11 minutes ago, Rockliffe said:

Hyperthreading off, no affinity mask, monitor refresh rate at 30hz, lock frames at 30 and use P3d Vsync. Do NOT be too optimistic with sliders. 😉

I second it.

CPU: HT Enabled

P3D: AM=340

2 hours ago, marc1184 said:

Not sure exactly why it worked, but I turned off hyper threading in BIOS yesterday and it took care of my stutters.  Of course, it worked for me, but may not for you.

This was going to be my next point for Intel owners if the above didn't work.

A lot of users with 8 or more core CPU's (like the 9900K) have reported better smoothness with Hyperthreading off.  

Now, for the AMD Ryzen users, many of those cpu's have 8-12+ cores.  I do not know if they have anything like Intel's Hyperthreading, and if so, whether or not it can be turned off or otherwise disabled.  I think it is worth experimenting with, if you have one of those cpu's, to see if it results in smoother flight.  Unfortunately I'm not currently very knowledgeable about AMD's processors...I'm sure others are in this subforum and may have more info.  I seem to remember a thread or two about Ryzen cpu's and P3D here.

4 hours ago, Mace said:

A lot of users with 8 or more core CPU's (like the 9900K) have reported better smoothness with Hyperthreading off.  

We don't ignore the AM with HT enabled or disabled since with many cored systems they need a setting to allow the least amount of cores they can use successfully.

Too many cores (or HT LPs) load up the system for no benefit. See how many cores are required to load the scenario fastest, 4 to 6 probably enough. After that they continue to load up the bus without gathering more stuff. The outcome burdens the main task.

HT enabled AM=340 should be a good starting point.

 

Here my server has a 14 core 7940 processor from which I use the 8 fastest cores for P3Dv4. Therefore I am using an AM. My one wide view is 2x 43” 4K so I choose 8 cores. Did I had only 1 display I would have choosen 6 cores for P3D ...

Not using an AM and using all cores is making the sim not smoother. 
 

The other 6 cores are for addons and using batch files with AM to execute them.

Imho if you are using addons on the same cpu then the correct AM will help smoothing the sim...

21 hours ago, marc1184 said:

Not sure exactly why it worked, but I turned off hyper threading in BIOS yesterday and it took care of my stutters.  Of course, it worked for me, but may not for you.

I checked if I had hyperthreading turned on which I did so turned it off and was able to oc the CPU up to 4.8 couldn't do this before as the temps went  too high but now temps are ok so no more stutters i didn't realize ht was automatically enabled in the bios.

thank you

Stutters are less because turning HT off halved the task count of P3D proving that P3D needs less tasks or over burdens the rig as I pointed out.

Turning HT off is a quick helper fix but not the right way, since it just reduces tasks load and saves one the hassle of understanding a proper HT setup, but wastes the expensive HT mode.

Better to understand what it is, than ignore it because it's hard to understand.

...I'm testing 9980xe and 9900x systems with 2080tis that's 18 and 10 cores, or with HT enabled 36 and 20 threads or P3D tasks with no AM. You don't need that many. I say ignore the HT off No AM crowd and work on your setups.

Let's look at a six core with HT:

HT enabled binary:
11,11,11,11,11,11
AM=4095

HT Disabled
1,1,1,1,1,1
AM=63

The equivalent with HT enabled:
01,01,01,01,01,01
AM=1365

 or 

10,10,10,10,10,10
AM=2730

They do the same thing; direct all threads to use each core single threaded as does disabling the expensive HT.

HT increases core throughput by saving half the switching time, maybe more that 8% on P3D system.

So we leave the system in HT, but must apply a suitable AM that keeps the P3D primary tasks to single Logical Processors of cores (or FSX or XP).

On a six core system we might use AM=340 which is 00,01,01,01,01,00 whereby we keep four P3D tasks on the six cores, each to a core to themselves.

And as Gerard mentioned, use the spare cores for addons, so they don't interfere with the sim cores.

