P3D V4.5 settings and Scenery

[dave2013 1,130]

I recently switched to P3Dv4 from FSX:SE.  I also built a new machine: i7 9700K, 32GB 3600MHz RAM, GTX 2070 Super Video Card, M.2 SSDs.

I haven't touched the configuration file and have managed to achieve very good performance. Of course, I am using reasonable graphics settings, i.e. FPS locked at 35 and no Vsync which is plenty for smooth rendering, no sliders at "max" or "extremely dense", no Dynamic Lighting or Reflections, shadows only for aircraft and clouds, 2048 texture resolution.  However, I do use 8x SSAA now as this greatly reduces shimmering of lights and autogen and doesn't adversely affect performance thanks to the GTX 2070 Super.

Frankly, I don't think that these config file hacks are necessary, or even helpful, any more providing you have adequate hardware and reasonable graphics settings.

Just my opinion.

Dave

[Noel 4,499]

9 hours ago, SteveW said:

OK well done Steve.

@Noel, No worries I understood your question and concern. I hope someone can come forward with the similar specification for you. I was using a much bigger CPU and strong GPU setup. 

The differences we get in the sim with changing around the sim cores is very subtle and hardly impacts the actual way the sim performs alone. I'm saying the big differences when we change the sim AM, even just rearranging cores, must be down to how the remaining systems find themselves operating, and how they become arranged. Also as you mention the important factor when testing is that the locked/vsync/tb/unlimited situation has a wide variety of effects. Then we have the type of monitor, and the cable, hdmi, display port, all play a part in the final outcome. My point is that when we see changes to performance with only rearranging the AM that means there's somehow some other thing or things impacting the outcome. So at this stage it's better to accept one good AM and concentrate on the remaining setup. The AM and corralling of addons is a final step to obtain maybe up to 5 or 8% more stable flow of the sim. We can't add performance.

My advice is that the latest high end CPUs especially 8 cores and more all work very close in performance on P3D because they are drawing from similar base hardware now like the memory SSDs and GPUs. get the better GPU rather than CPU if money permits. Extra cores or LPs only really gain loading speed, which does effect the flow of the sim if we can load it quicker at any time, but that burst of loading pushes out the system resources at the same time. So in short I think you will find more or less the same performance is seen around most those PCs in the long run.

Thanks for all of that.  Ever since unlimited + vsync to 30 was implemented I was able to achieve fabulously smooth video, reliably, provided I adjust to keep overhead such that frames don't dip below 30 which is accomplished by being careful where I fly and in what.  And that is with hardware entering its 7th year of operation!   I'm planning to do my next build when RTX 3080Ti arrives so by then not sure whether it's going to be 9900K or 9700K or Ryzen or what have you but in any case I will see a very meaningful boost in particular where I can set GPU related sliders as right now those eat up my old Titan.  

[Noel 4,499]

1 hour ago, Chapstick said:

I’ve tried AM=340 with my 8700k/HT on and got the worst stutters I’ve ever seen. Just panning around was unbearable. What could that signify? What’s the next setting to test?

Do you have perfectly stutter-free video normally Chap?  If not, check to see if your monitor supports 30mHz vertical refresh rate.  If it does all you need to do is use AM=4084, then use Task Manager to manually set affinity for all other apps and services running to Core0 or any other core but Core1 then set frames to UNLIMITED in P3D, VSYNC on, and set your monitor to 30mHz refresh.  I have a going on 7y/o system and have fabulously smooth video using these settings.  As long as you set sliders to where you never drop below 30 you will have perfectly smooth stutter-free performance--just watch CPU utilization and if it's approaching 100% this is your clue you've set some of the CPU related sliders too high for what your hardware can manage in that specific flight scenario.

[chapstick 2,757]

32 minutes ago, dave2013 said:

I recently switched to P3Dv4 from FSX:SE.  I also built a new machine: i7 9700K, 32GB 3600MHz RAM, GTX 2070 Super Video Card, M.2 SSDs.

I haven't touched the configuration file and have managed to achieve very good performance. Of course, I am using reasonable graphics settings, i.e. FPS locked at 35 and no Vsync which is plenty for smooth rendering, no sliders at "max" or "extremely dense", no Dynamic Lighting or Reflections, shadows only for aircraft and clouds, 2048 texture resolution.  However, I do use 8x SSAA now as this greatly reduces shimmering of lights and autogen and doesn't adversely affect performance thanks to the GTX 2070 Super.

Frankly, I don't think that these config file hacks are necessary, or even helpful, any more providing you have adequate hardware and reasonable graphics settings.

Just my opinion.

Dave

Turning off dynamic lighting is a major handicap. There are many native P3D add-on airports/aircraft that require DL to be on for any sort of realistic night flying, so keeping that off isn't a practical suggestion. That's why these CFG tweaks are suggested: so that you can find a balance of settings that work in 90 - 95% of scenarios without having to make sacrifices, like not flying at night.

