Havok, gpgpu and physics, Uberflow ?

Please don't post Bullet support questions here, use the above forums instead.
User avatar
Erwin Coumans
Site Admin
Posts: 4221
Joined: Sun Jun 26, 2005 6:43 pm
Location: California, USA

Havok, gpgpu and physics, Uberflow ?

Post by Erwin Coumans »

Havok is planning to use the gpu and shadermodel 3 for some of their rigidbody calculations: http://www.havok.com/%20index.php?optio ... &Itemid=77

Closest I can think of is http://wwwcg.in.tum.de/Research/Publications/UberFLOW

Has anyone experience in this area, that can do help doing basic rigidbody collision detection and dynamics?
Julio Jerez
Posts: 26
Joined: Sat Jul 23, 2005 12:56 am
Location: LA

Post by Julio Jerez »

This is quite interesting. A while back I was speculating that maybe companies like Havok and Renderware might seek licensing the instruction set from video card manufactures so they can make special version of their engine and bundle it with the drivers, in some kind of partnership.
I though for example that with the increasing cost of these later video card, some body will eventually has the idea that it will be easier to justify buying a $500 hardware accelerator that will accelerate all aspect of the game, rather than justify $500 for a video card, $300 for a digital signal processor for physics, and maybe another $200 for sound.

It did not occur to me that once again a solution might indirectly come from Microsoft. I had not read the details of the vertex shaders 3 specifications, but if it is as flexible as it sound then it is very good news for the independents developers, because at least it means they will have a shot at hardware acceleration.

If vertex shader 3 can be seen like a general-purpose signal processors, then is can be used not just for graphics and physics but also for many other things.
I am very excited about the possibilities.

Edit:
Um I see vertex shader 3 is an old dx 9 feature, I do not spend too much time on dx, for a moment I thought is was some new specification for dx 10. It would it be very nice if all that brute force floating point power could be addressed directly without having to write mombo-jumbo graphic geometry tricks. Just old plain John von Neumann style
Julio Jerez
Posts: 26
Joined: Sat Jul 23, 2005 12:56 am
Location: LA

Post by Julio Jerez »

Speaking of hardware, has anybody read about this.
http://www.clearspeed.com/
It looks like a dream vector processor with 96 double precision cores per chip.
It might be on the expensive side.