PhysX Dedicated Physics Processor Explored

Ned_Network writes "Yahoo! News & Reuters has a story about a start-up who have created a dedicated physics processor for gamers' PCs. The processor undertakes physics calculations for the CPU and is said to make gaming more realistic - examples such as falling rocks, exploding debris and the way that opponents collapse when you shoot them are cited as advantages of the chip. Only 6 current titles take advantage of the chip but the FAQ claims that another 100 are in production."

Comments

Physics Good, Fire Bad

[Cy Sperling]

I like the idea of offloading physics processing to a speciallized card. Seems like it should up the ante for games to move beyond just ragdoll physics for characters and into more environmental sims as well. I would love to see volumetric dynamics like fog that swirls in reaction to masses moving through it. A deeper physics simulation hopefully means more to do rather than more to look at as well. Playing with gameworld physics from an emergent gameplay standpoint has real play value versus larger prettier textures.

Re:Is that what I think it is.

[TubeSteak]

Chess games rely on brute computation to up the difficulty level.

Anything the programmers can do to examine more moves into the future is a good thing for them. Even Deep Blue couldn't look more than 30 moves into the future. Dunno about the 'son of' Deep Blue.

Animations, etc consume trivial amounts of CPU/graphics power compared to examining the next XY possible moves in a chess game.

Here's the problem with this

[SlayerDave]

There is no common, open API for physics. Rather, there are several proprietary, closed APIs which offer similar functionality, but have no common specification. For instance, there are Havok, Ageia, Open Dynamics, and Newton, just to name a few. The PhysX chip from Ageia only accelerates games written with their proprietary library in the game engine. Other games written with Havok, for instance, should receive no benefit at all from the installed PPU. On the other hand, Havok and NVIDIA have a GPU-accelerated physics library, but games without Havok (or users without NVIDIA SLI systems) won't get the benefit.

On the other hand, graphics cards make sense for consumers because there are only two graphics APIs, OpenGL and DirectX, and they offer very similar functionality under the hood (but significantly different high-level APIs). So a graphics card can accelerate games written with either OpenGL or DirectX, but that's not the case with the emerging PPU field. In graphics, the APIs developed and converged on common functionality long before hardware acceleration was available at the consumer level, but I don't think the physics API situation is stable or mature enough to warrant dedicated hardware add-in cards at this time.

However, I think there are two possible scenarios that could change this.

1) Havok and Ageia could create open or closed physics API specifications and make them available to chip manufacturers, e.g. ATI and NVIDIA, which have the market penetration and manufacturing capability to make PPUs widely available. I could imagine a high-end PCIe card that had both a GPU and a PPU on-board.

2) Microsoft. Think what you will about them, but DirectX has greatly influenced the game industry and is the de-facto standard low-level API (although there are notable exceptions, such as id). Microsoft could introduce a new component of DirectX which specifies a physics API that could then be implemented in hardware.

But unless one of those things happens, I don't think proprietary PPUs are going to make a lot of sense for consumers.

Already exists. Kinda

[Opportunist]

It's called "creativity" and is normally used only in the development of games. Actually has been for ages before studios found it too expensive, and realized it's cheaper to develop games without it.

Multiplayer

[lord_sarpedon]

There's a major flaw. Multiplayer gameplay requires certain clientside behaviors to be deterministic, otherwise clients will fall out of sync. Physics is one of those. If Bob uses a PhysX card and an explosion lands a box in position X, but Alice, without a PhysX card, has the same box in position Y, then there is a problem. Both can't be right. The server would have to correct for discrepancies such as that because the position of a box affects gameplay; bullets and players can impact it. Perhaps more position updates would have to be sent to make sure Alice ends up in the same spot as Bob. But what about midflight? I suppose this doesn't matter for blood smears and purely aesthetic effects, but as the videos show, thats not where PhysX really shines. This puts a physics accelerator in an entirely different class than a graphics card. You can adjust your graphics settings, but the quality of your physics simulation in multiplayer can only be as good as the least common denominator without killing gameplay for some of the parties involved. Sure, AGEIA could have non-accelerated versions for everything in its library when acceleration isn't available that produce the same result, but then you are offloading the entire functionality of an addon card on to the cpu...imagine running Doom at full settings using software rendering. Extreme example. But that defeats the very purpose of the card, if developers are limited because most of their customers might not have it.

Re:Great for single player, bad for multiplayer?

[MobileTatsu-NJG]

"How do they expect to run these huge physics calculations over the internet in a massive game like say for instance Battlefield 2?"

I can offer an uninformed theory. If an event is passed to the other players such as "barrel at explodes", then the processing is done at the client end for all of the players. If the event is done properly, they should all reach the same conclusion.

Unfortunately, as I'm writing this, I can start to see the problem. Okay, I apologize, but I'm going to do a 180 here. Imagine a car crashes through a brick wall and a hundred bricks go flying away. That alone should work fine. But if another player runs into the path of one of the bricks and it bounces off of him, suddenly it's no longer as predictable. His latency along with everybody else's latency means ONE of the computers has to make the decision of where everything goes. That, in and of itself, is probably okay, but then you have to pass a great deal more data along to let the other clients know what's happening.

So... yeah, I see your point.

History repeats itself

[rasmusneckelmann]

Back in 1995 game developers made 3D games using software rendering; then suddenly a company called 3Dfx introduced a dedicated 3D chip called Voodoo Graphics. Hardware acceleration of 3D was no new thing at that time, but 3Dfx was the first who would sell it to normal consumers. In the beginning, everyone thought it was insane to offer that kind of dedicated chip. Everyone was wrong, 3Dfx with their Voodoo Graphics was a massive success; soon all game developers supported 3Dfx's proprietary 3D API "Glide". Then came all the other "conventional" big players of graphics hardware, like ATI, nVidia, and Matrox, and started implementing similar features into their video cards. Microsoft introduced Direct3D to offer a uniform interface to consumer 3D rendering, and video card manufacturers even started to support OpenGL. 3Dfx and their proprietary API slowly faded away.

My best guess is that this is going to repeat. AGEIA have now done what 3Dfx did, introducing a dedicated hardware chip for something that until now has been done in software. They even have their own proprietary physics API. Soon ATI and nVidia will incorporate similar features into their GPUs, and Microsoft will create a brand new DirectX subsystem called DirectPhysics. And AGEIA will slowly fade away (if they don't learn from 3Dfx's mistakes).