Atom Rendering in video games.

it’s a scam!

Perhaps you’ve seen the videos about some groundbreaking “unlimited detail” rendering technology? If not, check it out here, then get back to this post: https://www.youtube.com/watch?v=00gAbgBu8R4

Well, it is a scam.

They made a voxel renderer, probably based on sparse voxel octrees. That’s cool and all, but.. To quote the video, the island in the video is one km^2. Let’s assume a modest island height of just eight meters, and we end up with 0.008 km^3. At 64 atoms per cubic millimeter (four per millimeter), that is a total of 512 000 000 000 000 000 atoms. If each voxel is made up of one byte of data, that is a total of 512 petabytes of information, or about 170 000 three-terrabyte harddrives full of information. In reality, you will need way more than just one byte of data per voxel to do colors and lighting, and the island is probably way taller than just eight meters, so that estimate is very optimistic.

So obviously, it’s not made up of that many unique voxels.

In the video, you can make up loads of repeated structured, all roughly the same size. Sparse voxel octrees work great for this, as you don’t need to have unique data in each leaf node, but can reference the same data repeatedly (at fixed intervals) with great speed and memory efficiency. This explains how they can have that much data, but it also shows one of the biggest weaknesses of their engine.

Another weakness is that voxels are horrible for doing animation, because there is no current fast algorithms for deforming a voxel cloud based on a skeletal mesh, and if you do keyframe animation, you end up with a LOT of data. It’s possible to rotate, scale and translate individual chunks of voxel data to do simple animation (imagine one chunk for the upper arm, one for the lower, one for the torso, and so on), but it’s not going to look as nice as polygon based animated characters do.

It’s a very pretty and very impressive piece of technology, but they’re carefully avoiding to mention any of the drawbacks, and they’re pretending like what they’re doing is something new and impressive. In reality, it’s been done several times before.

There’s the very impressive looking Atomontage Engine: https://www.youtube.com/watch?v=Gshc8GMTa1Y

Ken Silverman (the guy who wrote the Build engine, used in Duke Nukem 3D) has been working on a voxel engine called Voxlap, which is the basis for Voxelstein 3d: https://www.youtube.com/watch?v=oB1eMC9Jdsw

And there’s more: https://www.youtube.com/watch?v=xUe4ofdz5oI https://www.youtube.com/watch?v=lEHIUC4LNFE https://www.youtube.com/watch?v=Zl9CiGJiZuc

They’re hyping this as something new and revolutionary because they want funding. It’s a scam. Don’t get excited.

Or, more correctly, get excited about voxels, but not about the snake oil salesmen.


“But Notch, it’s NOT a scam!”

I’ve been getting a bunch of feedback that my last blog post is wrong for various reasons, and I’d just like to say that I would absolutely LOVE to be proven wrong. Being wrong is awesome, that’s how you learn.

If you want to read my reasoning behind various assumptions, click “read more”.

Why I assume it’s voxels and not point clouds:

* Voxels store only the information about each point, and their positions are implicit in the location of where the voxel is stored. Point cloud data stores both the information about each point and the position of each point.
* They mention “64 atoms per cubic millimeter”, which is 4*4*4 points per mm^2. While it’s possible they only refer to the sampling frequency for turning polygonal structures into point data, the numbers are just too round for me to ignore as a programmer.
* All repeated structures in the world are all facing the same direction. To me, that means they aren’t able to easily rotate them arbitrarily.

About the size calculation:

* I was trying to show that there was no way there was that much UNIQUE data in the world, and that everything had to be made up of repeated chunks.
* One byte per voxel is way lower than the raw data you’d need. In reality, you’d probably want to track at least 24 bits of color and eight bits of normal vector data per voxel. That’s four times as much data. It’s quite possible you’d want to track even more data.
* If the data compresses down to 1%, it would still be 1 700 three-terrabyte hard drives of data at one byte of raw data per voxel.

Animated voxels:

* Holy crap, people sent me videos of this actually being done!
* I was wrong!
* https://www.youtube.com/watch?v=tkn6ubbp1SE
* (But please note that just that single animated character runs at 36 fps)

Why it’s a scam:

* They pretend like they’re doing something new and unique, but in reality a lot of people are researching this. There are a lot of known draw-backs to doing this.
* They refuse to address the known flaws. They don’t show non-repeated architecture, they don’t show animation, they don’t show rotated geometry, and they don’t show dynamic lighting.
* They invent new terminology and use superlatives and plenty of unverifiable claims.
* They say it’s a “search algorithm”. That’s just semantics to confuse the issue. Sparse voxel octrees is a search algorithm to do very fast ray casting in a voxel space.
* They seem to be doing some very impressive voxel rendering stuff, which could absolutely be used to make very interesting games, but it’s not as great as they claim it is. The only reason I can see for them misrepresenting it this bad is that I assume they’re looking for funding and/or to get bought up.

If these guys were being honest with the drawbacks and weaknesses of their system, I’d be their biggest fan. As it is now, it’s almost like they’re trying NOT to be trustworthy.

All this said, voxels are amazing. So is raytracing and raycasting. As computers get more powerful, and storage gets faster and cheaper, we will see amazing things happen.

And a final word to the engineers who worked on this: Great job, I am impressed! But please tell your marketing department to stop lying. ;)