How many cores is BeamNG optimised for?

Recently, I went from a Fx-6100 6 core CPU, which is really just a 3 core with hyper-threading, To an Fx-8320. The Fx-8320 Is 4 actual physical cores with hyper-threading ( yes thats intel's technology but its the jargon I use.), and I've noticed a huge jump in Performance due to the extra core. Its also Overclocked to 4.8Ghz but thats not as much of a difference as physically adding in a core.

So, my quesiton overall is, how many cores does BeamNG scale to. If i were to get a 10 core Intel, would BeamNG even be able to use them, or where would the limit be?

 

Currently the game uses 1 thread per vehicle. If you search on the forum you'll find that AMD's older HT equiv (CMT?) gets along very poorly with BeamNG. IIRC disabling that or setting an affinity for each thread to the physical cores rather than the fake/virtual ones can help on AMD.

Basically on a reasonably capable processor which isn't hobbled by AMD's old CMT thing you shouldn't see any noteworthy performance changes as long as you have less vehicles spawned than CPU cores. Realistically you can actually have many more vehicles spawned than that with a fast CPU, Intel HyperThreading does help in certain situations. Look at bananabench numbers to get an idea if you want.

I don't know how things look on Ryzen, but normally for BeamNG the advice is to get an Intel CPU - almost any Intel CPU.

 

I think you should compare single core performance of those CPUs, lot of boost probably comes from that. It then depends a lot from which map you use, how many vehicles, what kind of conditions etc.

Skylake i3 and i7 are pretty much the same with 1 vehicle, because their single core performance is pretty much the same and i3 can run 4 threads, even it is dual core, difference comes mostly with more vehicles or more complex vehicles.

With BeamNG you would benefit from ability to run 4 threads well, that is with single vehicle, graphical stuff takes 1 core, physics takes 2 and then there is something that runs on 4th that is not so easy to identify, maybe collision stuff?

So, if it is so that HT equivalent of AMD is not really up to task in BeamNG, then that 3 core CPU might got some hard time of course, but I think effect of that one core less is not too great with single vanilla vehicle on vanilla maps.

 

I never realized that it was one vehicle per core. Sure seems like my FPS takes a hit when I have 4 vehicles, even on my 4-core, 4-thread i5-6600k. Must be a graphics thing.

 

That is interesting. IN MY OLD TESTING I don't experience a massive dip with the fourth car, but the fifth car on my i7. I believe this is due to the rest of the game (map or UI for example) requiring some processing. With my i7s hyperthreading it's able to do all of the background tasks and run four cars. Your i5 needs to run the rest of the game code on one of the four threads and that means you can't run the four cars well.

Here is a benchmark I did a while back: https://www.beamng.com/threads/performance-scaling-4-cars.21880/
UPDATED BENCH: https://www.beamng.com/threads/my-updated-system-benchmark.38206/
6 cars run on 4c/8t very well

 

Graphics, collisions etc. also needs one core, so it is number of cores -1 car actually, but depending from cars and intensity of graphics + collision on specific map you might get more.

CPU load varies a lot from map to map so it is completely impossible to make claims of whole game. Of course vanilla maps are rather consistent.

 

My laptop struggles with more than 3 but can run 2 okay, the UI really slows down though, even on grid small pure on lowest settings.
I7-6500U with integrated graphics

 

So waiting for that new version and working banana bench, sounds to be quite a improvement


There is some tests about multiple vehicles, but I guess banana bench will be better for valid testing:
https://www.beamng.com/threads/beam...l-ryzen-run-the-game-well-enough.36807/page-3

4 core 8 threads, so 7 x T75 for me and 8th starts slowing things down a lot, for maps that have more CPU load to GFX thread it might be only 6, but no more than 7.

For simpler vehicles maybe bit more, but I don't see really huge difference between D15 and T75.

Now then how much slowing down we accept before calling it can't run is of course one thing to yet be defined, but I'm looking when dropping below 60fps as limit.

