Building a more efficient foundation for real-time simulation

Trying to simulate the chaotic complexity of physical forces in real-time is nearly impossible without massive computing power. In large part this is because the computational data is represented in a fundamentally inefficient way for modern technologies. It is based on a 50-year old technological foundation and ideas which were never meant nor designed to be used at todays scale. They demand calculations of many separate system configurations for the same object usually multiple times per frame.

We are building a new type of data representation that will unify all descriptions and act as a better foundation for real-time physics simulation/visualization software. It enables reproduction and manipulation of physical forces on consumer grade gear for scientific, entertainment or business ends without the need for massive dedicated hardware architectures. The results are richly simulated virtual worlds previously conceived only in the realm of science fiction.

Quantum Neural Models

Our product is an application framework which can be integrated into any existing physics and/or rendering software that's used offline, real time or in the cloud. Instead of using billions of current primitives like polygons or voxels to define an object, it uses neural networks to model a probabilistic distribution of a primitive's state similar to how the quantum wave function mathematically describes a quantum state of a system.

High level repercussions

- a radical increase in general performance combined with fitting up to 10X more data into memory

- up to 100x more complexity in 3D simulations or visualizations for the same processing demand

- 4x to 6x reduction in the content size of software running 3D data simulation or visualization

For engineers

- Easy integration

- Geometry compression

- Polygonal to QNM model converter

- Custom ray intersection test

For artists

- Unlimited polygon counts

- Unlimited texture size

- No need for LODs

- Real-time path tracing quality

Low level use cases

- Path tracing

- Ray tracing

- Rendering engines

- Physics engines

High level use cases

- Video games

- CGI

- Scientific simulations

- Aerospace

Models are encoded with arbitrary number of dimensions, enabling animation, simulation, materials etc. It also features adjustable lossy model compression and reduces ray tracing complexity by multiple orders of magnitude

Daisyfield

We're developing QNM (Quantum Neural Models) in harmony with our rendering engine Daisyfield, the first engine to use QNM. For more information click here.

For additional information, please see our talk at Nvidia GTC 2019 / Silicon Valley: S9367 Beyond Polygons, Voxels, and Rasterization.