Remember that even a "puny 6% task" appearing on the main sim core interrupts that core and interrupts the pre-ordering of memory transactions unnecessarily.

 

Edited October 23, 20196 yr by SteveW

2 minutes ago, SteveW said:

HT increases core throughput by saving half the switching time, maybe more that 8% on P3D system.
 

But the question remains: how much % throughput gives you extra 300Mhz that you can squeeze out from your O/C by turning the HT off?

Thanks. 

Well, basically, heat=work done, so the same heat must have done the same work. It's all down to how the work is arranged.

Remember the general advice is to start with HT disabled.

Find the number of cores needed to load the scenario fastest.

Start tests with AMs like...00000011110 =(01,01,01,01,00 = 340 in HT) is a good starting point and avoids core zero

More cores added with no significant reduction in loading time are wasted.

Put addons on the spare cores.

Next go to HT enabled and set up with '01's in the AM.

Most high end CPUs more or less come out equal with P3D properly configured, use extra cash for GPU.

Edited October 23, 20196 yr by SteveW

On 10/22/2019 at 11:19 PM, SteveW said:

See how many cores are required to load the scenario fastest, 4 to 6 probably enough. After that they continue to load up the bus without gathering more stuff. The outcome burdens the main task.

Thanks Steve for your thoughts again.

You spurred me on to spend some time today testing my system out to see if I could eliminate my remaining small stutters, in search of the glassy smooth sim.

The system is a delidded 8700K (6 core +HT, HT on) overclocked to run at 5GHz permanently, all cores, including when running AVX2 instructions (ie AVX offset of 0). System is rock stable at this clock, used daily as a music production workstation where critical timing an low latency is a must. 32GB ram, GTX 980Ti overclocked. NVMe M.2 boot drive and 5xSSD SATA data drives. All rock steady and 100% reliable. All folders involved with P3D and it's addons are excluded from AV and Malware scans. I'm running P3D 4.5+HF1. All addons are up to date.

The time to load P3D from the desktop to the initial load screen was a consistent 11 seconds +/- 0.5 seconds. The times listed below are for loading P3D from the initial screen to in-sim, ready to fly, using the same scenario each time. I'm using full Great Britain photo scenery so load times are longer than they would perhaps be with non-photo scenery. For these tests, no background tasks were loaded, such as SkyForce3D, or weather engines etc. Just plain P3D.

So here is a list of the Affinity masks I used and the timings, sorted from slowest to fastest.

AM      Binary                       Load time (seconds)

4         00 00 00 00 01 00    120
20       00 00 00 01 01 00    115
84       00 00 01 01 01 00      90
340     00 01 01 01 01 00      90
1364   01 01 01 01 01 00      70
4083   11  11 11 11 00 11      58
1365   01 01 01 01 01 01      54
4085   11  11 11 11 01 01      48
4081   11  11 11 11 00 01      48
4095   11  11 11 11 11 11       44

Now you'd think from that list that AM 4095 would be the way to go, but it isn't. The sim runs smoothest with AM of 1364. I don't use core 0 at all, leaving that for operating system stuff and any other random background processes. And all my P3D addons are assigned to the other 5 core threads not used by P3D through use of a batch file.

With all that in place, the sim is maintaining 30fps at all times and is very smooth.

Except for one thing. I'm getting the most horrendous occasional sim hang, for maybe 10 seconds. All activity ceases, to the point that I'm expecting a BSOD, then after about 10 seconds, everything continues perfect. This is rare - it happened just once this evening in a two hour flight, and at the time I was in the middle of the Scottish Highlands, so far from any large airport scenery that might of been loading at the time.

https://photos.app.goo.gl/n6UbJnqtcKGMrTAx6

Has anyone had the same and found a cure for this. It's not a stutter. Looking at the screen grab on the link above, it's much worse and shows the CPU and GPU taking a dive in performance (red arrows). Core 0 Thread 1 (green arrow) got busy just before this pause, and Core 6 Thread 2 (purple arrow) also ramps up just before the pause. Core 0 is not assigned to the sim or any addons. Core 6 Thread 2 is assigned to FFTF Dynamic, and ASP4 (Active Sky).

