MSFS & Hyperthreading

[MrFuzzy 2,124]

14 hours ago, MarcG said:

I was intrigued...



🤔will have to see what it's like in a normal flight as that's different from a standard benchmark.
(VR, Reverb G2)

In VR you are most probably GPU llimited 99% of the time, try 2D with an airliner at a busy airport.

[Glenn Fitzpatrick 3,799]

9 hours ago, bobcat999 said:

I really wonder, imagine if you had only two cores but both had hyperthreading,

The minimum spec FS2020 CPU is 4 core (non hyper thread) so yes 4 threads but 4 real cores. I think two cores will struggle no matter what.

[MarcG 3,031]

47 minutes ago, MrFuzzy said:

In VR you are most probably GPU llimited 99% of the time, try 2D with an airliner at a busy airport.

I never fly 2D so doubt I'll ever do that (or fly airliners for that matter!).

[Donka 654]

2 hours ago, MarcG said:

I never fly 2D so doubt I'll ever do that (or fly airliners for that matter!).

I'm in the same boat. Got SMT disabled yesterday for some testing and absolutely zero difference in fps or even temps so leaving it enabled. Seeing a small but noticeable difference in fps and temps by undervolting a little though.

[Luis Hernandez 587]

Has anyone tried comparing behaviors while running both MSFS and FSHud (another known CPU hog)? I had to turn SMT back on for accomodating both P3Dv5 and FSHud (with lots of AI); but I haven't flown MSFS too much recently.

23 hours ago, Bob Scott said:

I used Process Lasso to externally set an affinity mask that keeps MSFS off of the paired virtual CPU on the core running the main thread, so the main thread doesn't share its core with other threads.

 

9 hours ago, 4L0M said:

I use it to allocate some of my e-cores on my 13600KF to run external apps like one e-core assigned to FS Realistic Pro, another e-core to run FSLTL, another e-core to run AnyoneFly etc etc.

You can also tell Process Lasso to always run MSFS with High Priority all the time with one simple click, or instantly make MSFS use maximum performance CPU mode with one click etc.

Looks like I can consider this scenario, instead of just turning SMT off.

 

23 hours ago, Bob Scott said:

I've observed that the main thread runs on the last core (core 9 in my case).

How did you find this out? Unlike P3D v5, I can't choose which core will run MSFS main thread.

 

 

 

[Bob Scott 7,750]

2 hours ago, Luis Hernandez said:

How did you find this out? Unlike P3D v5, I can't choose which core will run MSFS main thread.

I observed MSFS processor loading using Process Lasso.  The last processor available is loaded up at near 100%, while the others are typically much less, with occasional peaks, probably when processing/loading the next scenery tile.

I turned HT off on my i9-10900K's last core (core 9), and MSFS consistently runs the main thread there, and turning off HT on that core prevents it from having to share the core with other threads running on its companion virtual CPU.  Performance appears to be the same as when I managed it by setting a CPU affinity for MSFS externally with Process Lasso.  I do prefer doing it in the BIOS, though, as it presents no interference with the Windows scheduler that way.

[Luis Hernandez 587]

19 hours ago, Bob Scott said:

I observed MSFS processor loading using Process Lasso

Understood. I did so, but I found in my 5600x (with SMT on) that both Core0 and Core4 were the most heavily loaded ones. I couldn't tell which one handled the main thread. I guess I'll need to test more; FSHud eats more FPS than my AI, even using FSLTL in both cases.

 

[Bob Scott 7,750]

On 12/10/2023 at 8:48 AM, Bob Scott said:

In a variation on this theme, I had been running MSFS on my 10-core 10900K with HT off.  I've observed that the main thread runs on the last core (core 9 in my case).

As an experiment I lowered my all-core overclock to 4.8 GHz from 5.1 GHz, turned HT back on, and since MSFS doesn't have a config option to manage processor affinity, I used Process Lasso to externally set an affinity mask that keeps MSFS off of the paired virtual CPU on the core running the main thread, so the main thread doesn't share its core with other threads.  In other words, Process Lasso allows MSFS to use virtual CPUs 0-18, and blocks vCPU 19 so that the core running the main thread doesn't have to share its processor time with other threads.

This is an alternative to the all-or-nothing HT on or HT off approach, and it seems to have significantly mitigated the stability issues I was having with memory access violation (0xC0000005) crashes, and without any real performance hit.

So after some further testing, I've gone back to HT completely off--with HT on I was observing small microstutters in the Fenix displays in climbs/descents.  With HT off those microstutters completely smoothed out, and there was no drop in frame rate. 

I did leave the all-core overclock at 4.8GHz, and slowed my RAM down a notch from 3600 to 3500 MHz, and MSFS is still running stable.  My gut feel is that MSFS is hammering the IMC really hard, and that may have been the genesis of intermittent CTDs caused by memory access violations (0xC0000005  errors).

[MrFuzzy 2,124]

10 hours ago, Bob Scott said:

So after some further testing, I've gone back to HT completely off--with HT on I was observing small microstutters in the Fenix displays in climbs/descents.  With HT off those microstutters completely smoothed out, and there was no drop in frame rate. 

I did leave the all-core overclock at 4.8GHz, and slowed my RAM down a notch from 3600 to 3500 MHz, and MSFS is still running stable.  My gut feel is that MSFS is hammering the IMC really hard, and that may have been the genesis of intermittent CTDs caused by memory access violations (0xC0000005  errors).

Yeah, I also did a few more tests in these days to confirm the first results. 

This is the NY landing challenge with the 747 tested with the same addons I use daily, same settings and a fixed cockpit view (all 2nd runs to eliminate caching inconsistencies)

HT/SMT Off provides a better experience, right where hyperthreading is supposed to help (1% lows, 0.1% lows, stuttering...)

I remember having done the same tests with SU10 and the result was that HT on was slightly better. Asobo has changed something with the following sim updates and that's not good news. In particular with SU14, stutters on the ground at large airports have come back.

 

Edited December 15, 2023 by MrFuzzy

[GSalden 4,700]

Both my 13900 and 13600 are running at higher turbo speeds than with HT off. Same good stability.
The cooling never was an issue .Both liquid cooled. 

Enough cores on both of them so no need to turn HT on here ….