[ODE] hardware physics chip

[John Miles]

> I'm just waiting to hear about a new DirectPhysics library from MS now -)

I'd be surprised if that were too far behind. :-)

> Seriously though, the chip will have serious market penetration issues if
> they actually encoded the Novodex codebase into the chip.
> Physics is one of
> those things that often has to be tailored on a per-game basis, so while
> this would provide a minimalistic solution - it would not provide the
> open-ended access needed to go beyond the "look-ma, lots of simulating
> objects" first-pass google-fest.

I disagree strongly.  It is true that physics *parameters* need to be
tailored on a per-game basis, and probably always will, but a
properly-designed physics API is the most generic, black-boxable thing in
the world.  It is possible to design a physics API that makes everyone
happy... *if* you agree from the outset that certain aspects are going to
have to taken out of the application's hands if hardware acceleration is
going to work.

For instance, micromanagement of collision behavior via application
callbacks is probably not a Good Thing if you want accelerator hardware to
perform without being hamstrung by communication with the host CPU.

I've used Novodex, and found it to be a very solid API.  Nobody who's using
ODE is going to have a serious beef with it.  I have not used Havok... but
if I were Havok, I'd be sweating bullets right now.  Novodex is a hell of a
nice dynamics package even without hardware acceleration, and it's a lot
cheaper than Havok's offering.

> Not to mention locking out any progress
> made on new solution techniques etc.

There hasn't been much in the way of cool new hotness in the dynamics
business since Principia Mathematica came out.  It's all been implementation
details since then.

Besides, GPUs haven't exactly held back progress in the graphics world, have
they?  If a need emerges for the physics equivalent of a pixel shader, it
will happen.  Dynamics is much easier to fit into a fixed-function pipeline
than rendering, so I'll be surprised if this proves to be a legitimate
complaint going forward.

> I'm assuming as well that, given the
> boards have 128MB of ram, that when they talk physics - they mean both
> simulation and collision.  Opcode is a step-penetration collision system,
> while some games may need to do sweep tests for some objects.   Looking
> forward to seeing some real information on the chip as so far, I've only
> seen the PR propaganda -)

Agreed, collision is the scary part.

-- john

>