They really strike me as guys looking for investors that plan on cutting and running. That video is obviously not aimed at anyone in any relevant field. Who are they trying to sell their engine to? People who know what voxels are and the problems involved? Someone who knows voxels by the name voxels? Who knows the term level of detail? Or are they looking for some idiot with too much money that's willing to fall for the terms "UNLIMITED DETAIL" and then run away with his money once he realizes it's not digital magic?

This is old news (well, old news in a new nice box, nice paper even). It looks really nice when static, but try to animate it and it goes to the shits. Scripts, shadows...

Now, if we were talking some kind of a hybrid, that could be interesting. Memory could still be a problem, though.

Yeah, there's no way this could be true.

He's either lying about the amount of detail or lying about the computer required to run it.

With all due respect to Notch and his work, if you decide to argue somebody else creation, you should at least have the courtesy and use the data they provide, not some random guess that you pulled out of your rear ;)
Bruce Dell mentions the number of 'atoms'/voxels used in the island rather explicietly (21,062,352,435,000 polygons, though I'd guess he meant atoms, that still limits the amount of voxels) and that 22 trillion is rather far from Notchs guestimated 512000 trillion. At 12 byte data per atom (3 doubles for the coordinates and a 4 byte colour) that leaves you with about 240 Terabyte raw data. Still too much for a game to download in reasonable time, but that sure fits a single modern HDD. And that doesn't take compression into account, which would score massively here, given the amount of multiples you'd find object wise.

Does anyone even have eyes fast enough to see 120fps? :P

Pigeons maybe...

Wait... what. The biggest consumer drive I've ever seen was in the realm of single figures of terabytes (might have been 8).

Yeah... I don't trust him at all. It's a pity, because I remember wondering about this when I was little (10 maybe). I was aware that 3D graphics were polygon based (how could I not be in 1998) but I remember thinking "I wonder if in the future, CG things will be made of atoms?"

Dunno, I believe we're capped somewhere around 100fps, but 120 is double of 60 so it would make sense to cap games at that much. Just in case somebody improves his eyes somehow. :p

My bad, sorry. I meant the likely size of compressed data, not the raw one. I massively screwed up there. Good catch ;)

Edit:
If you can notice 100 frames per second, an old fashioned TV sure had to be 'paper cinema' to you with it's roughly 24 frames per second refresh rate :p

I was really trying to derail this into a Reboot nostalgia thread.

It's a nice buffer zone between "fps to the max" and "oh, so much action, so much lag". A bit excessive, if you ask me, but you need some extra fps to ward off the sudden drops that happen sometimes.

I did something similar that for a university project years ago.

Used self-contained dots (well, very small cubes) instead of your regular polygon mesh.

The problem with that is that while the polygon mesh can be fined-tuned to be as detailed as your computing power will allow, the dots/atoms/cubes/whatever are inherently disjoint so you need a bucket-load of them to present something that looks good.

It's not computationally efficient. It can be very powerful, but I think there are higher priorities given how good games look already.

24 frames of motion blur, that are roughly 2 frames per frame.

this one looks also promising:
http://www.atomontage.com/?id=dev_blog

You want virtual atoms? How about games like Worms? It is sort of built out of the same thing, except without the third dimension.

And as for Notch's calculation, he is sort of underrating the system, it would be more, as he mentions in his followup post, but there's a bit missing. Who says it all needs to be stored as voxels at that high resolution? It could work quite well as a hybrid. Data stored in polygons, translated to voxels when it needs to be, using something like pixel shaders to smooth out the polygons based on an algorithm depending on the kind of surface. If it retained the skeletal structure and translating to voxels with some transform data for deformation and such it wouldn't be as insanely heavy to animate, though how expensive the translation to voxels would be I have no clue.

If it's really as awesomely powerful as they claim it might be very nifty, but I'm fairly certain their engine has some rather crippling limits and I'd like to see something in action before believeing it. The idea of more focus on voxels might be nice though if for no other reason than better deformation and effects. Play something like Cortex command (I believe it was) and the dust you kick up would crop up somewhere. It's possible in 2D space because it takes so much less space, but in 3D space it'd still be insanely expensive, but it could be fun to play with.

Personally I'd still prefer to put gameplay over graphics. I've played jawdroppingly pretty games which have had me wondering how close I was to the ending because I was bored. And I've played downright ugly games which I haven't been able to put down because the gameplay is just so satisfying and entertaining. And this Voxel business seems to be touted mostly as "Look at how pretty we can make it", and honestly I don't care.

We're capped at about 25 the-eye-as-a-sensory-organ-wise.

The brain does heavy postprocessing to produce the final image though.