Ageia hasn't really been very open about exactly how its PPU works, which leaves us mostly guessing, trying to figure out how it works from the information released. Scientists speculate too, but they have a good understanding of how real physics works and how well it can be simulated. So let's first look at the facts.

Basic physics simulation, like objects in free fall or particles interacting with soft forces, isn't really that hard to do. Scientific literature on these problems goes back for decades. Even older is science dealing with celestial mechanics, such as planets. Many of these algorithms work well in a gaming environment, and are also used in gaming physics today.

When you move to matters like a very large number of particles and long-range interaction forces, which also change quickly, things start to get difficult. And when, for example, handling rigid multibody systems coupled with kinematical constraints, like robot arms, you suddenly have really tough problems. That's especially true when:

• There are closed loops in the system;
• The rigid bodies are rotating quickly (gyroscopic forces are problematic);
• There are both very heavy and very light objects interacting with each other;
• There are contacts; and
• There are collisions and impacts.

"Regarding heavy and light objects, as well as contacts and impacts, nothing published is 100% accurate," says Claude Lacoursière, chief scientist at CM Labs, a company specializing in simulated physics. Although he says there are good algorithms that deal with impacts and contacts, they have yet to find their way into gaming physics. According to Lacoursière, the current algorithms used in gaming physics are generally low-order approximations that usually cause stability problems, which are then fixed by crude tricks. For example, a thrown hammer in a game usually won't obey gyroscopic laws at all; instead, it just behaves more or less like a ball. This might not sound like a big deal, but for the folks looking for realistic physics, it likely is.

Claude Lacoursière: „Nothing published is 100% accurate."

Looking at all of this, one can see why simulating real physics is tough. To really solve these kinds of problems, Lacoursière says you need equations that are really heavy on calculations. For simulation purposes, these time-consuming calculations aren't a problem, as scenes can be rendered over a period of time. This is best done accurately with raw CPU power, where you can change or add new algorithms when needed. But in games this needs to be done fast, in real time, without slowing down gaming performance. To do that in an efficient way, you are forced to make some shortcuts - and shortcuts don't work well in science.