Did you run benchmarks to compare which sim racing game is more taxing on the cpu? And why are you linking the taxing on the cpu with physics calculations only and not count optimizations, or lack of optimal optimizations. You can't really benchmark the cpu taxing between games only for physics, because you can't separate the space that the graphics fill in the cpu work, from the physics.
Btw are you comparing the way tires are build in ac and rf2 or are you comparing the cpu performance while driving the cars on track? Because I heard that it takes several hours for the software to calculate-build the tire model in rf2.
I had an access to a suitably old dual core 2.0+ GHz system. The idea to try and run the most recent car sims on it came to my mind more as a joke at first. What I didn't really expect is to see AC feel at home on that rig. It ran pretty nice even on good enough graphics settings, let alone the lowest ones. Of course, I'm not talking about full grid AI races and all that. Just pure offline hotlapping.
Whatever settings I would choose for either rF2 or PC, they ran so sluggishly it was practically impossible to control the cars. Do I really need benchmarks to tell a very good performance from almost nonexistent one?
Of course you could counter this by saying AC is written in Assembler (which I bet is not the case) and is perfectly optimized, but let's be realistic here. If it would be possible to optimize these things that good, professional engineering packages would do their modelling in run time, not waste minutes of computing time for a single frame.
Sorry, I didn't quite get the question in your second paragraph, but yes, it takes enormous amount of time to make a tire for rF2, obviously. Not to mention that one has to have an actual tire -- or better yet, a set of those -- in their possession to measure the parameters needed. And this is what I call a proper modelling.
This, however, puts a huge question mark on the possibility of underfunded projects to get anywhere near what could be called a 'fidelity'.
If you want to go that way, BeamNG is the same thing, a bunch of nodes physically simulated, and has nothing remotely approaching realistic tire physics... the gross properties are there (kinda springy, turn by steering) but the fine details aren't. I would rather a correct lookup table or Pacejka curve to some complicated model with no guarantees it's in touch with reality. At least the Pacejka curve has real-world data to work from.
BeamNG and RoR are rather strange projects. It's really weird how they manage to perform that poorly when it comes to portraying rigid bodies interaction. I mean, you'd expect steel rods to behave like steel rods, yet they are pretty much rubbery. This product simply doesn't take itself seriously from the get go. And one can mess up even the best of concepts by simply not being serious enough.
Pacejka curves can deal with as precisely measured data as possible. The problem is, it's not enough by today's standards. As I said, Niels Heusinkveld makes the aged isiMotor engine do wonders. Push the car to and over the limit, however, and the suspension of disbelief starts to fade a little...
Yes, the newest brush models are not perfect yet. But since the beginning of this year they improved a lot. So much in fact, I start thinking the plausible simulation is somewhere around the corner.