So I scrounged up the time/gumption/$$$ to upgrade my PMDG 737NGXu / GoFlight rig with a new Z690 AORUS mobo, 64GB of DDR4-3200 CL 16 RAM, a new Core i9-12900K along with the necessary Corsair LGA 1700 Retrofit Kit (for my H150i cooler) and then fired up P3Dv5.2HF1.  During my initial flights with high-density city/airport scenery (such as New York City/Airports X, Flight Beam's KDEN, FSDreamTeam's KDFW, LatinVFR's KMIA at highest settings -- don't try this without a card with more than 11GB of VRAM -- etc).   At some point I noticed that both Windows 10 21H1 (via the Performance tab in Task Manager) and Intel Xtreme Tuning were reporting a "Max Core Frequency" of well under 5.0 GHz.  Actually the speed reported was under 4.0 GHz, which is about default speed of the Efficiency "E" Cores at 3.9 GHz (not o/c'd).  Note that the CPU was *not* being maxed out (according to Task Manager), but frame rate could have been a bit better (even though I've pretty much maxed out all the sliders except for Volumetric Clouds with EA, which drives my RTX 3090 to 100% GPU utilization according to GPU-Z, using instead HiFi's ASP3D with the latest ASCA for use with P3Dv5's EA). 

For laughs and giggles I disabled all eight (8) Efficiency "E" Cores via the motherboard's BIOS and re-ran my stress testing with dense scenery and mostly Ultra graphical settings and I noticed 100% CPU with clock speed staying at/near 5.0 GHz (with the eight Performance "P" Cores overclocked to that rate) with minor thermal throttling (according to Intel Xtreme Tuning) -- perhaps I used a tad too much thermal paste.  Other than minor stutters loading scenery (I've got the entire U.S. in photoreal scenery from MegaSceneryEarth and storing them entirely on SSDs is not enough to eliminate stutters) I was getting framerates ranging from 20 to 40 FPS (in 4K resolution), depending on whether I was near an airport.

So my questions are this:  Was the low GHz speed (with the "E" cores enabled) I wrote about in the 1st paragraph above an anomaly in Windows 10 because it isn't fully aware the design of the 12th Gen CPU's "P" and "E" cores?  Does P3D need some sort of patch/update to ensure it runs its busiest single-core activities on a "P" core at the CPU's full speed (i.e. avoiding the lesser performing, lower-clocked "E" cores)?  If so (or even if not), will P3Dv5 run better on Windows 11, which is supposed to have some sort of threading optimizations designed for this new (ridiculous for gaming purposes IMHO) P/E core architecture?  What about Windows 10 21H2?  Or is it best to just disable the "E" cores and call it a day (at least for now).  On a side note I'm wondering how MSFS2020 is performing with these new Intel CPU's and their being optimized under Windows 11.  Win-tel to the rescue?  That, or just give me a 12th Gen Intel CPU with 10 or 12 "P" cores!  Sheesh.

-Tim

I was under the impression that Windows 10 doesn't have the best thread scheduler (yet) for Alder Lake and you should use Windows 11.

Tip: change Windows 10 power plan to Ultimate Performance and test again. Don't disable cores.

Edited November 20, 20214 yr by Emerson67

10 hours ago, Emerson67 said:

Tip: change Windows 10 power plan to Ultimate Performance and test again. Don't disable cores.

Um, there are only three default (built-in) power plans included in Windows 10:  Power saver; Balanced; and High performance.  Did you create your own power plan?  (Perhaps you purchased your computer from a manufacturer that created a custom power plan and titled it "Ultimate Performance"?)  If so, what would be the relevant settings (in said power plan) to adjust?  Already in the default "High performance" plan the setting for "Minimum processor state" is at 100%...

Edited November 21, 20214 yr by tjahns

10 hours ago, Greggy_D said:

I was under the impression that Windows 10 doesn't have the best thread scheduler (yet) for Alder Lake and you should use Windows 11.

