Physics Processing Units (PPUs)...

[Covak]

http://www.theinquirer.net/?article=21648

 

Sweet I hope it takes off.

[Ryan FB]

I'm not usually a grammar/spelling pedant, but holy crap, do they have editors over there?

Big guys like Gabe Novell, the developer of Half Life 2, asked for more physics and Jon Carmak of Doom 3 wanted the same. The industry likes the marchitecture and developers want to programme for it.

Other than that, it sounds interesting.

[dogbert]

Big guys like Gabe Novell, the developer of Half Life 2, asked for more physics and Jon Carmak of Doom 3 wanted the same.

 

Which isn't the same as asking for dedicated cards in PCs for it I just don't see this taking off, especially when it's being pushed by one particular physics solution provider, one that isn't Havok

[Covak]

Which isn't the same as asking for dedicated cards in PCs for it I just don't see this taking off, especially when it's being pushed by one particular physics solution provider, one that isn't Havok

 

This particular card might not take off, but it's a start. And I'm betting that once gamers and developers see what can be done with dedicated physics processors, nobody will want to go back.

[Camp]

Ubisoft has signed up to support it.

 

Ubisoft and AGEIA Join Forces to Re-Ignite Game Innovation

 

Game publishing giant chooses AGEIA hardware-accelerated physics technologies to power next-generation titles

 

MOUNTAIN VIEW, Calif. ? March 08, 2005 ? AGEIATM Technologies, Inc., a company dedicated to delivering pervasive interactive reality to next-generation games, announced today that it has signed an agreement with Ubisoft, one of the world's largest video game publishers, to provide technology for Ubisoft's next-generation game titles.

 

AGEIA's PhysXTM chip creates an entirely new category of processor, the Physics Processing Unit (PPU), which represents the next milestone in the evolution of game acceleration hardware. AGEIA's NovodeXTM Physics SDK is uniquely equipped for this revolution in hardware-accelerated physics. NovodeX is the only multi-threaded physics API that will unleash the power of multiprocessor systems.

 

"The Ubisoft creative team thrives on working with the most advanced technology available, which drives both our creativity and our passion for making state-of-the-art games," said Yves Guillemot, Ubisoft's President and CEO.

 

Ubisoft is home to the second-largest creative team in the games industry and has developed some of the most successful interactive games in the world, including Tom Clancy's Splinter Cell, Tom Clancy's Rainbow Six, and Prince of Persia.

 

"It is exciting to see our technology embraced by a company with the stature of Ubisoft," said Manju Hegde, CEO and co-founder of AGEIA. "Together with Ubisoft, we are determined to re-ignite the passion and creativity of game developers, enabling them to integrate dramatic game play and develop totally immersive environments."

 

I like the idea of hardware specific to physics but only if that doesn't limit the developer to Ageia's tools.

It would be nice to see hardware support in the next gen consoles. Given Ubisoft's support of Xbox I'm almost curious if this isn't some sort of hint. Probably not.

[Covak]

http://ve3dboards.ign.com/message.asp?topic=18668945

 

Loads more info there. Sounds like they mean business

[dogbert]

It would be nice to see hardware support in the next gen consoles.

 

Nextgen hardware does not need dedicated 'physics units' - they're a bunch of parallel processing high speed units which churn through maths job, they're perfect for it already

 

Given Ubisoft's support of Xbox I'm almost curious if this isn't some sort of hint. Probably not.

 

Should be noted that Novodex is a software physics solution like Havok, as used by Epic, Cryptic etc.

[iainl]

I was about to ask just that - isn't a Physics Processing Unit just a pointlessly crippled FPU that you can't use for something more general?

[Cyberwoo]

All I know is after playing Half-Life 2 and enjoying the hell out of its physics model I can't wait to see where next gen games take physics. I've always been a physics whore. Hell I played through the entire game of Tresspasser just because of the physics.

[Ryan FB]

I was about to ask just that - isn't a Physics Processing Unit just a pointlessly crippled FPU that you can't use for something more general?

The answer is "not necessarily, depending upon how they're implementing it". An FPU does not always perform multiple operations in one pass, they can operate in a scalar manner like a "traditional" processor. That is, you could have two floating point numbers in two registers, then perform some arithmetic operation on them, but you couldn't have a whole list of floating point numbers and do something to all of them with just one instruction. The real advantage of a processor like this would be in its vector processing capabilities - its ability to perform the same instruction on a large amount of data with only a few cycles. For example, let's say you have a slew of meshes in 3D space (made up of connected points), and you want to transform them all by some constant (apply the effects of gravity) and perform collision detection, without having to go "ok, do it for this point, now do it for this point, now do it for this point" - that's the kind of thing this has the opportunity to excel at. Once that load of processing the physics of a scene is taken off the CPU, you have that much more free general purpose processing power to play with for other cool things.

[Ryan FB]

There's an informative interview with the developers over at Gamers-Depot.

[rustyjaw]

I'm having a hard time imagining this taking off in any way near the way GPU's did. But looking at that interview, these guys think it will:

 

The PPU in many ways will make changes in the game industry in similar ways as the GPU did in the late nineties, Similarly GPU?s made many leaps in graphical processing power available to the game developers. We anticipate the PPU to bring pervasive interactive reality to game play.

 

The main difference is that every home computer has a video card (excepting headless servers) already. It's the ubiquity video cards in the first place that allowed companies like Nvidia and ATI (and S3, 3DLabs, and others back in the day) to push for faster 3D performance in games, and faster 2D for everyday things like scrolling images and documents. This ubiquity also allowed prices to remain fairly low (at least once you got out of the very highest end of technology).

 

I don't see the same opportunity for an add-on product that has no reason to exist outside of gaming, unless they can find one.

 

I'm sure a number of gamers will want one, and I'm not doubting the results could be impressive, but I'm not convinced that's enough to overcome the 'barrier to entry' that video cards never faced.

 

EDIT: I initially thought that this technology should just be added to the video card, but then I saw pictures of the cards in that interview, and they look to be every bit as large as a video card. So I guess that won't happen yet, but I wouldn't be surprised if ATI and Nvidia weren't already working on this idea as a way to further enhance gaming performance via the video card.

[Whooter]

They'll have to be built into the video cards themselves if they expect anyone but Joe Hardcore to have one in their system...

[Ryan FB]

I initially thought that this technology should just be added to the video card, but then I saw pictures of the cards in that interview, and they look to be every bit as large as a video card. So I guess that won't happen yet, but I wouldn't be surprised if ATI and Nvidia weren't already working on this idea as a way to further enhance gaming performance via the video card.

I have to agree...I think this technology only really has a chance if the graphics card manufacturers acquire them, license it, or develop it independently (or for inclusion in a proprietary game console). I think it could work quite well if done correctly, most of the architecture you see on the cards in those pictures is pretty much the same as what's on a video card already, so integrating it wouldn't really take up any more "circuit board space". Since graphics cards already rely on fast vector processing to do what they do, they could simply abstract it a bit further and provide a seperate set of instructions for doing the physics processing as well (maybe with multiple vector processing pipelines, although they may already do this in newer graphics cards, I don't know). This would also be a good justification for moving to PCIe for the added bandwidth it provides, and make it so gamers only have one line of absurdly overdeveloped accessories to drool over.