Odd performance behaviour around wheels

FPS on the top corner with FRAPS.

-WITH WHEELS:

ON:


OFF:


-NO WHEELS:

ON:


OFF:

 

Without the wheels I get between 75-80fps with physics off or on it makes no difference..

Ok I'll test this later when I get on my PC

 

I think he's asking with all wheels.

 

Best would be all 4 cases:
with wheels with physics
with wheels without physics
without wheels with physics
without wheels without physics

 

HUD overlay is disabled

Physics ON, no wheels: CPU at ~7%


Physics OFF, no wheels: CPU at ~18%


Physics ON, with wheels: CPU ~ 24%


Physics OFF, with wheels: CPU ~ 10%


...so no big difference FPS wise if physics are enabled or not

Hope this is what you wanted us to do

 

-Wheels + Physics: 25 fps

-Wheels + NO Physics: 97 fps

-NO Wheels + Physics: 103 fps

-NO Wheels + NO Physics: 103 fps

Disabling physics seems to bump up the fps a lot.

 

Wheels + Physics on = 18.4 FPS


Wheels + Physics off = 68.8 FPS



No Wheels + Physics on = 73.4 FPS



No Wheels + Physics off = 74.1 FPS



So there is a massive Difference with wheels and no wheels
And here is my processor if that helps
AMD Phenom II X4 965 Processor 3.40GHz

 

I can tell you categorically that there's absolutely no difference with the physics disabled between wheels on or off. 120+ fps with wheels, same fps again without. Again, when the physics are disabled. This lag is in the physics somewhere.

 

The lag is due to the amount of nodes on the vehicle. The Moonhawk (which some people have trouble running) with all wheels has around 700 nodes while the T75 in its current state with all wheels has over 1200, with over 600 of the nodes coming from the wheels.

 

Its also the Amount of wheels on the T75 Lots of wheels are constantly moving Requiring more CPU power The T75 Has 10 wheels

 

Just a thing, the T75 is a 10 wheel vehicle and at 66 nodes per wheel we are looking at 660 nodes for the wheels. Incidentally the amount of nodes in the T75 without the wheels is 660. If you don't believe me, have a look



So having the wheels on should be an equivalent impact to having a second T75 without wheels, only with more beams. Absolutely not, my fps goes from 90 to 50. The T75 having more beams and collision surfaces being a vastly more complex object than it's wheels. However I did get the physics slowdown, so I wonder if there's a second spin off issue to do with the way collision detection works.

 

Not a fair comparison. 1 T75 gets its own thread. With wheels you have that 1 thread simulating the 1320 nodes on 1 core, without the wheels then the 1 thread simulates 660 wheels on 1 core. 2 T75's with no wheels will be 2 seperate threads, each on their own core. 660 nodes on 1 core, 660 on another core, not even nearly the same as simulating 1320 on 1 core.


May try dropping the number of rays down to 12 to see what difference that makes to the lag. May not work as nicely though due to larger size of T75 wheels. Maybe even trying to do the rear wheels as singles.

 

The trouble is that a single vehicle can currently only run on a single core/thread. If it were possible to split the load of the vehicle and its wheels, it would run better on many processors. The T75 relies on very powerful cores, it's just that most CPUs don't have cores powerful enough to competently run it.

 

If I'm correct, the game runs one vehicle per core. When you spawn in a second T75 without wheels there are 2 cores handling the 660 nodes each. With one T75 there are over 1200 nodes running on one core. Hence the lag.

 

per thread. Then windows tries to spread the threads over the cores where needed. Multiple threads can share a core. Multiple cores cant share a thread.

 

Alright. I see.

 

Then thats probably it, the lack of "sharing" is that wall which would explain why the problem isn't experienced by everyone. It could very well be that my processor can't handle a vehicle more complex than 660+264(for the 4 wheels from one set of back wheels)= roughly 900 nodes. But at the same time... there are computationally cheaper rigid body solutions that might be something you can implement and build a realistic tyre simulation off this basis. Because whats going on right now is frankly stupid. You can forgive a vehicle body being this expensive to run computationally but when you devote the same amount of power to simulating just the wheels also there's probably a need to look at other ways if only as something you can put in an options menu for "simplified tire model" so lower end users simulate the vehicles properly with an R factor style wheel model rolling these node beam chassis along. Its significant enough to be a non-crazy suggestion when 50% of my simulation is just the wheels.

Or get vehicles to share threads. At least allow wheels to be split off into their own thread

 

Could you assign the tire simulation to a separate thread? Would that possibly improve performance?

 

Just having the D-series in the grid map I have about 110 fps. When I have one T75 the fps is around 20, and spawning 2 additional T75 it only drops to around 18. If I have one T75 and remove all wheels I get 120 fps.