multi core usage qeustion

first of all.. please excuse me for any spelling mistakes english is not my first language
and i'm typing this at 1:21 am...
so first of all..
i've noticed with beamng drive each task uses 1 cpu core.
the game engine use 1
each car gets its own seperate core.
and so on.
is there any plans in the future to enhance this to multi threaded work?
for example.
i got a 24 core system that gets amazing multi core performance.
but its single core performance lacks a bit. most of the time alot of these cores gets wasted in this game.
or is there perhaps a way to improve the single core performance?
thanks in advance!

 

I'm not sure if they will move away from single core, but there was mentioning about more optimizations to come, which might help single core need.

It is mainly graphics, lua and such that uses single core.

Physics were change some versions ago so that if you spawn 100 physics objects load gets shared better, also I think it should do some stacking too, but might remember wrong.

Greatest limit with 24 core will be from graphics though, each different car takes piece from that one core pool and it does create the limit, only way to get more space there is to avoid LUA heavy cars and use graphics settings that ease the load, so no reflections, ssao or light rays and hide ui, also for me orbit camera robs fps compared to free camera.

Also browsers with several tabs seem to rob piece of single core pool as effect to fps is quite clear, when I had windows defender running, that took quite bit of fps too. I can easily get 10% increase in fps due to less load to single core when not running some software while playing BeamNG. Even GPU manufacturers fan adjusting software took quite chunk of fps.

I guess it has also something to do with how well or poorly windows chooses which core/thread software code is run at, sometime there are threads not used at all, I guess that spiking nature of CPU load that Beam has to single core is confusing it a bit, so clearing any running software seems to help fps.


Then there is increase of clock speed, that does help quite a bit sometimes, but 10-15% at maximum so if you get 32Mbeams/s with 1 car in banana bench, with overclocking that might increase to 35, with not running any other software and clearing all background processes to minimum you might get another 10-15% so maybe even up to 38 or so.

For me clearing software was from 35Mbeams/s or so to 40Mbeams/s, that included getting rid of Defender.

Also memory speed does have an effect, so running faster memory can help, depends from system a bit I guess.

With 8th gen intel turbo drops quite severely when more cores are loaded and it seems something always loads more cores so getting fastest or 2nd fastest turbo clocks just is not happening, not sure if you have such turbo related clock slowdowns, but if you have such turbo decrease, then tuning that might give some improvement.

Not sure how much that all helps, but maybe gives something to try.

 

Wow. And I thought 100 was the limit. I wonder how many cones could be spawned...

 

I 100% agree that the limit in a system like this is the graphics... even with a GTX 1070 in my system the graphics are still easily the bottleneck. That plus the fact that my system is an older Mac Pro, and you start to realize that I don't even think my motherboard is capable of keeping up with the data transfer speeds that the 1070 can do. This causes another bottleneck that basically caps the performance of my system at only about 70 to 80% graphics card usage. Granted Sixsixsevenseven warned me about that a long time ago... and I headed his advice, so I didn't buy one... but later when you end up getting one for about $50... it's hard to say no.

So I have an overpowered graphics card in an overpowered system that doesn't have the data transfer rates in the motherboard to be able to handle either of them.

That being said though... while it struggles to get over 100fps... you can keep it above 60fps all the way till like... 20 something cars.... which, you have to admit, is not bad.

 

It is quite a lot. I have had for example my flying wing vehicles 100 flying at 60fps at gridmap, with something like overclocked Ryzen 7, one could have lots more. Vehicles without textures, that use only solid color are such that you can spawn probably even more than cones, as long as node count is about same.

I think that I did put few hundred cones to autocross map as a test and sure enough it did not crash, but my poor old CPU was not quite up to a task, still got something like 20fps I think, there was something between 200-500 cones, can't remember so precisely anymore.

Those are really nice improvements though, one can make already some simple physics enabled objects for maps, etc.

All these limits are constantly moving further away with new updates, so it is really nice evolution we see here.

 

it would be quite intresting to think about in maybe 3/4 years that optimization has gone so far that we dont need dual 1080ti's to get reflections to max

 

I doubt that you can have reflections at max even if you had dual 2080Ti's at the moment