If you are making virtual reality in Matrix style, where you plug directly into the brain, instead of putting displays in front of your eyes, you can inject the sensations lot more directly. You can save lot of computing power.
If you are playing a videogame today, the computer has to take the models, textures, etc and render them into a photo-realistic picture that your eyes can see, then your brain breaks it back down, recognizes the parts and creates a mental model for you. You could skip the whole rendering part and just inject the model into the brain.
Two examples from everyday world:
- Dreams. When you dream, you can "see" all kinds of things, but your brain does not create a precise image that would be fed back into the nerves that come from eyes, the dreams are created deeper inside as more abstract concepts.
- Drawing what you imagine. All of us can imagine almost anything that is described to us - "a small cottage at the side of a river in front of beautiful mountains" - you can imagine it and "see" it clearly in your head, but unless you are trained artist, it is difficult to draw this to paper. It is partly because what you "see" in your head is not a 2D image, it is a mental model created by your brain as-if you saw it.
Today we have very little knowledge how is the mental model structured in the brain and how much it differs from person to person, but in the future we may know more and we may figure out how to affect that parts of brain directly. Then you would change the computer simulation to represent the world in a way that would be optimal for turning into mental models that are injected into people's brains, which is likely quite different from the representation that we use today for rendering.
Interesting outcome of this is that if your simulation is not good enough, people won't see pixelated textures or low-polly models. It will rather feel like a fuzzy dream. You see something, but you can't get yourself to focus on details.
A quote from Matrix the movie:
Neo: Do you always look at it encoded?
Cypher: Well you have to. The image translators work for the construct program. But there's way too much information to decode the Matrix. You get used to it. I...I don't even see the code. All I see is blonde, brunette, red-head. Hey, you a... want a drink?
The Matrix code is some kind of a transcript of the mental model that would be sent to person's brain in that area.
So at the end we don't know how much computing power will be needed to simulate the world in Matrix-style virtual reality, but it should be lot less than if you try to render the view of every person as an image. I would compare it to servers for massively multiplayer online games today, rather than to rendering farms. A MMO that puts all players into single world, rather than into multiple copies of the same world. Such as Eve Online.
Here is Eve Online's Server Model.
Each of EVE's 5000+ star systems is loaded as a separate process onto any one of hundreds of IBM blade servers, with some high-load systems being given a server all to themselves and many low-load systems being combined and run on servers together.
EVE's server houses over 300,000 players with a peak concurrent user record of over 40,000.
40,000 could be a population of smaller city. Of course simulating real-world-like experience will be harder than simulating game with ships in space.