Ageia physics engine - pro's and con's?

Rikan · Post by **Rikan** » Wed Apr 12, 2006 7:23 am

Since this one doesn't concern the Irrlicht engine per se but may apply to other engines (or frameworks) as well, I've started this thread right here and not somewhere closer related to Irrlicht.

I think the thread title says it all - what are the advantages and drawbacks of Ageia compared to other physics engines, mainly ODE or Newton?

Main concerns are

Speed of simulation (base load and with lots of moving/colliding objects)
Ease of use/integration into a graphics engine (namely Irrlicht)
Realism/Accuracy (jittering, explosions, collision errors)
Licensing politics

The last one is the big question concerning Ageia. I didn's see anything about the terms of use of Ageia on their web page, and downloading the SDK requires registration. And that's going a bit too far for me without knowing what comes out at the end.

bitplane · Post by **bitplane** » Wed Apr 12, 2006 11:07 am

after a quick glance at the website I have decided that I wouldn't go near it.
they lack a price and clear licensing info, their software is used in physics hardware and in unreal 3, they are targetting maya/max/xsi users, and the site is written in management/investor-speak rather than techie-speak.
something tells me that this is aimed at the big-boys rather than hobbyists/indies, and probably won't be within my price range this millennium

hybrid · Post by **hybrid** » Wed Apr 12, 2006 11:18 am

The hardware thing might be a chance, though. If they really want to be successful in selling their physics acceleration boards and gfx card add-ons they will have to open their APIs. I think there are already some organizations discussing an opengl like physics API which would of course make physic engine integration a lot easier. And ageia seems a promising candidate here, just by pure commercial power resp. presence. If they get nVidia and ATI in there will be no way around ageia anymore.

xhrit · Post by **xhrit** » Wed Apr 12, 2006 5:49 pm

what are we traveling back in time now?

co-processors were a bad idea in the 80's - and they still are. Soon the entire computer will made out of coprocessors instead of a CPU!

And if it comes down to that, why not just use omni-generic multi-processors? What is the advantage ov having a graphics co-prosescor, math co-prosescor, physics co-prosescor, etc? Tailored chips may be better at specific functions, but are wasted when said functions are not in use and cost more to produce.

What will that Ageia chip be doing when I am compiling software or surfing the web?

Is Ageia the new Emotion Engine? That is to say, a hyped up marketing-campaign for a product that was invented for the sole purpose ov marketing it?

CNN : "The Sony Emotion Engine: Will PlayStation2 replace your PC?"
/Me : "No."

http://archives.cnn.com/2000/TECH/compu ... ngine.idg/

cassini · Post by **cassini** » Wed Apr 12, 2006 6:08 pm

It is sad but the truth is that there is not comparison between PhysX and any of the free engines.
Ageia is a multimillion-dollar company, and have hired every PHD that has something do with physics simulation, to added more features to Novodex.
This is how they have Soft body collision, fluid simulation, fractures, and many other things. They are using the big corporation strategy of buying anything in their way.

Ageia marketing methods are very questionable. Instead of competing at a commercial level with other commercial technologies like Havok, they decided to bully out the small developers, but releasing a “free for nom commercial use SDK”.
There is not way a small developer can keep up with a professional developer with dozens of PHDs thinking of new ideas to incorporate to one engine.
Since they have the monopoly of the best software and the hardware they will exterminate the effort of all small developers. It may be good for them but in the long run it will only kill the creativity of other people, as they gradually see not support for their effort.

Notice that Ageia have not created anything new, everything they have done has been copied from other people, including the so call PPU.
Here is example of a physics acceleration adapter:

http://www.clearspeed.com/

I am hopping that a company like Microsoft or the openGL ARB committee come up with a started programming language for programming these hardware accelerators, in a more generic way.
I believe that some how this will happens as video cards are becoming so powerful that they have the feasibility to deal with physics algorithms.
If a generic language was created and companies like Nvidia, Ati, Havok, CmLab, RenderWare, and all of the free developers also adopt it, then any company manufacturing hardware will have not other choice than to do the same or be alienated.

