Why GFX thread has very high CPU usage at full screen?

On some maps effect is stronger than other maps.

Altitude is like this:


East Coast USA highway spawn is like this:


Update speed of graph was set to slow, because my monitor takes close to 10 seconds to switch from full screen to desktop.

Seems to happening more with mod maps than official maps, I'm just bit surprised it goes that way around, would of thought windowed being more of problem.

No Vsync or FPS limiter at windowed mode, not logical explanation as I made sure settings are perfectly same, except resolution in windowed mode is very tiny bit smaller because border takes some space and desktop resolution is 1920x1080 which I used for game resolution too.

There really is not much difference in FPS either, but on some mod maps windowed mode gives better framerate as GFX thread becomes too heavy for CPU.

Might be also happening only with specific settings too, I just noticed this behavior, my settings:


With Glass view mountains, it is completely other way around, physics thread does weird things though at full screen, FPS remains same around 64FPS


Update: Maybe it is Nvidia driver, Win 8 feature or Torque 3D feature, or something else, haven't got a clue really, or maybe it is just how things work / related way map is made? It could also be Task Manager being weird.

 

Hirochi raceway seems to also get this high GFX thread thing when using full screen, no effect to framerate though, but not much more load and it would, I don't understand why fullscreen can be so heavy to run on some maps compared to windowed and on other maps it is windowed mode that seems to have higher load.

 

It doesn't show up in framerate though, is it like that by design? GPU load remains at 100% also, which would indicate things running at same pace, but do they?

It would be easy explanation, but I'm afraid nothing is easy which I bump into

I did black out part where I did other testing so that it is not confusing this further, you can see GPU load, GFX thread CPU load and FPS in this graph, it is from Hirochi raceway, what do you think, could it still be syncing to desktop?

 

Could well be that, especially more complex part. Some maps Windowed mode is hammering CPU and full screen is not, so of course it can't be very simple that is going on, whatever it is. Based on experimenting it looks like that depends from the map if running windowed or full screen is giving lower CPU load, if CPU load can be trusted to show really what load there is.

 

Yeah, there is lot of variables, what I can tell is that from 4 cores / 8 threads, 1 runs stuff that is related to GFX, then at least 2 runs something related to physics, there are tiny loads that gets almost to 0 when I press J, so I guess those are physics related too, then there is one which I have no idea, but I guess might be UI related or then UI runs on main GFX thread, each start of BeamNG likes to rearrange them, but unlike Core 2 Duo this Skylake seems not to switch cores at all after that, Core 2 Duo did core swapping all the time (didn't run beam with it though), which I think was related to thermal balancing, maybe Skylake don't need that, who knows.

Update: Bios versions seem to affect my system at least, F21 bios is not giving more than 3.7Ghz and Turbo boost is not really working, it stays at 3.4Ghz unless I use mobo vendor software to get CPU to 3.7Ghz constantly.

F20 Bios allows 4Ghz of turbo clock and it does not go below 3.7Ghz, also it holds near 4Ghz when only single core is working.

F5 Bios allows 4Ghz, but only when CPU is at idle, with any load, even to single core, it allows only 3.7Ghz, but does not go below that. As a bonus that Bios allows overclocking and also it does core swapping a lot and very fast, which is a lot less with F20 and non existent with F21 bios.
However F5 is not doing that much core swapping when overclocked and kept at constant frequency.

None of bios versions seems to do real core parking though, tends to do some load on many cores when running BeamNG.

With F21 it is easy to tell which core does what in BeamNG, but with F5 and to some extent with F20, it is fairly challenging.

 

This is what makes things confusing, on some maps CPU load is high when windowed, but at full screen CPU load is less, however GPU load and FPS indicates still CPU limit, could it be as simple as CPU load reporting part of Windows malfunctioning?


Red line indicates point which from onwards driving was at same locations of map with pretty much same camera angles. Data collection frequency is 1000ms, CPU graph is Max Thread usage.

Probably I'm only one who finds this interesting, there is something going on that escapes logical reason, why on some maps CPU load appears to be more on full screen and on some maps on windowed mode?

CPU frequency shows how at full screen more cores are working as frequency is lower, 4Ghz with 1 core working and 3.7Ghz when 4 cores are working, add Hyper Threading to top of that to bring more confusion.

Low FPS and heavy CPU use on this map is because visual distance is quite long and Tennessee is very big map, but what is puzzling is why on some maps, like Tennessee, CPU load is higher on windowed mode instead of full screen mode, which would be more logical according to Darren's explanation about Vsync effects.