Crocotta Research & Development Ltd

“Be ambitious of climbing up to the difficult, in a manner inaccessible...”

ONCE UPON ANOTHERFANTASTIC DAY IN THE UK…

…we started to think about visualization, other than polygonized surface rendering, to bring us closer to reality.

NEXT DAY WE TURNED TO OUR “DEMON”:

We quickly realized that lots of people had been dealing with the problem already*, so we had to set a more future oriented goal.

What if we traveled 10 - 20 years ahead in time, and mimicked the real environment with the equipment of the future as much as possible.

Neither the “demon” criticized our project, no relevant search results were found.

* happens just too often

The idea was born!

VIRTUAL UNIVERSEA virtual world solely made of particles

BUT, there are fundamental questions:A.What is universe?B.What bottlenecks do we need to face?C.Can we do any part(s) of the experiment on today’s machines?

We want to know•the building blocks•and their interaction rules.Standard model of physics predicts•12 fundamental particles•which interact via 4 elemental forces.Seems modelable, so far…

A. WHAT IS UNIVERSE?

is made of•2 up + 1 down quarks for a proton, •and 1 lepton which is the electron.BUT:6x1023 hydrogen atoms per dm3.Sounds less modelable…

HYDROGEN ATOM

1. 1014 atoms in a cell2. 109 cells per cm3

3. 3x1011 stars in a galaxy4. Observable universe

• Diameter is estimated at about 93 billion light-years.

• Contains 1024 stars (1 septillion stars).

• Approximate number of atoms is close to 1080.

Huh!!?

GIGANTIC NUMBERS ALL AROUND

Obviously we need to do some compromise here.Possible modeling options:•Stay on subatomic/atomic level and model nanostructures•Organic material provided that a living cell is the smallest element•Galactic phenomena and have starts/planets as smallest elements

B. WHAT BOTTLENECKS?

Simulation speed:•Particle interaction•Measuring, scanning (visualization for instance)Even if we imagined 100,000 parallel cores, with fast common memory access, petabyte storage devices, etc., we could always enlarge/expand our simulation scenario to make the hardware struggle again.Amount of data:Obviously we are forced to think in smaller scale, even in 10 - 20 years term, as the amount of data is enormous.

C. WHAT CAN WE DO ON TODAY’S MACHINES?

Well, probably a lot, because:•If we designed the system scalable, we could deal with the problem - in small scale - straight away.•Due to the enormous task we can’t solely rely on hardware performance growth.We need to invent better algorithms anyway.

Let’s start!

DEFINE AREAS OF DEVELOPMENT

We split up the work to 3 major areas:A.Scanning & visualizationB.PhysicsC.Data compression, representation

In the current presentation we focus on the “Scanning & visualization” part.

A. SCANNING & VISUALIZATIONPARTICLES IN 3D SPACE

We deal with many particles, so a raster representation may be more feasible than working with individual points (point clouds).

3D VOLUMETRIC TEXTURES

(similar to 2D textures + 1 extra spatial dimension)

Definitions:

•2D textures have pixels•3D textures have voxelsTexel means a pixel in 2D, a voxel in 3D.

Pros:•Easy to scale up/down•Opportunities for cheap interpolation, pattern reconstructionCons:•Difficult to scan, visualize•Large data-size (empty space is also stored)

Instead of conventional intersection testing in ray-tracing, we march forward in tiny steps along the ray.

RAY-MARCHING

Pro: Can access all texels/matterCon: Damn slow

ACCELERATED RAY-MARCHING

Spatial data structures, adaptive grids:•Binary-trees•KD-trees•Oct-treesBetter, but still not effective enough.

Sphere tracingdistance fields•The trick is to estimate the distance to the closest surface or sharp change in the volumetric texture at any point in space.•This allows to march in large steps along the ray.

ACCELERATED RAY-MARCHING

Possible replacement for particles?Provided field construction vs. ray-marching speed up is a win.Is that possible? Yes.We’ve been successfully deploying gradient fields, and not for visualization purposes only, but to accelerate physics calculations too.Further benefits:•Scale extremely well (down/up).•Give lots of opportunities for guessing, interpolating.

INTRODUCING GRADIENT FIELDS

Gradient fields can be well used for physics:•Distance fields.•Dramatic speed up at photon-tracing.•Force fields, like gravity.•Energy fields, like kinetic energy.•etc.

B. PHYSICS

Our failed approaches:•Lossless compression•“Conventional” lossy compression , like wavelet or similarCurrent approaches:•Adaptive representation•Focus on interesting areas•Contour & pattern analysis•Reconstruction

C. COMPRESSION

The figure below highlights that a compression method has to be deployed.Texture’s sidein texels

Size in bytesside3

1 byte per texel32 32K64 256K128 2M256 16M512 128M1024 1G2048 8G4096 64G8192 512G16384 4T32768 32T65536 256T131072 2P… …

We traversed an exciting path so far, and the next months are going to be even more exciting for us.We don't want to close out the possibility of 2 - 3 magnitudes speed up comparing to brute force methods, once we get all our theories into practice.And we hope our friends at the hardware department won’t rest either…

SUMMARY

To be continued…Thank you!

Crocotta Research & Development LtdSuite 5, 39 Irish Town, Gibraltar

We are a small team of international researchers with the aim of conducting technology leaps in exciting fields of exploration like virtual reality, virtual synthesis of matter, artificial intelligence, and robotics.

www.crocotta.co.ukcrocotta@crocotta.co.uk

+44 20 3239 7007