Unlimited detail in computer graphics

Wrote this:
"Even if you just render XXXX by XXXX pixels, how much processing power will this take? And would it be time consuming creating an object with UPCD? It sounds time consuming, and if you're supposed to create multiple hundred different, for example, rocks - wouldn't that just be inconvenient?"

It looks cool, but I'm a bit skeptical.

Here are all the comments I wrote:

Quote :

Even if you just render XXXX by XXXX pixels, how much processing power will this take? And would it be time consuming creating an object with UPCD? It sounds time consuming, and if you're supposed to create multiple hundred different, for example, rocks - wouldn't that just be inconvenient?

Quote :

@dingumfCallofDuty I did.
They never explained it. I'd like to see some benchmarks.
I mean, a player moves in 30-60 FPS. 100's of atoms come on screen every frame. 10's of thousands would render every second.
That doesn't sound very conservative to me.
I would also imagine animation and physics are huge issues for this kind of system.
I mean, with all of the trouble making those backdrops, you'd think they would spend a few minutes showing some kind of animation.

Quote :

I won't deny, it looks awesome, but yeah, we gotta get stuff moving, then we got to get it moving dynamically, and then we gotta throw in all the actual game data to correspond with it.

I don't think this stuff is as close as that guy wants us to think it is. But I won't deny it's certainly interesting.

Quote :

One last thing. Google uses a server side algorithm. They have MASSIVE computers to run them. Thousands of servers actually.

For this to work not only would it be an enormous hassle (Probably) to program, but at the same time it'd require an obscenely large amount of money and more computers (3 dimensions, trillions of dots) then even the multi-billion dollar company Google has.

You've got to agree that it is a bit fishy.

Quote :

And when you scan an object, there's also a lot of work by hand.
Scanning isn't perfect.
Most 3D modelers don't use it.
Polygons are incredibly easy to cater. Vertexes are insanely hard.
Point cloud data bases itself around a vertex system.
This means it'll probably be fairly hard to make a good looking model unless they come up with some sort of brilliant SDK.

Quote :

@mercartax Of course the problem there is if you export a flat surface and create point cloud data, all you'll get is a flat surface consisting of many dots instead of a flat surface.