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

[SteveW 1,792]

Thanks for the kind words! Applause to all those giving it a go.

[Guest JustanotherPilot]

A very helpful read on P3D3 settings, I too find I'm spending more time trying to get it running smoothly with stuttering and low fps,than actually flying. I continually return to FSX Steam to remember what a complete flight is like.

 

If it's not breaching any of your secrets, cold you provide details of the software you used and the process to attain your results. I would like to experiment with my system because honestly I'm ready to bin P3D31.

 

My system comprises of an i7 4790K and GTX680 GPU - not high end but no slouch either, But I'm only getting about 15-20fps sometimes down to 8fps with frames locked at 30fps. I was running AF until recently when I tried without and HT on. Definite improvement but still crap framerates.

 

I would be very interested in experimenting using your methodology.

[glider1 90]

P3D is that hard to tune properly because of the lack of proper 1/2 VSYNC I've decided to completely drop all addon scenery from P3D31 and just use it as a dedicated aircraft simulator on default scenery and properly working VSYNC with frame rates > monitor refresh. P3D just doesn't cut it with addon scenery compared to FSXSE on 1/2 VSYNC no matter how you look at it unless you have a 30 Hz monitor or GSYNC but even then I'm not confident.

 

I have tried AF116 and AF85 and AF84 on my i76700k OC4.7 system and all of them end up stuttering after a while. I think the best way to go on default scenery is probably AF84 because you get slightly higher frames and texture loading isn't an issue plus you have a core free for addons.

 

One possible way forward that I haven't seen talked about. Is it possible to run the main sim core at a really high clock multiplier and the other cores at a much lower multiplier? I'm thinking that might be a nice optimization considering there is so much load on just one core. Maybe the fact that only one core is hot means you can run it at a higher clock (EDIT: because the heat is distributed away to the cooler cores)

[Guest JustanotherPilot]

P3D is that hard to tune properly because of the lack of proper 1/2 VSYNC I've decided to completely drop all addon scenery from P3D31 and just use it as a dedicated aircraft simulator on default scenery and properly working VSYNC with frame rates > monitor refresh. P3D just doesn't cut it with addon scenery compared to FSXSE on 1/2 VSYNC no matter how you look at it unless you have a 30 Hz monitor or GSYNC but even then I'm not confident.

 

I have tried AF116 and AF85 and AF84 on my i76700k OC4.7 system and all of them end up stuttering after a while. I think the best way to go on default scenery is probably AF84 because you get slightly higher frames and texture loading isn't an issue plus you have a core free for addons.

 

One possible way forward that I haven't seen talked about. Is it possible to run the main sim core at a really high clock multiplier and the other cores at a much lower multiplier? I'm thinking that might be a nice optimization considering there is so much load on just one core. Maybe the fact that only one core is hot means you can run it at a higher clock (EDIT: because the heat is distributed away to the cooler cores)

 

 

I might be inclined to follow your intentions if all else fails....or return to P3D25 which ran flawlessly with phenomenal fluidity no matter which addons where installed, I'm thinking along the lines that v3.1 is a step backwards, for me, anyways.

[SteveW 1,792]

I have tried AF116 and AF85 and AF84 on my i76700k OC4.7 system and all of them end up stuttering after a while. I think the best way to go on default scenery is probably AF84 because you get slightly higher frames and texture loading isn't an issue plus you have a core free for addons.

I disagree, 116=(01,11,01,00) which is just as 84=(01,01,01,00) is, also with one core free for addons. But 116 allows the sim to split into 4 jobs, moving the second job away from the UI which improves fps over 84 variants. 4 jobs is optimum at 4GHz. This has been proved several times by properly testing performance, not just fps. Fps can go high but can be sawtoothing like in Mbergsma Quote 252. Real performance can be said to be a measure of fps - fpsDelta (change in fps).

 

I would be very interested in experimenting using your methodology.

You can collect frame rate data with FSUIPC, and maybe some other programs, perhaps GPU testing software of some kind. You can calculate fps-delta from that trace, maybe use Excel with comma separated values to do it. I've built my own test software since FSX came out and included some of this technology with Ideal Flight 10. IF10 also measures fps-delta and outputs that trace too along with many other variables, like VAS used. Affinity masks can be used by the sim, other (addon) programs can be launched before the sim and affinity moved with Task Explorer, or assigned affinity by programs like Process Lasso, or IF10. Not sure how Process Lasso goes about setting application affinity, but IF10 creates a process space for the app to start rather than assign affinity afterwards. 4 core testing can be made with a 6 or 8 core by turning off cores in the MB setup. Graphs can be made of comma separated values with Excel, or other spreadsheet apps.

[Cruachan 705]

Hi Steve,

 

I couldn't help noticing the mention of Ideal Flight 10 (IF 10) coming up in your posts. I've researched this product and there seems little doubt that I should be looking to purchase. It's feature set is huge and, quite clearly, has been coded by someone or a team of developers who really do know what they are doing. However, I am unclear as to whether IF10/Pro is fully compatible with P3D v3.1. Can you please clarify this for me, please?

