Best BeamNG Hardware

So first of all I want to congratulate the BeamNG team for doing a really thorough job in developing the new physics. I mean, just look at this quote from Gabester:

WOW .......... So 40fps with around 15,000 beams on a mid-high end system!?!?! Which makes me wonder...... how far could you scale the physics if you had a really good system. For instance, in this video we see a 2 Ford Taurus' made from 28,400 elements each colliding in a (non-real-time) simulation by the NCAC:

I know 56,800 elements in many times more than 15,000; but imagine the level of detail achievable if you had the right hardware and experience to make extremely detailed models. It would literally be indistinguishable from an actual car crash. Which leads me to my main question: What sort of processing configuration is ideal for your physics? I am very excited to build models with these new physics and would like to just make sure I have made the right decision. Like: Are large core counts vs. clock speed preferred; or how much L1-3 cache is ideal and whether Hyperthreading makes a difference. No pressure though, I am just very excited about the possibilities with this new system, thanks!

 

I would set minimum requirements at:
Dual Core Intel processor or equivalent AMD @ 3.0 GHz
Nvidia 8800GTS or Radeon HD 3400
4GB RAM
4GB HDD space
An Internet Connection

 

You have to remember that BeamNG isnt built for that kind of simulation. Even if you were able to build something that complex chances are you would never be able to make it stable, let alone get a realistic simulation out of it. Weight would also be a huge issue, each node has a minimum weight before it becomes unstable, using that many nodes in a single model could cause many problems.

BeamNG is built for a specific type of simulation, at the level of detail (or close to) that Gabe has been building them, and it does that REALLY well. However something like this requires a whole different type of simulation, a different way of handeling physics and material properties.

With that said though, id absolutely love to see someone try something like this in BeamNG, you never know till you try right? Or until TDev states his opinion on the matter at least

EDIT: Also you have to think about the "elements" in that simulation vs. the node beam approach that BeamNG has. An element in that simulation is essentially a square, to create a square in BeamNG you need at LEAST 5 beams, and quite a few more if you want to hold it up (IE. a hood made essentially like a grid in BeamNG cannot hold itself up, so you need quite a few more beams to do so).

So lets assume 3 beams for each square or "element" then 2 to hold it in place, 5 total (3 and 2 because squares next to each other share beams so you dont need an entire set for each new square) So 28,400 X 5 = 142,000 beams... And that is at minimum... It would probably take quite a lot more then that to actually create any sort of stable structure that doesnt wiggle like jello.

Thats of coarse not in any way an exact number, but it at least gives you an idea.

 

For all you who want to have a crack at running these simulations;

http://www.ncac.gwu.edu/vml/models.html

There are some models. I haven't been able to find some software for little to no money, mind you. There's still half the research to do. But you can always find a way to acquire LS-Dyna solver, or borrow someone's machine. The cool thing is, if your PC is quick enough you can run the simulation with two ford sedans in a day or so. If you process the physics on a GPU, you can run one into a wall at about 5-11fps. I encourage you all to have some fun.

 

I wasn't trying to say that we should have a ridiculously high number of elements simulated - that would not only be extremely difficult to get right and offer no to little benefit to the end user anyway. What I was trying to ask is how far the physics can be scaled because even, say a 2x improvement in beaming would be great. Oh, and to ask what hardware would be best suited because I am in the process of upgrading. Weight would definitely be an issue though, but I am not that concerned about it as I was:

And about the square vs. beam thing, there was a discussion about this sort of thing/ automatic generation on the RoR forums a while back. One of the most interesting things with being able to have high beam counts is you could "feed" the simulation a unibody structure (and select the ASTM grade and thickness of steel) and then it could dynamically generate tet's to fit the structure and - presto - you would have a 95% accurate model with no beaming work having to be done (of course, you would still have to make the suspension and other rotating/mechanical parts). But who knows what could happen - I just wish for the team the best

 

That would be quite amazing. Id love to see that actually set up and working. It would be crazy cool to be able to design and test structures that way.