Anyone got any thoughts on what is happening here or found a fix? Is there a known issue with FFTF Dynamic or ASP4? I've had quite a few of these long pauses, and it would be great to get rid of them.

 

 

 

Edited October 25, 20196 yr by SledDriver

19 minutes ago, SledDriver said:

Thanks Steve for your thoughts again.

You spurred me on to spend some time today testing my system out to see if I could eliminate my remaining small stutters, in search of the glassy smooth sim.

The system is a delidded 8700K (6 core +HT, HT on) overclocked to run at 5GHz permanently, all cores, including when running AVX2 instructions (ie AVX offset of 0). System is rock stable at this clock, used daily as a music production workstation where critical timing an low latency is a must. 32GB ram, GTX 980Ti overclocked. NVMe M.2 boot drive and 5xSSD SATA data drives. All rock steady and 100% reliable. All folders involved with P3D and it's addons are excluded from AV and Malware scans. I'm running P3D 4.5+HF1. All addons are up to date.

The time to load P3D from the desktop to the initial load screen was a consistent 11 seconds +/- 0.5 seconds. The times listed below are for loading P3D from the initial screen to in-sim, ready to fly, using the same scenario each time. I'm using full Great Britain photo scenery so load times are longer than they would perhaps be with non-photo scenery. For these tests, no background tasks were loaded, such as SkyForce3D, or weather engines etc. Just plain P3D.

So here is a list of the Affinity masks I used and the timings, sorted from slowest to fastest.

AM      Binary                       Load time (seconds)

4         00 00 00 00 01 00    120
20       00 00 00 01 01 00    115
84       00 00 01 01 01 00      90
340     00 01 01 01 01 00      90
1364   01 01 01 01 01 00      70
4083   11  11 11 11 00 11      58
1365   01 01 01 01 01 01      54
4085   11  11 11 11 01 01      48
4081   11  11 11 11 00 01      48
4095   11  11 11 11 11 11       44

Now you'd think from that list that AM 4095 would be the way to go, but it isn't. The sim runs smoothest with AM of 1364. I don't use core 0 at all, leaving that for operating system stuff and any other random background processes. And all my P3D addons are assigned to the other 5 core threads not used by P3D through use of a batch file.

With all that in place, the sim is maintaining 30fps at all times and is very smooth.

Except for one thing. I'm getting the most horrendous occasional sim hang, for maybe 10 seconds. All activity ceases, to the point that I'm expecting a BSOD, then after about 10 seconds, everything continues perfect. This is rare - it happened just once this evening in a two hour flight, and at the time I was in the middle of the Scottish Highlands, so far from any large airport scenery that might of been loading at the time.

https://photos.app.goo.gl/n6UbJnqtcKGMrTAx6

Has anyone had the same and found a cure for this. It's not a stutter. Looking at the screen grab on the link above, it's much worse and shows the CPU and GPU taking a dive in performance (red arrows). Core 0 Thread 1 (green arrow) got busy just before this pause, and Core 6 Thread 2 (purple arrow) also ramps up just before the pause. Core 0 is not assigned to the sim or any addons. Core 6 Thread 2 is assigned to FFTF Dynamic, and ASP4 (Active Sky).

Anyone got any thoughts on what is happening here or found a fix? Is there a known issue with FFTF Dynamic or ASP4? I've had quite a few of these long pauses, and it would be great to get rid of them.

 

 

 

 

Good demonstration of the technique. The more cores you add the less time is taken to load the sim. After a point we only see a second or two gained. So it's there we should be looking. In your case the final big gain is at 1364 - 5 LPs on 5 cores. Shows how easy it is to get set up when you know how. Give addons at least two LPs.

Also your testing clearly shows how certain tasks can gain advantage in HT mode by using '11' on a core for some of those on CPUs with fewer cores where the advantage of HT provides opportunity.

Sim hanging:  Turn off a few addons, traffic maybe see where the hang comes from by elimination..