Intel buys Havok

[Dirk Gregorius]

http://www.intel.com/pressroom/archive/ ... -c1-183235

That is what Ageia was dreaming of. Maybe they are lucky and get bought by AMD now

[Antonio Martini]

http://www.intel.com/pressroom/archive/ ... -c1-183235

That is what Ageia was dreaming of. Maybe they are lucky and get bought by AMD now

surely Intel hasn't bought Havok in order to promote competitor processors.

So now the two main physics middleware companies are owned by hardware manufacturers.

I have the feeling that we will see more open source/in-house development in the future.

cheers,
Antonio

[Erwin Coumans]

http://www.intel.com/pressroom/archive/ ... -c1-183235

Interesting news.

That is what Ageia was dreaming of. Maybe they are lucky and get bought by AMD now

It is not wise for any company to buy Ageia anymore, because most of their key employees (and non-key) already left. I don't see much value in Ageia hardware strategy/IP/patents, so I would suggest to simply try to hire Ageia's Adam Moravansky, Richard Tonge and/or Matthias Muller instead.

If it is for an implementation, they better start with Bullet: it's parallel implementation for Playstation 3 and XBox 360 is quite competitive to Ageia PhysX, and Bullet is being used by several top game companies on those consoles
Erwin

[Dirk Gregorius]

I have the feeling that we will see more open source/in-house development in the future.

I agree. You can either use Bullet/Box2D/ODE or also write your own physic engine nowadays. There are a lot of good references out there right now and with Box2D and Bullet you have two very good open source implementations. The knowledge to write a physic engine is much better understood these days and more and more people know how to do it. I could imagine that this takes the same development as writing a 3D engine which is also understood quite well these days...

[Erin Catto]

Physics engine companies have a long history of coming and going. Have any of these physics engine companies ever been profitable? Better yet, have the investors ever got their money back?

Yes, hopefully these developments will lead to more collaboration in the area of game physics.

[Erwin Coumans]

Physics engine companies have a long history of coming and going. Have any of these physics engine companies ever been profitable? Better yet, have the investors ever got their money back?

Havok has been profitable. Bullet get's contributions and small donations now and then, I suppose it is the same for ODE. I doubt Ageia PhysX/Novodex is profitable.

All physics engines start very basic, with few features like Box2D.

Then the number of features grow, they are used in big projects, get optimized and and parallelized they end up like Havok, PhysX, ODE and Bullet. The physics engine becomes less educational and more complicated in this process. Writing a physics engine from scratch that is on par with Havok, PhysX, ODE or Bullet is a huge task unless you re-use the implementations (and their ideas) from existing / open source.

Professional game companies can and do contribute to open source, as long as the license it ok. That is why Bullet, Box2D and ODE choose ZLib/MIT or BSD licenses: the collaborations come from professionals and not just hobbyists. Physics engines tend to get thousands of features, and rather then pointing out some differences (2D versus 3D, integer sweep and prune versus float, one-shot contact manifold generation versus incremental, warmstarting, friction model etc) we should work on a large comparison matrix.

We need to collaborate rather then re-inventing the wheel.

Getting back to Intel: a few weeks ago they published a paper on parallelizing physics here

Erwin

[Dragonlord]

That is why Bullet, Box2D and ODE choose ZLib/MIT or BSD licenses: the collaborations come from professionals and not just hobbyists.

Sorry, but I have to disagree there. Why does GPL mean only hobbyists can contribute? This is nonsense. GPL simply means that the code has to be available to the interested people and prevents stealing code ( taking code of others and selling it as your own idea ). It's a protection and not a hindrance.

Back on topic. What exactly can be expect from this joint venture? I don't know yet what to think about it. Intel has done some good OS work but on the other hand also has shown to be a bastard blocking OS. So what will happen? What is their aim?

[Erwin Coumans]

Sorry, but I have to disagree there. Why does GPL mean only hobbyists can contribute? This is nonsense. GPL simply means that the code has to be available to the interested people and prevents stealing code ( taking code of others and selling it as your own idea ). It's a protection and not a hindrance.

Actually it was not my opinion, I am trying to describe the situation for professional console game developers (not XNA or PS3 Linux):

GPL is incompatible with most game companies and publishers policy. Also, GPL is incompatible with proprietary consoles like XBox 360 and Playstation 3: the NDA (non-disclosure agreement) and other license agreements explicitly forbid opening up source code, discussing the API and therefore mixing code with (L) GPL. Non-licensees don't have any means of compiling code for Playstation 3/XBox 360 game SDKs.