Yeah, that thought crossed my mind (of course).  Can anyone share their experience running P3D (and popular add-ons) under Windows 11?  Maybe I'll be brave and take the plunge.  Nothing to lose (but a whole lot of time) on the "bleeding edge"...

See https://www.avsim.com/forums/forum/905-win11-os-forum/ for Win11 experiences.

Well, I re-enabled the Efficiency "E" cores on my Alder Lake and also performed a Windows 10 Feature Update to 21H2 (a.k.a. November 2021 update), which apparently contains threading improvement with Alder Lake in mind -- there's an Intel VGA driver that reportedly makes use of it).  After these two mods, Intel's Extreme Tuning Utility now does not show the GHz drop it did when I was using Win 10-21H1 (with P3Dv5) with all 16 cores enabled.  Also, Windows Task Manager did not report the CPU dropping below 4.0 GHz (rather displaying apparently the average of all 16 cores, i.e. ~4.5 GHz or greater).

By the looks of the core temperatures, it appears P3Dv5 is keeping all Performance "P" cores fully engaged at 5.0 GHz (o/c'd) but with significant thermal throttling as temps neared 100C most of the time.  It is also apparent that Windows 10's threading spreads around P3Dv5's busiest core among the cores on the CPU.  Whether it is rotating them to the "E" cores is not something I can determine -- on occasion one of the E core's temps would spike, so perhaps that is indeed what's going on -- but my framerate did fluctuate on a consistent basis (which is not unusual).

Perhaps Windows 11 is the cure to keeping P3D's busiest thread/core off an "E" core?  P3D seems to like to use *all* cores, though, so not sure how Windows 11 "thread director" would determine the best way to run P3D.  Perhaps LM needs to release a HF (or new point release) that can somehow tag its busiest thread(s) as not to be executed on an "E" core?  Although this new Alder Lake hybrid "P/E" core business is good for streamers, it may not be optimal for flight simulators.   

However, theoretically if all my flightsim add-ons could be delegated to "E" cores, reserving the simulator itself to using "P" cores (and what ever left-over processing power remains of the "E" cores) then Alder Lake could be highly desirable.  Personally, I'd rather have a homogenous CPU with 10 or 12 (or more) "P" cores and optimize it with Process Lasso.  Btw, does anyone know whether Process Lasso is optimized yet for Windows 11 and/or this new Intel hybrid Alder Lake CPU architecture?

38 minutes ago, tjahns said:

Btw, does anyone know whether Process Lasso is optimized yet for Windows 11 and/or this new Intel hybrid Alder Lake CPU architecture?

https://bitsum.com/product-update/process-lasso-10-4-cpu-sets-and-adler-lake/

On 11/20/2021 at 10:41 PM, tjahns said:

perhaps I used a tad too much thermal paste. 

 

Extremely unlikely. Experiments by gamers nexus and others have demonstrated that the amount of paste barely makes a difference. 

 

 

 

1 hour ago, martin-w said:

 

Extremely unlikely. Experiments by gamers nexus and others have demonstrated that the amount of paste barely makes a difference. 

 

 

 

Yeah, it turned out 5.0GHz (on all Performance "P" cores) was too much for either my cooling solution (a Corsair H150i liquid cooler with only 1,600 RBG fans -- that reached over 38 degrees Celsius cooling temperature -- at 100% CPU utilization) or perhaps the particular CPU I ended up with just doesn't cut it at 5.0 GHz.  Anyway, dropping the max frequency down to 4.8 GHz (when more than 4 cores are active) lowered temps drastically (like 20 degrees C lower), without much loss in framerate.  (The difference between 4.8 GHz and 5.0 GHz is only about 4%, which is largely insignificant anyway.)

Btw, I have not yet tried Process Lasso.  My feeling is that using it to limit P3Dv5 to just the "P" cores would slow the sim down down with my photoreal scenery and weather software.  The sim apparently puts to good use all 16 cores (including the lower-clocked "E" cores) during flight.  Perhaps limiting my numerous add-ons (not to mention monitoring software and Edge browser windows) to just the "E" cores would be a worthwhile endeavor (in case Windows 10-21H2 isn't doing it automatically).

