Havok FX Supports GPU-accelerated Physics

There was that zombie game, Possession, that mentioned that. I looked into it and the anims wee good, but not realtime. They used Endorphin to create a load of different anims and then picked the ones they likesd the look of. OF course it's very early days to consider realtime Endorphin algorithmic animations, but I don't think there's anything out there that uses it 'properly' yet. I imagine a 2 character fighter would be the first place to see it.

 

Fight Night: Round 3 at Sonys E3 Sony Press Conference, potentially had realtime Endorphin algorithmic animations or some similiar tech.

Quicklook bellow. Realtime demostration from E3 sony press conference gives a lot better picture.

http://www.gametrailers.com/player.php?id=5830&type=mov

 

Allot of animation stuff that the cpu used to try and do has been offloaded to the gpu, so maybe physics and eventually all animation will be handled by the gpu.

 

Did they? I actually read on some internet forum (this one) in some thread (this one) that it's tech which will be used in next-generation games. Havok seems to be pushing it for future PC games, it will probably (IMO) not be used in console games unless it's terribly useful (apparently you have to pay extra for it).

 

Then what'll be handling the graphics?! It's not like we've reached perfect graphics with GPU cycles to spare.

 

That is subjective.

 

Compromise

It will surely be design compromise. For cost of a little graphics, game can have much better physics and animation. Therefore, I think (hope) capable developer will try to do this on xbox360. Since most threads used (yet) on Xenon is 4 and shown trailers have unimpressive physics, GPU physics is important for real next gen look.

 

Wait a second... Are you saying that the tessalator in xenos can produce 1 extra vertex per 2 cycles and still have the 500kk limit? I always thought that the tesslator would limit the throughput of vertices to the setup engine by half if used in all cycles... Is this true or the tessalator can actually give an extra of 250kk vertex?

 

The problem I have with this argument is what are the CPU's supposed to do? It's not like XeCPU is incapable tot. It's got a lot of performance to power the games, physics, animation, procedural synthesis and AI and stuff. Now if the GPU is doing rendering, phsysics and animation, that leaves the CPU doing very little. Of course if the animation can be rendered somehow on GPU more effectively than on the CPU it kinda makes sense, and obviously you balance resources of the entire system to get the effect you want. But really, if the GPU is to handle everything related to visuals and physics, that seems to defeat the point of the CPU's in my mind.

 

Theoretically the tesselator unit can provide 250 MPoly/sec, but I don't think that would be in addition to the 500 MPoly/sec hardware limit... rather towards, as the setup limit still remains. Just means that if used there may be less work that they unified pipelines would have to deal with for geometry. Tesselation is helpful for LOD and should be very helpful for displacement mapping (so you can draw in extra geometry detail into objects). Should be interesting how all of that turns out... XENOS is a geometry monster.

Currently there are no XBox360 games that use the tesselator unit yet as far as I know.

 

Or perhaps, CPU, since I believe tesselation doesn't usually occur in shaders (because they can't, for example, create vertices). It wouldn't spare the vertex/unified shaders their "normal" role.

 

No, no, sorry for my bad english but I didnt mean adding 250kk polys to the 500kk limit.
What i thought is that the tessalator would give a 250kk limit to xenos, meaning that when it tessalates the setup engine perfomance drops by half.

And i asked if the statement above is true, or if Xenos can do all the tessalation it wants and still "achieve" the 500kk setup limit.
Because if so, from the roughtly 24 Gflops needed to transform 500 millions vertx only half of that would be needed since the rest of it would come from the tessalator... And that would mean that from utilizing the tessalator as much as you can you would gain a roughtly P4 3.0 Ghz (12 Gflops if i'm not mistaken) just to do for exemple physics on Gpu...