Game News Sacred 2 to use some cool physics SDK

[Psilon]

Hmm, it'll probably be a few years or so before we reach the point where that's realistically attainable in a game. It would be fucking awesome, though.

It's already in one of the Novodex physics demos. No defenders scattering in terror, but there's about two dozen catapults flinging projectiles (and a few ragdoll bodies) at a stone tower. Quite slick.

 

[triCritical]

I am too lazy to read, but I would curious to find out what is soooo special about a physics processor. Honestly I think its a lot of hype. Being a physicist who essentially is trained in modeling and simulation, ie simulating and modeling physics on a computer, I find the most used operation is straight up arithmetic. Mainly because integrals and differential equations for the most part are solved numerically. Henceforth, whats wrong with the the x86 arithmetic CPU?

Now for operations heavily involving graphics, we have a GPU, makes sense. For operations requiring extensive DSP algorithms, there are Digital Signal Processors, which make things like convolution a breeze. I think this is mostly hype to sell to video game producers, who don't know a fucking thing about physics. Which is probably why most of academia still uses workstations and PC's.

 

[Psilon]

As I understand it, the Ageia PPU is designed to accelerate the collision detection phase, which is a drag on the CPU. Collision detection's O(n^2) in the worst case, and is a pain to compute once you switch from simple primitives (spheres, capsules, axis-aligned bounding boxes) to arbitrary triangle meshes.

The spring, fluid, and rigid body dynamics accelerations are nice, but really it's the collision queries I'm interested in.

 

[Spazmo]

Most of mechanics basically amounts to F = ma. I think having a whole expensive card to do that is kind of silly.

 

[triCritical]

As I understand it, the Ageia PPU is designed to accelerate the collision detection phase, which is a drag on the CPU. Collision detection's O(n^2) in the worst case, and is a pain to compute once you switch from simple primitives (spheres, capsules, axis-aligned bounding boxes) to arbitrary triangle meshes.

The spring, fluid, and rigid body dynamics accelerations are nice, but really it's the collision queries I'm interested in.

In other words, the physics engine strong point is to determine at what point to polygons intersect? That is not even physics, nor for the most part can it be determined as a deterministic system, especially between two characters.

I guess it would be applicable in the FPS realm, but those games already have adequate physics engines, especially in the case of Half Life 2.

I guess my new point is, that physics engines of right now are already very optimized and very believable. Hence, I can't see any real use for something like this when the fidelity of the simulation, ie. the game, is very very very low.

However, I am too lazy to read about their processor, and somebody told me about this about 3-4 months ago, and blew it aside then. Either I am jealous, or I just don't think its all that.

 

[truekaiser]

Do you actually need to buy one of these PhysX things to get it to work with S2?

Yes. Hopefully they will also include a software based physics in the game like Havok so that we won't need to buy that useless PhysX

no you do not actualy need to buy the ppu to use their physics. the sdk is designed so that people who do not have the chip can still have some physics but due to the limitations of software physics it will pale in comparasian to having the ppu.

 

[Sarvis]

I am too lazy to read, but I would curious to find out what is soooo special about a physics processor. Honestly I think its a lot of hype. Being a physicist who essentially is trained in modeling and simulation, ie simulating and modeling physics on a computer, I find the most used operation is straight up arithmetic. Mainly because integrals and differential equations for the most part are solved numerically. Henceforth, whats wrong with the the x86 arithmetic CPU?

Now for operations heavily involving graphics, we have a GPU, makes sense. For operations requiring extensive DSP algorithms, there are Digital Signal Processors, which make things like convolution a breeze. I think this is mostly hype to sell to video game producers, who don't know a fucking thing about physics. Which is probably why most of academia still uses workstations and PC's.

Well, keep in mind that everything you just said would apply to graphics too. Why calculate graphics on a seperate chip when the math co-processor would do just fine?

This is just basically throwing another processor into your system, which is dedicated to a specific task. That plus it comes with an SDK to make the physics stuff easy, and probably some hardware optimizations and such that I'd never be able to understand.


Is this all needed? Probably not. Then again, graphics cards probably weren't either, but once they existed developers found uses for them.

 

[Psilon]

In other words, the physics engine strong point is to determine at what point to polygons intersect? That is not even physics, nor for the most part can it be determined as a deterministic system, especially between two characters.

Two polygons is easy, yes. Two polygon meshes is hard, especially when their movements aren't easily predictable or you don't have convex hulls/bounding boxes available. Multiply by all the non-crate actors in a scene, and things get even more fun.That's why people are interested in accelerating the collision queries. Our group doesn't have anyone to do all the 3ds Max triangle decimation and bounding-box-fitting, so anything that lets us be lazy artists and still get 200 FPS is a Good Thing.

Still, there's always the software Novodex and Tokamak APIs, both of which are free for academic/noncommercial use. And, as I said earlier, Novodex/Ageia is still a pretty sweet library.