[SteveW 1,792]

14 hours ago, Chapstick said:

I’ve tried AM=340 with my 8700k/HT on and got the worst stutters I’ve ever seen. Just panning around was unbearable. What could that signify? What’s the next setting to test?

It wasn't because you used four cores or an AM since that is something that I noticed never happens in a decade of testing FSX and P3D.

Your poor performance must mean that something very odd about your PC or the test went wrong or was tested while the system was busy on something else. Even a bad AM won't cause the

"worst stutters I’ve ever seen"

Steve

[SteveW 1,792]

To test performance of an AM we measure the loading time of the scenario, fps is relatively unaffected.

When we fly into a new zone of scenery that scenery is fed in gradually as we approach the scenery.

That is where performance matters. That is where we don't need all the LPs of the big CPUs to light up as that takes throughput from the main task causing variation in the fps - and that is stutter as long frames appear.

Panning smoothness is down to the camera setting, how much it turns per frame that is always an issue. So this is different depending on fps.

That and turning of the plane is affected greatly by VSync and locked and limited and monitor refresh cycle modes.

That's where stutter comes from. If the incidence of frame to refresh cycle is out then even a small INCREASE in fps can get more stutter from frames arriving early.

The AM/HT debates roll on and I see no reason for it.

The AM and HT are not Tweaks in the normal sense of the word. HT is not a tweak it is a CPU type. The number of cores we have or choose to use for a program is not a tweak, it is simply the number of cores put to use.

Folk with P3D and FSX have been getting OK performance form four core CPUs for a decade, any similar CPU with four or more cores, especially recent CPUs should be able to perform the same.

An AM makes a difference by restricting use of cores to enable more throughput to parallel process to those that run alongside the sim, system resources file and network, addon resources network Simconnect client and so on. Don't restrict cores unless you need to. It doesn't improve performance, we use it to allow performance in the accompanying tasks and to prevent the Sim sharing the main task with another.

Bad performance with an AM or HT, that's because something else is wrong. I would put AM 340 onto a six core +HT as a starting point. That's four straight cores and two for system and addon resources. Then to improve loading speed we can go up to AM4084 as Noel suggested. The difference in actual performance will be very hard to spot, the loading speed increased will be in the region of 15% the fps will be the same.

 

Edited January 12, 2020 by SteveW

[Noel 4,499]

9 hours ago, SteveW said:

To test performance of an AM we measure the loading time of the scenario, fps is relatively unaffected.

When we fly into a new zone of scenery that scenery is fed in gradually as we approach the scenery.

That is where performance matters. That is where we don't need all the LPs of the big CPUs to light up as that takes throughput from the main task causing variation in the fps - and that is stutter as long frames appear.

Panning smoothness is down to the camera setting, how much it turns per frame that is always an issue. So this is different depending on fps.

That and turning of the plane is affected greatly by VSync and locked and limited and monitor refresh cycle modes.

That's where stutter comes from. If the incidence of frame to refresh cycle is out then even a small INCREASE in fps can get more stutter from frames arriving early.

The AM/HT debates roll on and I see no reason for it.

The AM and HT are not Tweaks in the normal sense of the word. HT is not a tweak it is a CPU type. The number of cores we have or choose to use for a program is not a tweak, it is simply the number of cores put to use.

Folk with P3D and FSX have been getting OK performance form four core CPUs for a decade, any similar CPU with four or more cores, especially recent CPUs should be able to perform the same.

An AM makes a difference by restricting use of cores to enable more throughput to parallel process to those that run alongside the sim, system resources file and network, addon resources network Simconnect client and so on. Don't restrict cores unless you need to. It doesn't improve performance, we use it to allow performance in the accompanying tasks and to prevent the Sim sharing the main task with another.

Bad performance with an AM or HT, that's because something else is wrong. I would put AM 340 onto a six core +HT as a starting point. That's four straight cores and two for system and addon resources. Then to improve loading speed we can go up to AM4084 as Noel suggested. The difference in actual performance will be very hard to spot, the loading speed increased will be in the region of 15% the fps will be the same.

With all of this clear explanation of the role of an AM and assigning tasks off the main thread core, I'm still not sure which CPU I should upgrade to for my next build for P3D (and the new MSFS of course), a 9700K, or a 9900K.  You state, "we don't need all the LPs of the big CPUs to light up as that takes throughput from the main task causing variation in the fps - and that is stutter as long frames appear." and yet w/ HT enabled using AM 4084 I have not only superior static load time but for reasons I can't appreciate I also get a steady stutter when in complex scenery when I set up the 3930K w/ HT disabled using a similarly configured AM, for example 62 (111110) and reassign other tasks off of Core1.

