I always thought that high autogen is the biggest fps killer. However when flying over those big cities like Paris (OpenLC EU) and Seattle (Orbx PNW) I did some quick experiments and I found increasing autogen from none to extremely dense only makes maybe 2fps difference.

 

Road traffic appears to be the biggest single fps killer. Increasing it from 0% to 15% can eat easily more than 5 if not 10 fps over Paris. Shadows are bloody expensive on fps too. With trees and building shadows on, increasing shadow quality from low to medium can cut as much as 10 fps if not more.

 

I'm curious whether anyone has done some systematic experiments to show how different graphic settings affect fps differently. I'm surprised that autogen is so light on performance in P3D, as I remember that in FSX autogen is considered to the biggest fps killer. So it seems true that P3D really has optimized autogen by instancing etc as they mentioned before.

 

 

I remember that in FSX autogen is considered to the biggest fps killer

 

I also find I can run with high autogen settings in P3D with little FPS penalty.

 

Biggest FPS killer for me in P3D V2.4 is still volumetric fog. I have this ticked off now.

AI air traffic, road traffic, urban autogen and volumetric fog all are major FPS killers.

well what I found out specifically is that uban autogen is not as heavy as we thought in P3D. As I said, over those dense urban area moving autogen slider from none all the way to extreme only makes couple fps difference.

 

I agree that traffic and shadows seem to be the biggest killers. I fly default weather and haven't found volumetric fog makes any differences.

I always thought that high autogen is the biggest fps killer. However when flying over those big cities like Paris (OpenLC EU) and Seattle (Orbx PNW) I did some quick experiments and I found increasing autogen from none to extremely dense only makes maybe 2fps difference.

 

Road traffic appears to be the biggest single fps killer. Increasing it from 0% to 15% can eat easily more than 5 if not 10 fps over Paris. Shadows are bloody expensive on fps too. With trees and building shadows on, increasing shadow quality from low to medium can cut as much as 10 fps if not more.

 

I'm curious whether anyone has done some systematic experiments to show how different graphic settings affect fps differently. I'm surprised that autogen is so light on performance in P3D, as I remember that in FSX autogen is considered to the biggest fps killer. So it seems true that P3D really has optimized autogen by instancing etc as they mentioned before.

 

I can be sitting at a medium gate at KJFK in the QW757, FTXG 1.2, be looking at the side of an airport terminal in light clouds, and see a frame rate of 20, unlocked, on a SB-E hexacore at 4.44Ghz, GTX Titan.   The direct view forward is very simple:  remote horizon w/ almost no visible autogen.   Just the terminal, which really looks simple to render as graphics tasks go.  And yet, the main thread CPU core is pegged at 100%...AND...the sensor says my GPU is only running at 52% of peak.  Even its core clock is down throttled the work load is so light.   That is a bad problem IMO, to leave that much resource poorly used.  I hope LM looks at whatever it is that is doing that because that type of programming could conceivably be sprinkled here and there affecting performance significantly.   Get those tasks offloaded to the GPU!

Very interesting, Donor. Have you brought this to the LM team on their forum?

Yes I have mentioned this in several posts early on in P3D 2's history, but alas they don't seem particularly responsive.  Of note, in that same scenario I just described I can dial back scenery and autogen sliders, take tessellation from ultra to zero, remove building shadows and there is hardly a difference in displayed frame rate.  There apparently is something about those legacy default buildings which their approach to rendering misses in a big way.  

For me its Scenery complexity and Autogen if I turn those to none in a big city I get 120 FPS, if I turn them to max I get 25-30. I'm not sure why they hurt FPS so bad they are just simple models. I use Orbx regions I wonder if they kill FPS so bad because the models were made for FSX and DX9?

My computer is 5 years old, but it is an i7 Quad Core at 3gig speed with a GTX 660 PCI Express graphics card, so it does pretty good. I have FTX Global, FTX Vector, and most of the regional sceneries. My worst place is Seattle, Washington. I take off from KSEA on Runway 34 R in the Manfred Jahn C-117 and climb up to 1500 feet and settle in at 150 knots low and slow and aim for downtown. My frame rate is set at unlimited and during the take off I get 30 to 40 FPS. As soon as I get up ion the air it starts to slide back. With most of my scenery sliders pretty far right and road traffic set at 25% I get 10 to 15 FPS at 1,500 feet. With road traffic set to 0 I get 15 to 20 FPS, so I agree that Road Traffic is the biggest hog. I find that nothing else seems to matter much, not even shadows. I do have my Scenery Objects slider all the way to Extremely Dense as I find that is where I see the most detail in the scenery and set any lower than that, I do not see the red loading cranes at the big ship basin at the south end of the city. Moving all the other sliders up and down doesn't seem to make more than 2 to 3 FPS difference, but Road Traffic does. Moving Ship and Leisure Boat Traffic to 50% doesn't do an harm. Maybe someone can come up with an add-on Road Traffic program that doesn't kill the sim?

Just a nod in agreement about road traffic being huge frame rate factor. But there are other areas that are killers, especially Seattle. Some urban areas are very FPS friendly, but others are not.

Just a nod in agreement about road traffic being huge frame rate factor. But there are other areas that are killers, especially Seattle. Some urban areas are very FPS friendly, but others are not.

 

I think one difference could be the amount of vector data. For example, with ORBX Vector some urban areas have vector roads a lot. I've noticed that if you keep all the smaller roads enabled, they can have huge impact of FPS even without any traffic turned on. This might be one explanation, so that the autogen buildings aren't actually the main culprit.

For me its Scenery complexity and Autogen if I turn those to none in a big city I get 120 FPS, if I turn them to max I get 25-30. I'm not sure why they hurt FPS so bad they are just simple models. I use Orbx regions I wonder if they kill FPS so bad because the models were made for FSX and DX9?

I concur. The old default scenery objects seem to have a disproportionate impact on frames given the number of objects.

If you want to see what is possible with great programming try Aerosoft's Thessaloniki scenery. I can run that with everything turned full blast and it runs over 25fps. There are hundreds (thousands? lol) of buildings in sight at any time and it is still butter smooth.

Get those tasks offloaded to the GPU!

 

 

LM has recognized that the main CPU thread is often the limiting factor for performance on high end systems and I would assume that if there was a simple solution, they would have made those code changes already. What we are seeing is a generally limitation of parallel processing on an Intel CPU. All threads must be synchronized and one thread in essence, has to keep track of that synchrony. Don't feel too bad. I have 12 physical cores and P3d uses all of them. But from my reading of what LM has posted regarding the degree of parallelism in P3d, most of those threads are loading scenery and that keeps them only marginally busy. As a result I rarely see more than 60% CPU usage rates.The same is true for the GPU, although with certain IQ settings, I can overwhelm my GTX 680 and get the GPU usage up to 99%.

 

I believe that I stated the following in a thread a few months ago. At some point, buying faster and faster GPUs is not going to improve frame rates, just because one CPU thread will be holding everything back. The only good news embedded in this is that with a faster GPU, higher resolutions like 4k are plausible options.

 

According to MS, DX12 (native to Windows 10) is a lot less  of a burden on the CPU then even DX11, but that will probably have to wait until P3d 3.0.

I have the exact same settings in P3D V2.4 as I did in V2.3 and I'm getting a drop of 10fps and stuttering. Same computer, same hardware. Although adding to  the P3D cfg is not recommended by those who are smarter than me, I think some serious tweaks need doing. I can now only use P3D v2.4 for VFR flying in light aircraft - not the Axe A320 IFR flights I could do in P3D v2.3.