Intel buys Havok

[Antonio Martini]

http://www.developmag.com/interviews/81 ... vid-OMeara

[bone]

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

In answer to the second part of your question, the paper basically admits as much in the introduction: "A radical re-design of the memory hierarchy may be necessary for the multi-core tera-scale era to provide good scaling for this type of application."

In answer to the first part: I don't think multi-threaded programming is ever going to be dead simple. However, my opinion is that multiple uniform chips (like multi-cores) would be easier than the current challenges of trying to divvy up tasks among CPUs, SPEs, graphics and physics cards. The chip companies also need to continue lowering the overhead (context switching, thread creation and synchronization) of using multi-threading: the current recommendations for Intel's Thread Building Blocks (TBB) library is that you need to have a chunk of code with several thousand instructions before it's worth trying to break it up. That's okay when optimizing your biggest bottlenecks, but the cost/benefit seems to break down quickly from there.

/my two cents

[DavidWu]

While I do agree with Erwins comments in principle, I would think that $110m for a handful of fairly good programmers is not a great deal.

My 10 second valuation of Havok would be:
- Customer base, Inertia, Trust, Name recognition
- Marketing
- Average physics engine with
- Cross platform support
- Documentation
- Stable enough, mature code.
- An architecture that is not completely resistant to parallelization
- A handful of good programmers
- A number of average programers
- Cutomer Support infastructure.

Which of these will benefit Intel?

David Wu
Pseudo Interactive

[bone]

1) Marketing and
2) an architecture that is not completely resistant to parallelization (which will indirectly lead to better Marketing).

[Antonio Martini]

Average physics engine with

although Havok may not be an ideal Physics engine are we sure it is average? i mean if we take the list of the available physics engines known to the public and we order them based on some quality criteria, are we sure that Havok is in the middle of the list?

i would not also fall in what i call the "programmer's fallacy" that is "if i had to write it, i would rewrite it better" because that's true today also for the same people who wrote Havok if they had to start designing/writing it from scratch, so which "better" is better?:)

cheers,
Antonio

[DavidWu]

Okay, so you caught me on the "quality criteria" bit.
For most developers doing cross platform development across x360 and Ps3 I would say that Havok would have been the safest bet.

If I did not have my own physics engine, I would want to re-write my own.
However, that would cost a lot of time and money, so pragmatically I would probably chose something open source that I could modify. Preferably written by someone that I know. Bullet seems like a good starting point.

I do think that it is rather amusing that many people working on the "many core" project did not know that their company was going to purchase Havok.

[Dirk Gregorius]

>>Okay, so you caught me on the "quality criteria" bit.

I think most current engines use some PGS/SI iterative solver. The devil might be in the detail and from my experience small changes can make big differences. But what do you think how a physic engine 2.0 should look like? What will be the differences? Not only from a mathematical/technical side, but also what kind of physics does the player want to see in next-gen games.

I made the experience that even though you can do something with physics you not necessarily should do it


Cheers,
-Dirk

[Antonio Martini]

But what do you think how a physic engine 2.0 should look like? What will be the differences? Not only from a mathematical/technical side, but also what kind of physics does the player want to see in next-gen games.

deformation, fracture, clothes, hair, physics based animaton, 3d fluids/gases? anything else?:)

if a world looks real i would expect it to behave realistically. Physics and animation are behind graphics in such respect at the moment. There is no need to mention AI of course..

cheers,
Antonio

[Dirk Gregorius]

deformation, fracture, clothes, hair, physics based animaton, 3d fluids/gases

Very true, but do you think you can have everything simulataneously in one game? On current hardware you might consider specializing on one of them (the one that supports your game the best) and concentrate on this.

[Antonio Martini]

deformation, fracture, clothes, hair, physics based animaton, 3d fluids/gases

Very true, but do you think you can have everything simultaneously in one game? On current hardware you might consider specializing on one of them (the one that supports your game the best) and concentrate on this.

i think it's possible if all those things dont happen simultaneously or not in large quantities. i mean if rather than having clothes on n characters you have clothes on n/2 instead, maybe you can fit in a bit of fracture in etc... maybe you cannot have 3d fluids in the same room where a lot of fracture happens.
Surely it's not an option that i would discard without giving it a go, resources permitting of course;)

cheers,
Antonio

[Dirk Gregorius]

Capcom replacing Havok:

http://www.developmag.com/news/28512/Ca ... use-engine

I would like to hear the justification on this. Anybody?

[bone]

I don't see where they claim it's useless. They're just concerned about language discrepancies, turnaround on bug, and lack of understanding on how their entire engine works. So they've concluded it would be better to develop it themselves.

Why not Bullet?

[Dirk Gregorius]

Oops. I misread this - I really should read more carefully. Fixing it in my original post. Anyway my question is if this might be somehow related to the deal with Intel and also the lawsuit between Silican Knights and Epic...?

[Antonio Martini]

Capcom replacing Havok:

http://www.developmag.com/news/28512/Ca ... use-engine

I would like to hear the justification on this. Anybody?

there is an interesting related article at page 17 of this month's Develop that can be downloaded from:

http://www.developmag.com/print-edition

cheers,
Antonio