A Frame-Time Analysis of P3D v3 -- Effects of CPU Affinity, Frame Lock, and HT

[SteveW 1,792]

Steve

 

What an extremely interesting topic. So much to test, it seems one setting doesn't suit all. Are you able to publish your findings in one document for easy reference. Your research is much appreciated, thanks.

 

Hi Steve, I'm hoping the message gets across, and someone better than me at describing things will have a go at that. But I've made several info pages on the codelegend forum Installation section which you could have a read of at your leisure!

 

 

 

That would be true if one didn't also have concurrently test ASN, GSX, TrackIR, STB, the Traffic Optimizer and a few other addons that are executables. A few like Fraps don't enjoy being told what core to use.

But Jay, of course that's true. However, to make the performance tests with specially constructed software is a breeze. Having the correct approach to work this out saves time, but that approach requires a working understanding of HT and the way Windows apps work. That's why I've put my time into it, to save you yours. With your 12 core, there should be ample cores for running other stuff away from the sim cores utilising the AM techniques I've described. As long as you protect the first sim job from sharing a core, the sim will work fine so long as the other three jobs get reasonable bandwidth. When we have more than 4 cores we don't need to mix apps on so many cores, but if we do, avoid mixing the main sim job.

[Cruachan 705]

Hi Steve, Daniel

 

Any tips on how to post a nice clear image of a FRAFS Viewer trace? You are right Steve, when I compare my image with those uploaded by Daniel in his original post mine is very fuzzy. Also, I note Daniel tested over 5 mins whereas I did 2 mins yet my trace appears much more dense and, as you said, Steve, hard to read.

 

Regards,

Mike

[Sirxamoc 22]

After reading this post and another about stutter free frame rates I decided to try a program last night on my PC, Process Lasso.  I was really impressed with the quality of my system.  I was able to see in real time what the effect was when I restricted a core from one program and directed it to for only certain programs.  I have an I5-4670k 4-core processor with no HT.  I was able to use core one for certain functionary programs and able to spread the work load for P3D across core 1-3.  With my processor turned up to 4ghz, and my Nvidea 680 it was awesome to takeoff out of JFK in a PMDG 737 with Traffic 360-ON & traffic at 40%; EFB-ON; Next-ON with multi layered cloud base; ACARS-ON; and some other background programs going - NO STUTTERS!

 

It is not going to happen at once or automatic, after watching a couple of Vids I learned that I had to really load my system up, then fly with it and tune it at the same time.  Brilliant.  Plus the nice thing is if you change by adding another program, scenery file, or whatever you can make immediate adjustments to level the work across the cores.

[luke97 8]

Hello Simmers;

 I need to find out the best config. for Nvidia Inspector for my Prepar3d V2.5 and V3

 

I can`t find a good config for it and I think because of that I`m getting instability issues.

 

 My PC specs and tweaks are:

 

Windows 10 64x pro

 

Intel i7 2600K OC to 4.3Ghz using Corsair Hydro liquid cooler

8GB of RAM 1866

SSDs

GPU: 2 x ASUS GTX 660 Ti DirectCu "SLI"

 

Maybe some has the same config.

 

I`ll be so grateful for the assist and if you need my .CFG info, let me know.

 

Thiago 

​

 I have i7 2600k too, and I always thought NO HT and NO AM on .cfg , were the best options for Prepar3D 3. Now, after all those posts, Im not so sure.

 

SteveW, do you recommend AffinityMask=14 for i7 2600k and no HT (4 cores) ?  Sorry if it was asked before, but I cant find it.

 

Thanks very much for so interesting thread.

[SteveW 1,792]

I have i7 2600k too, and I always thought NO HT and NO AM on .cfg , were the best options for Prepar3D 3. Now, after all those posts, Im not so sure.

 

SteveW, do you recommend AffinityMask=14 for i7 2600k and no HT (4 cores) ? Sorry if it was asked before, but I cant find it.

 

Thanks very much for so interesting thread.

You're right to be cautious, "NO HT and NO AM" is not the optimum for P3D or FSX, and as we get CPUs with more and more cores, that will become more and more obvious.

 

For the four core, if you are not going for ultimate GHz, turn on HT and use AM=116 or AM=85. Otherwise with HT off use AM=14 or no AM (AM=0). Generally they will give you good results, see how they go.

[bthiago 8]

I`m actually stayed with P3D V2.5 with a nice tweak without the Nvidia Insp.

What is those really kills your FPS in the TExture at or above 2048. I kept mine on 1024 and didn`t see no differences.

