Complex aircraft, what to sacrifice to increase performance?

[Noel 4,487]

You're hoping 'proper driver support' changes how complex aircraft will perform in P3D, is this part of what you are saying?  That would be a very pleasant surprise, emphasis on surprise.

 

You've twisted my comment too, which was not 'starting from square one'.  It was: 

 

The software needs an overhaul, that's the main issue [in order to allow complex aircraft to run well].  My hope is that when LM reworks P3D for 64-bit they continue working on the rendering engine using DirectX 12, exploiting more of what Dynamic Parallelism offers, and also rework how multicore/HT systems can be better exploited for components in the simulation amenable to this.  When this happens SLI might offer more for P3D than I believe it current can with respect to complex aircraft.

 

The overhaul I mention is, again, 64-bit & DX12 optimization, and perhaps better multicore utilization beyond texture loading.  You also seem to be stating there is no benefit w/ having for example AI run on separate threads, because of the synchronization need?  Every time I bring this up Rob you retort w/ the same patent dismissal.  You seem to be saying on no uncertain terms, this comment from LR is total nonsense:

 

If you have 20 processors, then you can run TWENTY AI PLANES WITH BASICALLY ZERO FRAME-RATE HIT. Crank the number of planes up to 20 in X-Plane 10 and watch what happens to the frame-rate. Try it now: Set the number of planes to 1 and look at the frame-rate. Then set it to 20 and look again. See the hit? That is because all of those flight models are running on ONE CPU, one after the other, in order. With X-Plane 10, each flight model can run on it’s own CPU, all at the same time… if you have 20 CPU’s, running 20 planes is no slower than running 1.

 

Marketing hype you're saying?  Or some magic not possible w/ ESP?

[Guest]

 

 

Marketing hype you're saying?

 

1.  Are you aware that AI aircraft flight models do run on separate threads in P3D (and FSX SP2)?  

2.  I can't predict how you define "overhaul", but that word to me, represents "starting from square one".

3.  Where did you determine that threading is only related to texture loading - I think you mean "terrain texture loading" not all texture loading?

4.  So what happens with 300 AI planes in XP10?

 

Does LR have access to the P3D source code to make such a comment?  Or is he speculating?  And what product is he specifically referencing, FSX, FS9, FS8, P3D V2.4?

 

The flight models running on a separate thread isn't anything "new" and it doesn't resolve the need to synchronize them so they position correctly in the 3D world.  If it did, then XP10 could only have 3 AI aircraft for a quad core processor ... and as we know, that is NOT the case.

 

Cheers, Rob.

[Noel 4,487]

Nope, not aware of AI models in separate threads, but I am also not aware of most of the intricacies of programming P3D.   My comments to the OP started w/ statements about how complex aircraft performance seems only modestly affected by changing various sliders in P3D V2, and proposed general reasons as to why.  I then conjectured on what avenues might be available to improve the picture, so that complex aircraft might run well in complex scenery, which I maintain now is poorly achievable on current hardware.   The overhaul was defined as 64-bit and optimizing for DX12 essentially.

 

With P3D in its current state, what is the upside for SLI in 'complex aircraft' given an 'optimized driver'?   I TO from default KLAX in the CS777 and show the main thread pegged at 100%, the other 10 handling terrain texture loading etc, but my Titan is lugging along under clocked and lower temperature w/ a reported utilization of say 45%, and a dismal frame rate of let's say 20.  If I now add SLI w/ a 2nd Titan, what happens?   Hard to imagine how a driver revision alone changes this picture.   Dynamic Parallelism looks like a mechanism to help shift the CPU/GPU balance, but it also seems like you need to revise the core engine to accommodate this, so driver revision alone wouldn't cut it, or so it seems.

[Guest]

Driver "upside" is removing work the CPU is doing that the GPU should be doing and overall efficiency of processing GPU tasks.  Driver updates have provided significant increases in performance with many other nVidia "key" applications in the past ... as much as 70% in some cases.  They've also corrected graphical issues and provided better AA options.  Look at nVidia driver history, they have broken P3D numerous times with WHQL drivers ... frame "buck" stops with nVidia for those using nVidia GPUs.

 

Getting nVidia to no longer ignore P3D I think is a good thing regardless of what it brings to the performance table.

 

I don't think anyone (well actually I'm sure there are a few that would not be interested) interested in a P3D future would disagree with 64bit and DX12 support ... it would be a product that would grow along side existing 32bit product ... but either way, it will be key to get nVidia providing driver support.

 

Cheers, Rob.

[Noel 4,487]

Sounds very helpful, DirectX 12 optimization thru collaboration w/ NVIDIA in 64-bit P3D.  

 

What do you know about how having VAS for current practical purposes unlimited in 64-bit might impact coding strategy?   What is the upside, beyond simply not running out of VAS, in having 64-bit AND lots of physical ram?  Intuitively, it would seem to create more opportunity to cache more and in doing so perhaps smooth out some elements of the sim.  

[Guest]

1.  Free to allocate

2.  OS level memory fragmentation will be less of a concern

3.  GC can deal with de-allocation "in time" so it less critical for force de-allocation which actually slows overall processing down, believe it or not, maximizing VAS usage is going to be "slower" ... in fact, I think that's what the FSX:SE folks are seeing, better VAS usage but lower FPS (althought I'd like to see Pete's results on FSX vs FSX:SE)

4.  I don't think Cache scenery in RAM is going to be that much of a benefit ... caching textures in VRAM will be a big benefit but 64bit doesn't really come into play here

5.  LM will need to deal with the current tessellation limitation so as to allow higher LOD Radius ... significant change but probably good for a 64bit product

6.  Open the door WIDE for 3rd party content providers ... and I do mean WIDE!

 

Fluidity will most likely come from driver support and not 64bit.

 

My 2 cents.

 

Cheers, Rob.

[Noel 4,487]

caching textures in VRAM will be a big benefit

 

Do you understand what happened between P3D 2.0 and 2.1 or 2.2 that caused reported VRAM utilization to change in such a big way?  I remember reported VRAM utilization would slowly uptick while flying thru complex scenery all the way to close to the max of my Titan's 6Gb.   Since I think it was 2.2 VRAM never goes over maybe 2.2 or so and I never heard the reason why.  Didn't seem to hurt anything performance-wise, but it was interesting to see such a big difference.  I think it had something to do with improving the hit you got w/ vegetation, or something along those lines.  I remember when 2.0 happened I was thinking it was a good thing I had all of that VRAM, now it seems it's not very relevant for me w/ my one 1920x1200 display.

[Guest]

 

 

Do you understand what happened between P3D 2.0 and 2.1 or 2.2 that caused reported VRAM utilization to change in such a big way?  I remember reported VRAM utilization would slowly uptick while flying thru complex scenery all the way to close to the max of my Titan's 6Gb.   Since I think it was 2.2 VRAM never goes over maybe 2.2

 

Not sure what you mean, I've gone above 4GB VRAM usage in v2.4 ... see here:

 

[GSalden 4,656]

I use the 2D cockpit (iFLY 737).

When running the Ifly 737 in 2D on 1 screen I always have 45+ fps.

And then I have a 3 monitor desktop, so 2 are showing the Windows desktop...

[Noel 4,487]

Not sure what you mean, I've gone above 4GB VRAM usage in v2.4 ... see here:

 

Ahh, so maybe this is a function of 2.2 which is what I'm using currently, it is absolutely the case, and very different from prior versions.  Sounds like maybe they discovered this and changed it for what whatever reason in 2.3 or 2.4.