, systems, and operating environments for a one-size-fits-all answer.

The 13900K/KF/KS has the ability to easily challenge pretty much any cooling solution short of LN2.  More than any CPU I can remember, a system config needs tailored to the operating environment (system/software) to get the most performance out of the system without hitting the thermal wall.  You can certainly get away with doing none of that and have a good experience, but a little work can pull more performance out of these expensive components.

And maybe don't knock it if you haven't tried it.  When someone running a 5-generations old 6-core CPU declares definitively that hitting thermal limits with a 24-core 13th-gen CPU simply means your cooling solution sucks, it's hard not to bust out laughing--the 13900K 's primary resemblance to an 8700K is that they both have a "K" in their nomenclature.  And when folks declare that manipulations of hardware configurations are a thing of the past, I'd suggest that's because modern hardware is now powerful enough to perform satisfactorily without touching the config, yet while still leaving a fair amount of potential performance on the table.  If a stock configuration is "good enough" for you, that's fine.  I get it--lots of people don't want to mess with combing that extra 5-10% or more out of the system.  But some of us do.

[Bob Scott 7,745]

32 minutes ago, odourboy said:

So I did a test. Just one, so I really need to ensure this is repeatable. I created a scenario where I was main thread limited then looked at performance with and without hyperthreading. To my surprise, the hyperthreading enabled yielded better FPS performance by about 8%. I also saw something more unexpected... the W11 scheduler had half the P-cores parked with HT enabled. With it off, it seemed to shift tasks running on E-cores to P-cores, and all the P-cores were being used.

With four of eight P-cores parked, you saw an 8% increase in FPS?  Something definitely does not compute there.

BTW, I use a utility called ParkControl (made by the same folks that produce Process Lasso) that prevents my P-cores from parking.

[odourboy 1,290]

6 minutes ago, Bob Scott said:

With four of eight P-cores parked, you saw an 8% increase in FPS?  Something definitely does not compute there.

BTW, I use a utility called ParkControl (made by the same folks that produce Process Lasso) that prevents my P-cores from parking.

I know eh!? It makes no sense to me either. But that's what I saw. Like I said, I need to repeat this a few times!

[HiFlyer 9,372]

Believe me, I've chased FPS with the best of them over the years, with all of the dollar costs associated with that pursuit.

Yet once you hit 100 fps or more in a heavy airliner, with everything on ultra over New York City....

Well, just a few months ago that was a fevered fantasy.

Having reached this point then, for myself, I kind of think that at least for now, I'm a dog that actually caught the car. (And didn't get kilt)

HT on or off then becomes kind of a moot point.....

YMMV

[odourboy 1,290]

Just now, HiFlyer said:

Believe me, I've chased FPS with the best of them over the years, with all of the dollar costs associated with that pursuit.

Yet once you hit 100 fps or more in a heavy airliner, with everything on ultra over New York City....

Well, just a few months ago that was a fevered fantasy.

Having reached this point then, for myself, I kind of think that at least for now, I'm a dog that actually caught the car. (And didn't get kilt)

HT on or off then becomes kind of a moot point.....

YMMV

I'm not actually chasing FPS. I want to get the most efficient configuration without performance compromise. Less power/heat... quiet fans!

[odourboy 1,290]

Follow-up note. I repeated the test a few more times... I've decided there's no appreciable performance or power consumption difference between HT on or off in my testing scenario.  I am consistently seeing a difference in the way the W11 scheduler uses cores in the two scenarios: 4 P-cores parked, 4 E-cores used with HT On, versus 8 P-cores used with HT off, but this doesn't translate into any observable FPS or Power consumption change.

My test was essentially static and not comprehensive. My gut tells me that when the going gets tough, having all P-cores in operation is going to help, so I'll leave HT off for now.

If any others do any testing, please share with the community! 

[Bob Scott 7,745]

35 minutes ago, odourboy said:

Follow-up note. I repeated the test a few more times... I've decided there's no appreciable performance or power consumption difference between HT on or off in my testing scenario.  I am consistently seeing a difference in the way the W11 scheduler uses cores in the two scenarios: 4 P-cores parked, 4 E-cores used with HT On, versus 8 P-cores used with HT off, but this doesn't translate into any observable FPS or Power consumption change.

What's reporting that the cores are parked?

[odourboy 1,290]

24 minutes ago, Bob Scott said:

What's reporting that the cores are parked?

Windows Resource Monitor 

Edited February 14, 2023 by odourboy