[Richard Sennett 2,499]

You're right to be cautious, "NO HT and NO AM" is not the optimum for P3D or FSX, and as we get CPUs with more and more cores, that will become more and more obvious.

  :smile: Steve is a very smart Guy  :wink:

[Guest]

Experiment ... see how it works for you.

 

For my 5960X , I'm now testing out 8C HT off and (with TrackIR)

 

[JOBSCHEDULER]

AffinityMask=254

 

Also removed Texture_Bandwith_Mult entry and found that resolved LODs faster and overall performance is good at extreme graphics settings.

 

FYI, I assume most know that P3D has a Performance Tuning section in their Learning Center.chm file?

 

But I think it's important to note that P3D V3.1 has some issues with Building autogen and DR that has been resolved for next release.  Not saying this will make a difference in tuning, but just like racing, don't "chase" the setup, wait for the track to come to you.

 

Cheers, Rob.

[Guest JustanotherPilot]

 

 

FYI, I assume most know that P3D has a Performance Tuning section in their Learning Center.chm file

 

Rob,

 

Wasn't aware of this can you point to it's location.

[Skywolf 992]

 

 

Also removed Texture_Bandwith_Mult entry and found that resolved LODs faster and overall performance is good at extreme graphics settings.

 

Rob,

 

Why are you removing that entry from the cfg file?  This is the first time I am reading about this.  Isn't this entry default in the cfg

[Cruachan 705]

Also removed Texture_Bandwith_Mult entry and found that resolved LODs faster and overall performance is good at extreme

 

Yes, I agree. I have tried several higher values and none produced better results than the default 30. I believe the sim assumes 30 if the entry is removed. My system has similarities to Rob's with three important differences - I run two GPUs in SLI whereas Rob is currently preferring one; my memory is running at 2400 whereas Rob's, I think, is clocked at 3000 and, lastly, my monitor refreshes at 120Hz (No DSR) while Rob's = 30 with DSR. Not sure how much impact all this has other than the fact that my two GPUs will be delivering performance closer to Rob's more powerful singleton.

 

Steve, just in case you are waiting with bated breath (LOL!) I haven't managed to get around to running those tests as yet - will do so ASAP

 

Mike

 

Edited: first paragraph.

[Skywolf 992]

Yes, I agree. I have tried several higher values and none produced better results than the default 30. I believe the sim assumes 30 if the entry is removed. My system has similarities to Rob's with two differences - I run two GPUs in SLI whereas Rob is currently preferring one and my memory is running at 2400 whereas Rob's is clocked at 3000.

 

Steve, just in case you are waiting with baited breath (LOL!) I haven't managed to get around to running those tests as yet - will do so ASAP

 

Mike

 

My setup is similar to both you guys - only thing is I am using i7-5820K.

 

Are you running with HT OFF or ON and what are your AM settings?

 

I am getting good performance with HT off and AM set to 30 (4 cores)

[Cruachan 705]

Are you running with HT OFF or ON and what are your AM settings?

 

Currently HT=OFF and NO Affinity Mask. However this may change.....quite soon! Watch this space

 

Mike

[Guest]

Why are you removing that entry from the cfg file?

 

Experimenting and because it's not in LM's performance tuning section.  I think Steve has been pretty consistent in his approach and testing and has repeatedly suggested "try and see".  I have a lot of variants to try with a focus on stable OC ... for example 4C (no HT) and I can run 4.8Ghz very stable, but at 8C (no HT) 4.625Ghz seems to be my cap.  8C 16T produces the best "numbers" but it seems to consume more VAS and P3D performance is down and stability is reduced.

 

Also keep in mind I'm talking 2 fps difference (in higher OC) ... but when I'm at complex scenery with complex aircraft and just dropping to occasional 28 fps, I get the occasional stutter, so for 30Hz Vsync that extra 2 FPS will get me back to solid 30FPS and remove the occasional stutter.

 

I'm using AM to free up Core 0 for TrackIR.  So as of now 8 real cores, 1 reserved for OS/TrackIR/ShadowPlay (these all seem to work well on Core 0) and the rest for P3D.  I run ASN and FS Commander on a separate PC.

 

That's my focus, which maybe entirely different from others desires.

 

Cheers, Rob.

[DylanM 1,069]

 

 

But I think it's important to note that P3D V3.1 has some issues with Building autogen and DR that has been resolved for next release.

 

This is very...very exciting Rob! If this one is licked, that's one significant tick-up in performance.