I still think it'd be nice if LM would add some code (if the Windows 10 SKD provides for it) to limit its "Core 0" activity (which actually seems to get rotated out among all available cores via Windows 10 threading logic) to just the "P" cores, which of course have better single-thread performance than the "E" cores.  I'm not sure even Windows 11's thread director logic can pull this off automatically (via Win11's own threading logic) when dealing with P3D's appetite to use all available cores during flight...

 

 

On 11/21/2021 at 6:48 AM, tjahns said:

Um, there are only three default (built-in) power plans included in Windows 10:  Power saver; Balanced; and High performance.  Did you create your own power plan?  (Perhaps you purchased your computer from a manufacturer that created a custom power plan and titled it "Ultimate Performance"?)  If so, what would be the relevant settings (in said power plan) to adjust?  Already in the default "High performance" plan the setting for "Minimum processor state" is at 100%...

https://www.howtogeek.com/368781/how-to-enable-ultimate-performance-power-plan-in-windows-10/

To enable the Ultimate Performance plan on a computer not running Windows 10 Professional for Workstations you've got to jump through some hoops (no big deal):

Add or Remove Ultimate Performance Power Plan in Windows 10 | Tutorials (tenforums.com)

It's mostly about transitioning the computer from sleep to wake states, so increasing performance of a computer that is already fully awake (e.g. while the user is gaming) is not something the plan is intended to do, as is stated in the article you cited, Emerson.  Thanks for the tip, but mostly I'd say it's a dead end in terms of improving Windows 10 threading capabilities with Intel's new Alder Lake CPUs.

Personally, I'm thinking that Lockheed-Martin needs to do something to enable Windows 11 (and/or Process Lasso) to distinguish its "Core 0" (i.e. single-threaded) code from its other code that does things like load scenery in parallel.  That way a PC could be configured to ensure P3D's busiest "Core 0" thread runs on the "P" cores while all the other threads are free to run on both "P" and "E" cores.  

There's an interesting discussion on Windows 11's thread directing capabilities with the new Alder Lake CPUs at a popular hardware site I follow:

Thread Director: Windows 11 Does It Best - Intel 12th Gen Core Alder Lake for Desktops: Top SKUs Only, Coming November 4th (anandtech.com)

Given what I'm reading, Process Lasso would definitely be a wise investment (particularly with Windows 10).  It'd be an even better investment if P3D's busiest "Core 0" thread could be "lassoed" to the "P" cores, but the rest of P3D's threads still able run on the E cores, in addition to the P cores.

 

 

Edited November 23, 20214 yr by tjahns

Well , "CPU0 " is on the first core, therefore a P-core. So without affinity mask , the main thread will always be on the first Core. 

1 hour ago, tjahns said:

Personally, I'm thinking that Lockheed-Martin needs to do something to enable Windows 11 (and/or Process Lasso) to distinguish its "Core 0" (i.e. single-threaded) code from its other code that does things like load scenery in parallel

There's a technical guide on how to do it : https://www.intel.com/content/www/us/en/developer/articles/guide/alder-lake-developer-guide.html

13 hours ago, vincentrouleau said:

Well , "CPU0 " is on the first core, therefore a P-core. So without affinity mask , the main thread will always be on the first Core. 

I am merely using "Core 0" as a reference to how P3D (and FSX before it) always ran on the 1st core of a CPU under earlier versions of Windows.  Under Windows 10, executing threads are rotated among all available cores on a CPU.  (You can verify this by examining Core temperatures while sitting on a ramp in P3D -- Windows Task Manager will show 100% on "Core 0", but in actuality the hottest core rotates among all available cores.)  Hence my concern over P3D executing it's busiest single-thread  activity (which is what I'm for simplicity sake's referring to as the "Core 0" thread -- perhaps I shouldn't refer to it that way) on E cores.  You may wish to examine the rest of this "thread" with that in mind.