For me it's GFX stuff that CPU does which limits mostly and that depends hugely from map and even from location of the map. Vanilla maps generally are quite easy for CPU.

 

For me, I can get up to about 14 cars before seeing a FPS drop, but according to the in game monitoring, it is because of the fact that I have maxed out my graphics card. (further backed up by the fact that when I hit J my FPS does nothing and also my computer is only running at about 20% processor load on the task manager.) My game crashes at about 16 cars... seems to just overload my graphics card.

my computer is running dual hexacore intel processors with hyperthreading at 3.2GHz though, so I don't envision any real problems with keeping up physics wise.

 

It depends really a lot about map and graphics settings. I'm using settings from performance testing thread (link in sig) except 720p here, CPU has not much slowdown because of physics, but that darn GFX thread is causing issue of GPU load starting to decrease and FPS starts to slow down, ECA is quite hard on CPU maybe hardest of all vanilla maps, Hirochi would probably not have issue likewise many other maps won't have this level of CPU GFX load, that is however what limits my FPS mostly. Now, put those 14 cars on map and tell AI to flee, it's a bit chore, but I'm sure resulting chaos will be rewarding




Update: Physics paused, BeamNG set to normal window, which is minimized and that thread is still going crazy, when window is brought up increase in CPU usage on some cores can be observed, so I'm not really sure if it is GFX thread or some collision or something else, what I know is that at ECA that limits number of vehicles more than vehicle physics at least on my system and GPU is rarely limiting number of vehicles on my system/settings, but of course with only 7 vehicles or less that is half of less of polys atv_123 runs for example.

Chased those 4 AI a little, data polling rate 1000ms, so not all spiking showing and not sure if this adds much, but 0 is paused, then chasing a bit, as you can see GPU load increases as one and then second AI car gets left behind, at end no AI car on sight. 5th AI car would surely be possible but with 7 total vehicles 60fps might not be possible to reach with those settings at least.

 

Well on BananaBench it used to say I could run over 11 cars... (I don't remeber the actual number) BananaBench only tests how many cars can run on the processor which allow the 2000 Hz simulation speed. My thing is that I want to run the game near 60 FPS. After one car per physical core on my i7 the FPS drops from around 55 FPS to near 20 FPS. This is generally not playable to me, although all the physics are correctly operating in "real time".

 

I can get 8 cars with a i7-4790

But then a good fps for me is 25.

 

5 per core? Am I simply not understanding something here or is that some insane optimization? Is it April fools again?

 

He is refering to the old BeamNG benchmark, which determines how many vehicles can run on your processor can run at "real time". "Real time" means that all of the physics are able to be calculated for the physics engine to work correctly. As he has an extremely fast and overclocked processor he is able to run many cars before the processor overloads. This does not reflect FPS at all though, as the game can be running at real time at very low FPS numbers.

 

My 4790K doesn't see lag until I hit 9 cars typically. 7 T-Series' finds me lagging down to 40 fps, but not much worse than that.

 

24

 

No BeamNG crashes at around 14-15 cars no matter what, so 24 is way overkill.

 

For me 60fps tends to be lowest enjoyable as if FPS is lower by CPU limit things get choppy and usually my graphics settings are such that my GPU can run at 60fps.

For amusement I took 7 vehicles to ECA, surely my FPS stays above 30fps but it was really difficult to drive even at 60fps as there was kind of choppiness, so usually I try to stay away from near 100% core load, FPS is quite nice considering how much calculation must be done:


Start is paused, only one core has high usage.


At 720p even 1050 Ti can do pretty much anything at 60fps or more. FPS and GPU usage gets higher as less AI cars around. With single car I'm GPU limited, but whenever that one CPU thread (gray in above pic) gets high usage my GPU usage falls down.


No hope to really run more cars than that with those settings on that map, but then again I don't usually need to run much more than 1 or 2.