That is why we need licenses like BSD, MIT or Zlib for physics libraries.

Thanks,
Erwin

[Eternl Knight]

Sorry, but I have to disagree there. Why does GPL mean only hobbyists can contribute? This is nonsense. GPL simply means that the code has to be available to the interested people and prevents stealing code ( taking code of others and selling it as your own idea ). It's a protection and not a hindrance.

Using GPL also requires that any code it is used in also be licensed under the GPL. In other words, commercial (otherwise known as "professional") game development cannot use it and still expect to make money off of it. The commercial realities of "game software" development are quite unlike "platform" (operating systems, databases, web servers, etc) development. One cannot make money off of "game services" (with the exception of MMO* games, most of which will not require physics on the level discussed in these forums). In order to survive as a profitable entity, you pretty much need to keep source code proprietary... this is without considering the needs of proprietary console platforms such as Erwin mentioned above.

I only make this post in order to "enlighten". That is, if this is about "GPL is better than BSD/Zlib/etc" rather than "Why isn't GPL feasible in the game physics areas of development?" - then I am simply not going to respond. I respect the boards too much to be involved in a religious/political discussion on them.

--EK

[Dirk Gregorius]

Back on topic. What exactly can be expect from this joint venture?

That is really the question. I really wonder what Intel expects from this deal?

[wyd124]

Now, Havok become a part of Intel. What does this mean? Will new physics acceleartor born with the buy?

[Dragonlord]

Sorry, but I have to disagree there. Why does GPL mean only hobbyists can contribute? This is nonsense. GPL simply means that the code has to be available to the interested people and prevents stealing code ( taking code of others and selling it as your own idea ). It's a protection and not a hindrance.

Using GPL also requires that any code it is used in also be licensed under the GPL. In other words, commercial (otherwise known as "professional") game development cannot use it and still expect to make money off of it. The commercial realities of "game software" development are quite unlike "platform" (operating systems, databases, web servers, etc) development. One cannot make money off of "game services" (with the exception of MMO* games, most of which will not require physics on the level discussed in these forums). In order to survive as a profitable entity, you pretty much need to keep source code proprietary... this is without considering the needs of proprietary console platforms such as Erwin mentioned above.

I only make this post in order to "enlighten". That is, if this is about "GPL is better than BSD/Zlib/etc" rather than "Why isn't GPL feasible in the game physics areas of development?" - then I am simply not going to respond. I respect the boards too much to be involved in a religious/political discussion on them.

--EK

This is not the point. Erwin went about it the right way. If you look at it from a "console" ( aka locked down hardware ) then yes GPL doesn't work out this is correct. I also said only that GPL does not "shut out" commercial companies. Everybody can contribute no matter what level they are on. GPL simply makes sure that what you contribute stays open and accessible to all others also in the years to come and that you are "not" responsible for cranked up modifications ( author protection ). This is not in any way a religious fight. Different licenses exist for reasons but sometimes licenses are "misunderstood"

[Erwin Coumans]

Let's not go too much into detail on the GPL versus other license in this 'Intel buys Havok' topic.

Basically (L)GPL doesn't mix with console development, and most PC game companies don't want to be forced to open source their game for obvious reasons. And even if they want, they will have a hard job finding a publisher that allows them doing so.

If you want to discuss licensing issues, please continue the discussion on this existing licensing topic 'Zlib/MIT/LGPL/GPL Collision Detection Library License'.

Thanks,
Erwin

[Erwin Coumans]

Back on topic. What exactly can be expect from this joint venture?

Intel is investing heavily in many-core chips, in their latest Tera-scale physics paper they mention 64 cores. Such 64-core machine can accelerate physics, graphics and many other tasks.

On 64 cores, we achieve 30x to 56x speedup for production physics and 36x to 61x speedup for game physics.

So my question is how they make programming easier for such architecture, and how they deal with memory/cache/DMA for all those cores.

Parallelizing a non-trivial piece of software takes a lot of time, as we experienced during the Cell SPU porting work of Havok, Bullet and PhysX. To get some estimated numbers: parallelizing Bullet's collision detection + constraint solver for Cell SPU/multi core took almost 2 man year in total, and Havok invested over 10 man year in their Cell SPU port, using their best engineers.

Intel wants their upcoming massively multi-core chips to be the best in running relevant gaming applications, so acquiring Havok makes a lot of sense to me: they ensure enough Havok resources are put into their upcoming many-core architecture, rather then on GPU (NVidia, ATI/AMD) or Cell (IBM, Sony)

[jdolan]