Micro-Stutter Easy Fix

[Bohemond 17]

Hey, Everyone!  I thought I'd put my 2 cents worth in on a very hot topic here at Avsim concerning terminating micro-stutters.  Many of you may already have tried this or do it now, but maybe some here haven't tried this simple fix.  And it may not work for everyone.  It's the "Disable Core 0 and Immediately Enable Core 0" procedure after starting P3D, so it's nothing revolutionary and it's probably well-known, I think. 

My specs are listed in my signature and I run P3D v4.5 HF2 at pretty much max graphical settings except for low shadow quality.  Basically all my sliders are to the right including autogen (draw distance and quantity).  In ASP4 I have it set to 20 cloud layers and 180nm draw distance.  I'm also using FFTF Dynamic to dynamically adjust FFTF and it yields an extra 4 to 5 fps. Frames are limited to 30 with VSync on.  My fps ranges, usually,  between 20 and 30 which is good enough for me since I'm a "graphics lady of the night"!  

I stumbled upon this in an old post here by SledDriver:

https://www.avsim.com/forums/topic/552330-p3d-multicore-usage-anomoly/page/3/

Also, I made a short video of how I go about this:

https://www.youtube.com/watch?v=uQV5jv9HCcs

I hope this helps some of you and thanks to everyone at Avsim for all the help I've received on this forum.

[chapstick 2,757]

This feels like it’s working for me, but stutters are hard to measure so YMMV. What I can say for sure is that the basic premise works on my system: P3D will go 100% on core 0 until I I reset the affinity with your instructions, then the load seems to spread across all six (8700k). 

Does anyone know the reasoning behind this? Is it a bug with P3D or maybe a system-specific thing? 

[Rogen 1,087]

All I see is that the load previously running on core zero moves to another core...

Although I did have P3D glitch on me about a week ago.

Basically I was flying around PNW and thought this is too smooth, something must be going on.

Popped open task manager and saw P3D barely using any CPU - less than 10 % across all cores (GPU was up around 75% though).

Switched off vsync and P3D was outputing mid to high 50s FPS, which is not normal for the area, then I noticed the ground textures suddenly became blurry and it was like I was trapped in a small area of non-burriness that didn't need any rendering.

I was so surprised I flew around enjoying the bubble of great scenery and high fps for a while before saving, killing P3D and restarting, where upon all was back to normal.

Cheers

Edited December 13, 2019 by Rogen

[newtie 533]

This has been around for a while, with varying opinions.

[Bohemond 17]

Definitely nothing new or revolutionary here and, as I stated, it has been debated here for a long time.  All I'm saying is it works for me.  Seems to vary per system.  Cya!

[SteveW 1,792]

Yes it's an observation of what can typically happen on systems that expect the jobscheduler to sort it all out. Each system is different and some may see that behaviour other different.

For over a decade I've been showing how to partition the system to avoid those anomalies with appropriate AMs. Better to set up a system properly in the first place than hope for the best.

[Bastlwastl 0]

41 minutes ago, SteveW said:

For over a decade I've been showing how to partition the system to avoid those anomalies with appropriate AMs. Better to set up a system properly in the first place than hope for the best.

Could you explain that to me please ?

[SteveW 1,792]

When rearranging affinity in Task Manager shows better results, it's possibly due to tasks sharing cores that become unshared after the move. Instead with Affinity Masks on those tasks sharing can be avoided in the first place.

Edited December 14, 2019 by SteveW

[SteveW 1,792]

Say with a six core we can use the AM 30 on the simulator (011110 = 30) and that leaves two cores free. Hopefully tasks causing the stutter stick to the free cores.

When we move threads in task manager, that's what happens.

Another way we can be more sure of, leave the sim with no AM (111111 = 63) and use a tool to assert affinity on the task so that is avoids the rightmost cores AM60 = 111100.

 

Edited December 14, 2019 by SteveW

[CaptKornDog 281]

No difference here, may have even introduced more stutters on my i7 7700k (HT on).  I use AM=85 already; tried it with no AM as well.  So seems to be truly a case of YMMV.

[Bastlwastl 0]

@SteveW

 

Ok thank you, but you said something about 

22 hours ago, SteveW said:

how to partition the system

and

22 hours ago, SteveW said:

to set up a system properly in the first place

What did you mean with that?

 

Thank you

[SteveW 1,792]

I refer you to my comments above.

[SteveW 1,792]

So to recap then:

The OP and other have noticed that stutter can be alleviated by altering the chosen CPUs in Task Manager for the Simulator.

The reason sometimes this works is that the result of altering the chosen CPUs for the Simulator program allows the Jobscheduler to rearrange tasks so that sharing of the main Simulator core is avoided.

My point is that we can avoid that in the first place by paying attention to where programs run alongside the simulator program.

So in my posts I showed how to partition the Simulator and the other exe apps so that they cause less interference.

Another aspect of partitioning is when we have many cores we find the simulation works better with four or six cores rather than eight, ten or more.

 

On 12/14/2019 at 10:20 PM, Bastlwastl said:

@SteveW

 

Ok thank you, but you said something about 

and

What did you mean with that?

 

Thank you

So I hope that helps.