And in my opinion that is fear and good for every body, and not these Ageia monopolistic method when they own everything and every body is just a spectator.

Rikan · Post by **Rikan** » Wed Apr 12, 2006 6:11 pm

Maybe I'm wrong, but it looks to me that the physics engine makes use of the hardware mentioned if present - not that the engine itself is embedded into the hardware.

Doesn't matter. As for the targetted audience... First and foremost, everything is about advertising. So it's just clear that the first impression you've got is 'marketspeak'. As stated in their FAQ, the real juicy information comes once you've been registered.

But yes, that makes me suspicious as well. Needing to register first, waiting one to three days for an email before getting to the REAL information? I think I steer clear of Ageia and consider using something where I would know beforehand what I get.

Heh, how about registering just to ask why they are so tight-lipped about pricing and licensing?

xhrit · Post by **xhrit** » Wed Apr 12, 2006 7:08 pm

"Ageia physX is a rigid body dynamics engine that shares a geneology with ODE (both were largely written by ex-MathEngine employees) as well as many architectural similarities. However, physX is a much more mature engine, with a fully fleshed-out API, and contains several important features that ODE lacks."

translation : we are just like ode... but better!

Eternl Knight · Post by **Eternl Knight** » Thu Apr 13, 2006 1:29 am

we are just like ode... but better!

This should not be a big surprise.

Firstly, ODE is BSD licensed and as such can be used in this manner. Hell, that's one of it's greate benefits!!! The BSD license is what makes it so popular with game developers who cannot afford the license fees Havok & Co require, have someone halfway competent in physics & constraint-solving mathematics on their team, and who are not interested in including a hundred and one third-party DLL's.

Secondly, ODE uses proven methods to solve physics simulations. While there are tweaks & changes that can improve the speed of the system (via parallel code execution, contact caching, friction assumptions, etc); the fundamental algorithms used are the same used in your big-time (and small-time) engines. This is not an "Everything is a copy of ODE" statement, but a "There is nothing new under the sun" one. There are only so many ways one can calculate a physics simulation in real-time, and ODE implements one of the most popular and flexible methods.

Finally, real-time physics is not an incredibly large research field. As many have commented elsewhere - unlike graphics algorithms, real-time physics does not have alot of funding behind it. As such, there is a limited number of professionals with PHd's & significant academic time dedicated to the area. A company that could convince enough venture capitalists that the "physics card" is the next big thing (remember, there was a time when people scoffed at the 3D video card too) would have enough money to buy out the top dogs. After all, they're probably thrilled at the fact that there is money flowing into their research area.

As for my opinion on Ageia - I'm holding off on it. I already have a free engine that does not require me to remain uncommercial. Their site is (as someone already mentioned) more a marketting/sales site and, as such, has no real information I can use to determine if it's worth the effort. And finally - I'm waiting on the success to kick in. I once chose a single-vendor API (Glide) that is now dead. I'm not doing that again...

--EK

Tim · Post by **Tim** » Thu Apr 13, 2006 2:56 am

Eternl Knight wrote:Secondly, ODE uses proven methods to solve physics simulations. While there are tweaks & changes that can improve the speed of the system (via parallel code execution, contact caching, friction assumptions, etc); the fundamental algorithms used are the same used in your big-time (and small-time) engines. This is not an "Everything is a copy of ODE" statement, but a "There is nothing new under the sun" one. There are only so many ways one can calculate a physics simulation in real-time, and ODE implements one of the most popular and flexible methods.
--EK

So for you it is “like father like son”. What I like of the ODE fanatics is how they see every thing been ODE. Once again this time by using reverse psychology you did say that another engine in a ODE copy with shall we say some “tweak and changes” to improve speed.
You know your last statement remind me of movie Amadeus when the Emperor ask the advisers for their opinion about Mozart’ symphony. And the adviser replied that there were too many notes, the emperor said yes that is right “there are only some many notes that human can handle, take out a few notes”.
In your mind there are only some many way physics can be done and ODE cover all of them. So for you life start and end in ODE, I never stop learning things this totally amazing engine.
So there you go people ODE is the origin and the end of Physics, every thing else is just a copy with some twist and tricks.

