It will probably end up like most computer components. Enthusiast and mid-range. A dedicated PPU would most definitly have more power than a GPU split between graphics and physics. So the Enthusiast market would be more likely to have 2 cards. However, if ATI (and you can bet Nvidia will follow) can swing a good Physics processing on their card I bet you’ll find a lot of people using it for midrange systems.

Overall this means after a couple years the majority of PC gamers will have something in their machine capable of running physics beyond just a few hundred objects in the level. This opens new doors for developers and the kind of games we make. Good stuff.

I disagree with Kelmon as well. I feel physics can enhance a game experience. While if you gave me the choice of AI or Physics I would definitly go with AI (I would love an AIPU) I will not reject some fantastic game mechanics and visuals just because I am not getting other things.

Granted, at the moment, the ideas seem to be for dual card solutions - one GPU to process the graphics, one to process the physics, etc… but I think it’s a move in a really fascinating direction. We could end up with “gaming” cards which not only do graphics, but also Physics and AI calculations as well. We have multi-core CPUs now… we could end up with multi-core GPUs, so you could have it all on one card. I’m not sure if that’s exactly the best idea, but it is an interesting one to think about.

Realistic physics in games is an overblown concept and there’s more important things to work on

Seriously, I really don’t care about physics in games that much; they’re already fine. I’m more in favour of letting the developers determine what I can or cannot do in a game than incorporating the facilities that could potentially allow me to break the game. This is perhaps going a bit too far but consider the FPS, Red Faction, and it’s Geomod engine that allowed the player to deform the environment. Originally an exciting concept it was found to be rather dull in practice because the player was only free to destroy aspects of the level that had been “marked” as destructible. A complete implementation of physics in a game would allow the player to destroy absolutely anything in much the way that we imagined that the Geomod engine would. In practice, however, I think that this is “a bad thing” as it will make it much easier for the player to take paths that weren’t intended and, therefore, potentially break the game. Freedom of choice is great and all but there needs to be something fun at the end of each choice and that requires advanced knowledge of all the possibilities.

My position is that games need to focus more on other areas, such as the Artificial Intelligence of the Non-Player Characters in a game, be they enemies in a FPS, characters to interact with in an RPG, or troops under your “control” in an RTS. Improving, for example, the path-finding abilities of troops in an RTS would add dramatically more to the overall gaming experience than realistic (or distorted) physics. Once developers have sorted out core aspects of their games, then they can play with the physics as much as they like for all I care. Just don’t be surprised if the end results aren’t all that impressive…

As for ATI, I don’t much care. Making efficient use of a computer’s resources is important and therefore I fully support the use of the GPU for performing such calculations if it has the spare capacity. It is, however, up to both developers and ATI to demonstrate how the GPU could be used and create the necessary drivers and libraries to make use of them. Until this happens then this announcement is nothing more than hot air.

The same can be said with regular CPUs. They aren’t built for/good at it, and using them for physics in games takes away processing power that could be used for other things. But you need to remember that GPUs are a lot more complex and it appears as they they would do a better job at physics processing than a CPU would. Plus, at the moment, we’re finding that with these stupidly powerful graphics cards, people are CPU-limited in most games.

So if we’re CPU limited, it’d be less trouble to take some untapped power from the GPU and put it towards physics calculations, so as to free the CPU from having to do them. And all this without having to spend $250 US on a separate physics card. I think it sounds promising, and if ATI can do it I don’t see why Nvidia can’t as well. Hell, maybe it means AGEIA could do the reverse and start processing graphics themselves.