[SteveW 1,792]

29 minutes ago, Noel said:

With all of this clear explanation of the role of an AM and assigning tasks off the main thread core, I'm still not sure which CPU I should upgrade to for my next build for P3D (and the new MSFS of course), a 9700K, or a 9900K.  You state, "we don't need all the LPs of the big CPUs to light up as that takes throughput from the main task causing variation in the fps - and that is stutter as long frames appear." and yet w/ HT enabled using AM 4084 I have not only superior static load time but for reasons I can't appreciate I also get a steady stutter when in complex scenery when I set up the 3930K w/ HT disabled using a similarly configured AM, for example 62 (111110) and reassign other tasks off of Core1.

Still stands exactly what I said. I'm not using 18 cores for P3D due to precisely the reasons I stated. If you have a six core then how's that compare? Not too well. Mostly you need all six.

So you are looking for balance with the addon exe apps and the system resources invoked. If that's all your cores then that's OK you only have six.

When you apply an AM the system resources invoked reside inside that AM cost use throughput. That throughput is drawn from the system. That draw changes the main task.

Think of adding more cars to a coal train and putting less coal in each. You are now pulling more cars. You need the optimum carried by the car to reduce the total cars weight. Not exactly a great analogy but kind of works like that.

 

Edited January 12, 2020 by SteveW

[Noel 4,499]

Seems like you're ignoring what I keep saying.  I keep stating 12 LPs is superior to 6 Ps in my configuration because as I said I get a regular interval stutter in very dense scenery when I disable HT, yet when I ask does it then follow that 16 LPs will be possibly also be superior to 8 Ps, I get the THEORETICAL reason why it clearly won't be.  

[SteveW 1,792]

I'm saying if it doesn't work out better, then that will be the reason. Perhaps I misunderstand your question. maybe rephrase it.

[SteveW 1,792]

@Noel, I think I long since answered your question. Eventually you reach a limit of cores or LPs, whatever that is. HT enabled means you can double that count so it Is theoretically possible to come up to the limit quicker. That's one reason why HT is not so well favoured.

 

[SteveW 1,792]

Having spoke about this with Gerard and some others, and testing here, the feeling is that for a big system you might need ten or more cores with P3D on eight of them.

[Noel 4,499]

20 minutes ago, SteveW said:

@Noel, I think I long since answered your question. Eventually you reach a limit of cores or LPs, whatever that is.

It's the 'whatever that is', that limit of cores or LPs, that I was trying to get a hard answer for, and apparently it's not easy to answer this question precisely.  This doesn't surprise me because P3D is a fickle app that is very dicey to pin down because of the vast range of demand it can dish out.  OK I'll take this conclusion to mean go for the 9900K, and play around w/ restricting some of that capacity as required to optimize the various tradeoffs.   And pick up that RTX 3080Ti 

Edited January 12, 2020 by Noel

[SteveW 1,792]

14 minutes ago, Noel said:

It's the 'whatever that is', that limit of cores or LPs, that I was trying to get a hard answer for, and apparently it's not easy to answer this question precisely.  This doesn't surprise me because P3D is a fickle app that is very dicey to pin down because of the vast range of demand it can dish out.  OK I'll take this conclusion to mean go for the 9900K, and play around w/ restricting some of that capacity as required to optimize the various tradeoffs.   And pick up that RTX 3080Ti 

Yes, that's it in a nutshell. Since every single system has to be set up per the expectations of the user and the limitations of the PC and the scenery size and detail levels play a big part.

For example you could try 20 locked on the slider, or a similar division of your monitor frequency, and Vsync = Off. But turn settings down for fast jets and gliders for higher fps when needed and VSync then.

[Noel 4,499]

29 minutes ago, SteveW said:

Yes, that's it in a nutshell. Since every single system has to be set up per the expectations of the user and the limitations of the PC and the scenery size and detail levels play a big part.

For example you could try 20 locked on the slider, or a similar division of your monitor frequency, and Vsync = Off. But turn settings down for fast jets and gliders for higher fps when needed and VSync then.

I'm actually very comfy w/ how to set up the sim in large part thanks to your comments over the past few years.  I'm the guy using a 6.5y/o system and bragging about wonderfully smooth, stutter-free video, maintained at 30 frames via the monitor sync, in the context of plenty here with high end systems still asking about stuttering w/ an AM of 340!  It will be great to have some more processing power to play with, and again I've been so happy w/ vsync to 30mHz I will just follow the same approach and it should work great, to the tune of about 30+% better overall and that's huge.  The true holy grail IMO would be to vsync to 60, but that isn't going to happen w/ a 9900K/RTX 3080Ti so we will be happy w/ 30.   I'm especially excited to see what the GPU brings as right now the old Titan is pretty hamstrung vis a vis the GPU related sliders.   Do have have a sense of what role the GPU plays in autogen/vegetation, i.e. the traditionally thought of as CPU related sliders?  I understand in the evolution of P3D the GPU may be playing an enhanced role for those sliders versus earlier iterations of P3D.

Edited January 12, 2020 by Noel