Eternl Knight · Post by **Eternl Knight** » Thu Apr 13, 2006 5:19 am

Here we go again. Tim is back to beat the dead horse some more.

Look, before you start you little hissy fit arguments again - do some research. Point blank, there really is only a limited number of ways to run a real-time physics simulation when multiple objects are either in simultaneous contact and/or there are constraints involved on the physics bodies (such as joints, friction, etc). One of the most popular methods is via iteratively solving the Linear Complementary Problem (i.e. a large sparse matrix) of the simulatneous constraints using the Projected Gauss-Seidel method. There are others (such as using sequential impulses rather than proper LCP solving) however the limitations with these methods (numerical stability, jittering, speed, etc) tend to mean they are used in areas where their flaws can be easily hidden and/or ignored.

I did not say ODE is the father of all. Nor did I say that ODE covered all the way in which a physics engine can be implementedd. For fucks sake - you even quoted me on that, but in case your little tirade is getting in the way of your reading skills; here's the bit you seemed to miss... "ODE implements one of the most popular and flexible methods." Did you notice that? It says "one" not "the only".

I suppose you also missed the quote that reads "Ageia physX is a rigid body dynamics engine that shares a geneology with ODE (both were largely written by ex-MathEngine employees) as well as many architectural similarities". Well, gently caress me blind and call me Mary - it looks like the folks behind XSI (Softimage) are either in on the conspiracy or just might know what the hell is going on given their integration of PhysX into their new product. But hey, this is Tim's little world of paranoia. What has logic and supporting evidence have to do with it?

You know, I've tried to be patient with you. I initially tried helping you understand what I was talking about. But it appears you have a wood on for some other engine and just HAVE to malign ODE at every opportunity. To which I have to take a leaf out of Midnight's game book and tell you to bugger off. Come back when you have a clue what your talking about.

--EK

jam · Post by **jam** » Thu Apr 13, 2006 5:24 am

Rikan, I know you seem to have locked in on ODE and Newton, but I thought I might point you to two other physics engines

Tokamak link

and

True Axis link

In the least it gives you more options to choose from, which I always think is good thing

Tim · Post by **Tim** » Thu Apr 13, 2006 7:06 am

Wow what a reaction, did I made you mad or angry? Let us not forget you were the one mentioning the prase “Everything is a copy of ODE”
you use reverse psychology, good try on that one, but you did put the idea on every person that everything is a copy of ODE.

Eternl Knight wrote:Look, before you start you little hissy fit arguments again - do some research. Point blank, there really is only a limited number of ways to run a real-time physics simulation when multiple objects are either in simultaneous contact and/or there are constraints involved on the physics bodies (such as joints, friction, etc). One of the most popular methods is via iteratively solving the Linear Complementary Problem (i.e. a large sparse matrix) of the simulatneous constraints using the Projected Gauss-Seidel method. There are others (such as using sequential impulses rather than proper LCP solving) however the limitations with these methods (numerical stability, jittering, speed, etc) tend to mean they are used in areas where their flaws can be easily hidden and/or ignored.
--EK

How do you know those are the only method? Could they be different ones, don't you think it would be more accurate saying that those are the only methods known by you and the ODE fanatics.

For all I know no other technology is saying that, or are you saying that Havoc is a copy of ODE too? Havoc predates ODE for over three years.

Eternl Knight · Post by **Eternl Knight** » Thu Apr 13, 2006 7:43 am

...did I made you mad or angry?

