To scale with as many CPU and graphics cards/cores as you can throw at it?
Feels like we're getting there, but more leaps will be made once this is possible.
What do you guys think?
It doesn't feel like we're getting there. Parallel computing is the holy grail, but it also isn't possible to make games truly parallel.
The likely answer to your question is, as FT said, never.
--
Slightly longer answer: if you throw a ball against a wall, the ball bounces off the wall. The instant the ball tries to occupy the same space as the wall, a physical interaction occurs. The ball doesn't actually need to know anything about the wall. The ball doesn't need to know the wall even exists. The ball is just travelling and reacting to things in its path.
Throw a virtual ball at a virtual wall and things are suddenly different. There is no physical interaction anymore. Instead we are describing a ball, and describing a wall. It is a simulation. For the ball to hit and bounce off the wall, the thread needs to know about both the wall and the ball, and have accurate information about the state of both of them.
Now lets think about a simulation of several balls which all move, and could all hit and bounce off each other. Now each ball needs to know about the other balls in its vicinity, needs to test if they are colliding, etc.
In other words (and this is hard to explain properly), you can't just give each ball it's own CPU or thread and have at it. At some point you need to collate information about every ball together to check for collisions. Each ball could try to use predictions about where the other balls would be, but sometimes those predictions would be wrong.
What I'm trying to say is that in a situation where you can have dynamic interactions, you can't go fully parallel, as far as I'm aware.
For other software, like video processing software, this isn't the case. You can divide up the picture into as many small sub-pictures as you like, as what happens in one doesn't affect what happens in another.
