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

[Mochi 82]

Fantastic, informed post.. Should be sticky or merged into a P3D tweaks post.

[SolRayz 392]

To the OP...Not sure why your having so many issues "out of the box". To be honest, no prior P3D version has been as stable and enjoyable as v3. I'm running a 2700k OC@4.6 with 970GTX at 1920x1080 with 4xMSAA in game and NI set to 4xSGSS, fps set to unlimited. Performance has been pretty spectacular over most scenery flying the NGX with settings much higher or max compared to the settings you posted. I did have some hiccups over the western coast of Italy with Orbx OpenLC Europe. Not sure whats up with that. I tried to isolate the issue, disabled options, but never could find the source of stuttering in that case. I also messed around with affinity mask settings in the cfg, and HT off and on, since that is all the craze again, but found no perceivable benefits to default.

[Poppet 709]

@Marc J.

 

As you know every system is different.  I find  V 2.5 much smoother "out of the box" then V3,  there is obviously something in my set up that V3 is slightly tripping over. This is my problem though.  

[Nyxx 6,993]

Incredible, well thought out and easy to see, credit to you.

 

I could not help smile as the guild I did for the 777/NG in by sig 6 months ago, the guild line was

 

HT ON

AM 84

FPS lock @ 30

FFTF 0.01 

 

As you have found the above is a good base to start with.

[carlito777 712]

Nice analysis. I like the fact based style of it as there are way too many posts on the forum which claim to have found the next silver bullet setting claiming the sim runs "sooo smooth" when applying it. Or the people claiming that P3D is so smooth out of the box simply becsuse they are unable to see the stutters that are definitely there.

 

One quick question though: Have you turned on vsync and triple buffering in P3D?

[Guest]

One quick question though: Have you turned on vsync and triple buffering in P3D?

Check the opening post... you may notice a few screenshots there...

 

I will give the overal best settings a try today (AF=85, HT=ON, FPS @ 30 and even FFTF=0.01). Curious to see how that runs, even though I use quite different settings (like extremely dense autogen).

[pmb 5,137]

If i use frames unlocked, it gives butter smooth performance in near all conditions, however, my GPU usage skyrockets....especially in clouds it goes to 99% !!

That's exactly what I observe. Contrary to what I read, I get better performace PLUS fluidity (just subjectively) with unlimited frames. Same CPU as the OP (except non-K). According to GPU-Z my GPU (EVGA 760 OC'd 4GB) runs always at 99%.

 

Kind regards, Michael

[virtualstuff 758]

That's exactly what I observe. Contrary to what I read, I get better performace PLUS fluidity (just subjectively) with unlimited frames. Same CPU as the OP (except non-K). According to GPU-Z my GPU (EVGA 760 OC'd 4GB) runs always at 99%.

 

Kind regards, Michael

 

Well you have to read this topics and results with a grain of salt.

Because the results are based on local hardware and skills ;-)

 

For you local it can be a total different story ;-)

 

Nevertheless always nice to read what the results are local from other simmers...

Maybe some simmers can benefit also with similar hardware and settings.

Thanks for sharing your endeavour local :-)

 

Cheers,

[carlito777 712]

Check the opening post... you may notice a few screenshots there...

 

Thanks for the hint, Jeroen. I did miss that one... ;-)

Well you have to read this topics and results with a grain of salt.

Because the results are based on local hardware and skills ;-)

 

Is there really a need to discredit the OP? Your answer comes across really arrogant. If you are Mr. Super-expert then come up with a better analysis.

[virtualstuff 758]

Thanks for the hint, Jeroen. I did miss that one... ;-)

 

Is there really a need to discredit the OP? Your answer comes across really arrogant. If you are Mr. Super-expert then come up with a better analysis.

 

Not arrogant your read things between the lines lol just based on facts ;-)

It's tested on one system local thus the results are based on that...

[AviatorMoser 353]

No, I think you're correct that this is not a one-fits-all analysis.

 

What I wanted the community to get out of this is a quantitative method in evaluating how tweaks can change the performance of their own systems. We see a lot of qualitative statements, but it's actually rare to see any of the raw data to back up such statements. All one needs is a simple combination of FRAPS, and a plotting program for the frame times CSV files FRAPS creates (http://sourceforge.net/projects/frafsbenchview/ or Excel, Matlab, etc.) Such a quantitative analysis is powerful because it can detect roughness in the rendering performance beyond what the naked eye can sometimes discern. It also removes subjectivity from performance evaluated from just your eyes alone.

[Penz 45]

Nice analysis!

 

I don't have a HT CPU so I can't really contribute, my i5 runs without any AM set.

 

My experience with very low FFTF values is that it may work if you're doing a circuit around an airport, but if you try flying low and fast over a larger area, say in a fighter jet, terrain texture will not have time to get properly rendered/processed, and you will get "blurries".

 

One thing I'm curious about is how you calculate the TEXTURE_BANDWIDTH_MULT value you use Daniel?

 

- JP

[virtualstuff 758]

No, I think you're correct that this is not a one-fits-all analysis.

 

What I wanted the community to get out of this is a quantitative method in evaluating how tweaks can change the performance of their own systems. We see a lot of qualitative statements, but it's actually rare to see any of the raw data to back up such statements. All one needs is a simple combination of FRAPS, and a plotting program for the frame times CSV files FRAPS creates (http://sourceforge.net/projects/frafsbenchview/ or Excel, Matlab, etc.) Such a quantitative analysis is powerful because it can detect roughness in the rendering performance beyond what the naked eye can sometimes discern. It also removes subjectivity from performance evaluated from just your eyes alone.

 

I agree it's always better to do thorough test and analyse instead of emotions, and thank you for your effort Daniel I applaud you for that :-)

But not everybody has the skills to do it that way, that's why I mend the skills part? 

Interesting would be if Lockheed would build in a extra tool to measure exactly the above...

[S_D 7]

I agree it's always better to do thorough test and analyse instead of emotions, and thank you for your effort Daniel I applaud you for that :-)

But not everybody has the skills to do it that way, that's why I mend the skills part? 

Interesting would be if Lockheed would build in a extra tool to measure exactly the above...

 

Bingo! A tool would be perfect. A kind of 'benchmark' mode that allows pre-programmed looped tests where you can setup 'loop 1 has AM, loop 2 doesn't', different resolutions etc etc etc. That would be amazing, or maybe something for a 3rd party to knock up specifically for the FS community.

[ryanbatc 12,851]

You're welcome, everyone.

 

I'm quite impressed by the Fiber Frame Time Fraction = 0.01 tweak. After reading it shouldn't make a difference for multi-core machines, I had disregarded it. But now I see the results firsthand. It does certainly make a quantifiable difference. Here I can now push the internal frame lock to 35 FPS while keeping LOD_Radius at Max.

 

AF = 85, 35 FPS Upper Limit, HT on, Overclock to 4.7 GHz, FFTF = 0.01

 

 

 

Doesn't your scenery blur up at FFTF .01?  I tried this in FSX for fun and anything over 100 kts my scenery will blur and not ever load.  Of course I'm only at 4.0GHz it 2500K.