The sentence is "...did I make you mad or angry", but hey - you make everything else up - why not your own grammar?

Again, I reiterate (for the dummies, yes that means you Tim) "Where did I say everything is a copy of ODE?". I explicitly state the reverse, but hell you didn't bother with the facts last time you got a hankering to harrass me, and nothing has changed.

For the record - I have never stated that ODE predates Havok (note the spelling dimwit. It's a "K" not a "C"). Although, even there you are incorrect. Havok was unveiled at the GDC of 2000. First version of ODE was released in 2001. By my mathematics - that's ONE year - not three. But again, why concern yourself with the facts?

I never stated there was ONE method either. I simply stated (and you have yet to prove me wrong) that the use of Projected Gauss-Seidel for solving the multiple constraint issue in realtime dynamics engines is "one of the most popular".

How do I know these things? I'm an engineer. I studied these things at university. Not only that, but I also (to this day) pursue research on the topic. I read papers, look at code, and discuss issues with people who do this for a living. You know, the kind of things one does when you want to know something about a topic. But then again, this is Tim talking - we can just make things up as we go along. Must be a nice world you live in, where one has no need of knowledge in a subject area to make definitive statements about it.

"Reverse psychology"... Definition: "Reverse Psychology is the term that describes the outcome where advocacy of one course of action persuades someone to do the opposite."

In other words Tim, you are saying that because I am stating that ODE is NOT the be all and end all of dynamics simulators - I am making people believe that it IS. Riiight... You know I also tell people the Earth is round because I want them to believe it's flat. Yup, the world according to Tim "the cowardly Guest poster" is more beyond repair than any Lewis Carrol book ever written folks. Step right up and smoke the drugs!

As I said before - come back when you have some FACTS in that vacant noggin of yours. Until then, I reserve the right to call it as I see it. That "Tim" hasn't a friggin clue what he is talking about and cannot even spend five minutes getting his facts straight before attacking supporters of something he clearly dislikes.

--EK

Rikan · Post by **Rikan** » Thu Apr 13, 2006 8:54 am

Whoa, hold the reins, please.

I never intended to start a holy war about the origins of each physics engine or company politics.

I'm just interested in a fair comparison (as in let them do the same test case, like having a stack of boxes and/or arbitrary shapes collapse) of the advantages and shortcomings of each engine.

Every list of an engine's features looks quite good to read, but sometimes flaws and shortcomings may only be found if one reads the code's comments or the forum posts carefully. For example, the source code for Tokamak's tutorial states that too heavy or too small boxes tend to (or tended to, maybe they fixed it) fall though the floor into the abyss. Not very good if you need to animate a bowshot or a gunshot for example.

Next, how do the engines behave under stress? How far do the frame rates drop if there are many things in movement? And does the quality of the simulation suffer if the simulation steps need to represent larger time spans just because of that?

Everyone, please... IMHO there is no 'best' engine - just one the most fitting for a special purpose, one may need a lot of features (like soft body physics or fluid dynamics), another may need rock-solid accuracy and a third one may opt for speed at the cost of both.

I don't fancy myself knowledgeable about physics engines, that's why I'm asking you. Perhaps one of you is willing to do such a comparison or maybe has done such a thing already?

hybrid · Post by **hybrid** » Thu Apr 13, 2006 10:07 am

Eternl Knight wrote:
How do I know these things? I'm an engineer. I studied these things at university. Not only that, but I also (to this day) pursue research on the topic. I read papers, look at code, and discuss issues with people who do this for a living.

So maybe you should also show your academic wisdom in the way you participate in discussions. So why do you freak out just because someone's telling nonsense (nosense in your eyes - I cannot judge this). I don't think that those people you were discussing with always had the same opinion, but I hope for all of you that those discussions were on a different level (w.r.t. emotions).
Maybe it's time to come back to the facts and accept different opinions as long as they won't become personally.