I have found novodex is the most accurate physically compared to all other physics libraries I've used. Newton is the only other that can be considered acceptable. (I have not used Havok or Vortex)
But keep in mind that unless you are actually trying to simulate the real world these inaccuracies are not too much of a problem.
Novodex actually copes quite well with large numbers of objects. It really depends on how your objects are interacting though.

Novodex is available on linux, and I assume other platforms. Its my understanding that you need to pay for this privilige though.

Novodex is superior to all 3 free libraries (except possible accuracy wise compared to Newton; i have yet to see convincing figures either way).

The only real problem with Novodex is that it's not free for commercial use. Also, only Windows is free; you have to pay for the other platforms.

Quote:Original post by Ljilekor
what about ode? (open dynamics engine) http://www.ode.org

How does it compare with other dynamic engines?

We've been playing with different physics solutions at work; we used Karma (Renderware Physics) for one of our released titles, ODE for some commercial simulation work one of our guys did, and Novodex for an FPS pitch prototype.

Karma was (at least back in 2003) pretty badly integrated into RW, and the tools were quite flakey. There's no Linux version, which made running it against our Linux-based servers... interesting... I hear they've improved it's integration and tools, though. Apart from that, stability was rock-solid, and it's fast. Like all RW libs, it's really primarily optimized against the PS2, which is nice if that's where you're going.

ODE is the nicest of the 'free' versions.. it'll run on pretty much anything (one guy was porting it to fixed-point WinCE; God alone knows how that will work with numerical stability issues, but it shows how portable the core codebase is). With the new solver it's pretty fast, and the user community is extremely good (it's worth subscribing to the mailing list just for background info on the how's and why's of building a physics engine). Tools are so-so (but Juice is a pretty neat little app), and most commercial applications that use it end up having to re-write the trimesh collision system.

From looking into Novodex, it seems rock-solid, and fast as blazes. I've only tried it on Win32, though, so I can't speak for the console libs. I love the tools (both the in-house ones, and what Ratcliff has done with Novodex and Rocket). If I was going to use a rigid-body physics system for a commercial project right now, I'd probably go with Novodex again. I've also encouraged the students I mentor to look into using it for their projects.

Hope that answers your question;

Allan

Quote:Original post by Ljilekor
thx odin.
So Novodex' trimesh support is seamless?

Yupp. The guy who wrote what became ODE's default trimesh system (OpCode), Pierre Terdiman, is one of the key guys behind Novodex. From what I understand, Novodex is using an optimized and improved codebase derived from OpCode. OpCode, btw, remains one of the best plug-and-play arbitary polygon cloud collision systems; it's just the bindings and contact generation between it and ODE that is a bit flakey.

One of the things Novodex does really well is convex hulls; normally Trimesh-Trimesh is a real pain (so unstable it's not even funny). Novodex, and the work Ratcliff did for Rocket, allows the programmer to turn arbitary polygon-meshes into simplified convex hulls, and for the collision between the hull and everything in the world (including C.H. vs TriMesh, and C.H. vs C.H.). As such you get close to Tri-Tri functionality at significantly better performance and stability.

I had hours of fun playing with the LUA script-files for Rocket, especially Siege. There's just something fundamentally disturbed about lobbing ragdolls from catapults to break down a giant Jenga tower :)

Allan