 

[bryce777]

I never got the big deal about a physics engine. There is like one equation for kinematics. You assign each object in the game a mass and each collision a speed and there you go...maybe I have gone crazy but 'real' physics are not that hard and I have modelled them myself and it's a snap so why do people get so bent out of shape???

 

[DarkUnderlord]

Will it let you throw rats? 'Cause we all know how much more real the role-playing is when you can throw rats!

 

[Sarvis]

Will it let you throw rats? 'Cause we all know how much more real the role-playing is when you can throw rats!

Nah, you guys have pretty firmly established that more options and flexibility are bad. It's not roleplaying unless it's on rails!

 

[Naked_Lunch]

It was a silly example, you have to admit.

 

[Sarvis]

It was a silly example, you have to admit.

No, not really.

It was something that happened in a PnP game. It illustrates quite well the lack of flexibility of computer games by comparison.

Was it a silly thing to DO? Sure, but it was pretty funny at the time.

 

[triCritical]

In other words, the physics engine strong point is to determine at what point to polygons intersect? That is not even physics, nor for the most part can it be determined as a deterministic system, especially between two characters.

Two polygons is easy, yes. Two polygon meshes is hard, especially when their movements aren't easily predictable or you don't have convex hulls/bounding boxes available. Multiply by all the non-crate actors in a scene, and things get even more fun.That's why people are interested in accelerating the collision queries. Our group doesn't have anyone to do all the 3ds Max triangle decimation and bounding-box-fitting, so anything that lets us be lazy artists and still get 200 FPS is a Good Thing.

Still, there's always the software Novodex and Tokamak APIs, both of which are free for academic/noncommercial use. And, as I said earlier, Novodex/Ageia is still a pretty sweet library.

I never said it was easy. I was forced to do 2d interpolation on meshes and it was really friggin hard and that is not ever3d. However, my point was that its not physics rather geometry.

And tokamak are those plasma torus'. They use simple non interacting particle collisions? I find that hard to believe.

 

[Psilon]

No, I'm not talking about the plasma torus sense of "tokamak." I'm talking about the Tokamak library.

 

[triCritical]

No, I'm not talking about the plasma torus sense of "tokamak." I'm talking about the Tokamak library.

Anyhow, it seems what you are interested in is a physics engine. And that is all fine and good. But my original point is that all physics calculations nearly always boil down to arithmetic. Hence, what does this new PPU offer, in hardware, that can't be done in an optimized way with the x86 architecture. From what you are describing the answer is really a GPU, which some cards like 3Dlabs might already be well suited to accelerate. Since they allow for a higher level of fidelity and offer more programability.

I am not sure what internally a collision query does. Probably a sophisticated algorithm for detecting a collision, but how does new hardware optimize this, and at the same time utilize a physics built into the hardware. Again, I have always been too lazy to read about this, but it sounds like a scam.

 

[Psilon]

Yeah, I agree that implementing the dynamics part of a physics engine is really easy. Collision detection is the big research area. The main reason I like this news is that I've got several relatively big codebases right now that use Novodex/Ageia tech and the licensing fees Ascaron's forking out should ensure that I don't have to switch engines around in the near future.

 

[DarkSign]

Collision detection and edge detection for that matter has always been a bitch.

There are several physics integrations out there:
Havok, Novodex, Newton,Tolamak, ODE (suckiest of them all).

triCritical, remember when Intel came out with its Pentium with MMP? basically logic just for multimedia? It boils down to specific parallel processing. An additional chip with extra processing algors =ok but An additional chip withOUT extra processing algors = better.

We're going with Newton. The cost is right and its already been integrated successfully into Torque before.

 

[triCritical]

Collision detection and edge detection for that matter has always been a bitch.

There are several physics integrations out there:
Havok, Novodex, Newton,Tolamak, ODE (suckiest of them all).

triCritical, remember when Intel came out with its Pentium with MMP? basically logic just for multimedia? It boils down to specific parallel processing. An additional chip with extra processing algors =ok but An additional chip withOUT extra processing algors = better.

We're going with Newton. The cost is right and its already been integrated successfully into Torque before.

I am pretty sure that MMX and all the other x86 architecture features are areas of the die that are completely unrelated the the math coprocessor. Could be wrong. And with dual-cores this means twice as many arithmetic units.

I can see where edge detection is a bitch especially with very irregular poly's, but two things to be considered?

1) Isn't this still a geometry problem?
2) Where talking games, here what level of fidelity are you guys after. There should be a very acceptable tolerance, especially with our frame rates, that doing this in a very approximate way and still not tell the difference.

Case and point, walking in Morrowind, something that will be far more complex then the simplistic Sacred isometric walking . Is there crude collision detection, which leaves, probably slightly unacceptable clipping levels for are day and age, really that bad. Especially if we expect good improvements in the future. And won't the limiting feature be the animation as well. It really to me, without really knowing what goes on down deep sounds like a graphics engine, GPU, geometry problem.