 

Thanks.

Regards,

Mike

[SteveW 1,792]

We can chat more over on the codelegend forum Mike, or by PM. But briefly, the P3D v3.1 version is available in beta from the forum, it's a fully working product.

[Cruachan 705]

Many thanks, Steve

 

Regards,

Mike

[331BK 96]

Hey Steve,

 

i have a question or more a thought  

 

What when the Core for the maintread has for example 4,7 Ghz and the 3 other cores have 4,5 Ghz. 

 

In this case the main schedule schould be "faster" then the terrain treads. Do you think that can help with performance ? 

 

brgds

[Guest]

Thought I'd share some testing I did with my 5960X with cores On/Off and HT On/Off ... for P3D, I've switched to 5C/10T using AM 340.  Seems to work well with TrackIR, and very high graphics settings ... can't say this will be my final resting place but thought I'd share.  

 

These tests are using SiSoftware Sandra 2016 exclusively, nothing as extensive as Steve's FPS testing ... some of the numbers are interesting.

 

 

Cheers, Rob.

[Cruachan 705]

Hi Rob,

 

Great stuff! However, I'm feeling a tad confused....just a tad

 

You say you are favouring 5C/10T with HT=ON for P3D yet an Affinity Mask of 340, according to my calculator, appears to indicate that 4 physical cores are being used: 0101010100

 

Whereas, Inverse Binary would produce: 0110101010 giving 5 x LPs in use.

 

Also, how did you apply the Affinity Mask while running tests using SiSoftware Sandra?

 

I assume that I may have misunderstood something here. Perhaps you or Steve would be kind enough to clarify.

 

Regards,

Mike

[Guest]

 

 

You say you are favouring 5C/10T with HT=ON for P3D yet an Affinity Mask of 340, according to my calculator, appears to indicate that 4 physical cores are being used: 0101010100

 

Correct, 4 physical cores being used for P3D, 1 core free for other stuff (OS, ShadowPlay recordings, TrackIR, etc.).  But I didn't say I was favoring it, I'm testing it out and haven't really come to a verdict yet in regards to P3D.

 

 

 

Also, how did you apply the Affinity Mask while running tests using SiSoftware Sandra?

 

I don't, I enable/disable cores and turn on/off HT via EFI/BIOS before each test.

 

The numbers I found somewhat interesting were "File I/O access time", "File I/O read", "Inter-Core latency", and "Inter-core Bandwidth" ... the other numbers are pretty much what I expected.  I have not (on purpose) made any connection to how these numbers relate to P3D operation.

 

With that said, I have discovered and odd correlation between certain "view" in P3D and the Affinity value.  For example 5C/10T AM=340 when using Instant replay in spot view is horrendous, jitters and jumping all over the place.  8C no AM and instant replay is much smoother in spot view.  It's almost as if the Instant Replay system isn't using the same threading model ... in normal flight both settings (8C or 5C/10T) are very smooth.

 

Cheers, Rob.

[Cruachan 705]

Correct, 4 physical cores being used for P3D, 1 core free for other stuff (OS, ShadowPlay recordings, TrackIR, etc.). But I didn't say I was favoring it, I'm testing it out and haven't really come to a verdict yet in regards to P3D.

Okay, my apologies.

 

On the face of it those figures appear to confirm that for general use 8C/16T delivers best overall performance from the 5960X which, presumably, is the way Intel engineered/intended it to behave. Bench testing with RealBench bears this out and my scores are higher with 8C/16T than with 8C, although not by a huge amount. 8C, of course, has the advantage of lower temps under over-clocked conditions. My CPU (water-cooled) has not yet exceeded 45C while running for extended periods at 4.4GHz (100x44) so, currently, I couldn't be more happy.

 

I followed your lead in another thread a few weeks ago and by doing so confirmed 8C along with VSync ON, Triple buffering, Unlimited frame rates and no AM provided the best performance overall in terms of smoothness and, yes, frame rates in Prepar3D v3.1. I know this runs contrary to current thinking but, on my rig, the evidence is there. I have not tested by viewing Instant Replay in Spot view using 5C/10T, but I can say that with 8C and Tower View Instant Replay performance is very good. Indeed, I often take pleasure in sitting back while watching the last few minutes of a flight as I turn to base leg and then finals

 

My settings are reasonably high and I do use SGSSAAx4 in NI along with MSAAx4 in P3D and with thick cloud cover (REX4 TD Soft Clouds) it doesn't seem to hurt too much.

 

Edit: I should perhaps say that I have not yet tested performance while using TrackIR or ShadowPlay.

 

Cheers, Mike

[Slayer 190]

Do you guys get windows reactivation nags when enabling/disabling HT or cores in BIOS? I ask because I did in Win7 but haven't tried yet on 10.

[Cruachan 705]

Do you guys get windows reactivation nags when enabling/disabling HT or cores in BIOS? I ask because I did in Win7 but haven't tried yet on 10.

Hi Steve,

 

No, and I have enabled/disabled HT several times in the UEFI BIOS over the past few weeks. I have not touched the cores, however.

 

Mike