[ODE] Future of physics processors and IM API

[David Black]

Erin Catto wrote:

>I've seen no evidence that Ageia's card is anything more than a daughter
>board with an off-the-shelf CPU and some local RAM. I don't think this is
>nearly as interesting as multi-core CPUs, such as in the Xbox 360.
>  
>
I tend to have the feeling that CPUs should basically be there to 
control  and feed data to other specialized processing elements. For 
example graphics, physics and environmental audio. CPUs are designed to 
execute general instruction sets, and for that there is a 
price/performance trade off.

I think ultimitly, to see they kind of performance benefits a GPU gives 
an application, vert specialized physics hardware will be needed. At 
least the constreaint solving problem appears to me to be ammeanable to 
really massive paralization,which I dont think a general purpose chip 
can best deliver.

>The problem with developing an open physics API is that physics algorithms
>are not nearly as well published and researched as graphics algorithms. It's
>not a solved problem, at least not in the public view. Given the current
>situation, I don't think it will happen.
>  
>
I think people are being a little pessimistic in this respect(though it 
is still a very strong point). From what I see, at least the major 
libraries, seem to be settling on a sparse mass matrix/state 
vector/jacobian/iterative solver kind of formulation.

The exact nature of this iterative solver  can presumably be left 
implimentation dependant(be it gauss seidel, conjugate gradiate etc). 
Although providing guarantees dealing with numerical issues and poorly 
conditioned input seems very tricky area.

David

>Erin
>
>
>
>_______________________________________________
>ODE mailing list
>ODE at q12.org
>http://q12.org/mailman/listinfo/ode
>  
>