How has the core pursuit of simulation evolved?
When I was younger, I thought simulation was about accuracy. Then I thought it was about speed. Then I realized it was about scale.
But today, in the age of AI and robotics, I believe simulation is about volume.
Let me explain.
Why does capturing real-world physics demand a multi-scale approach?
Imagine you want to simulate something simple: wind around buildings, a drone flying in a city, heat inside a data center, airflow around a robot, sound propagation in a room, smoke in a tunnel.
These are not toy problems. These are real-world physics problems. And the real world has a property that many people underestimate:
The real world is multi-scale.
A city is kilometers large. Buildings are tens of meters. Vehicles and humans are meters. Boundary layers and turbulence are centimeters. If you want to resolve the physics properly, not with rough engineering correlations but with real physics, very quickly you end up with hundreds of millions to billions of cells.
At that scale, simulation stops being a small engineering task. It becomes infrastructure.
What computational barriers define traditional billion-cell simulations?
A billion-cell simulation typically means:
- ~100–300 GB of memory
- ~10¹⁵ floating point operations
- Traditionally: HPC clusters, hundreds or thousands of CPU cores, days or weeks per run
This is why, for a long time, high-fidelity simulation was rare, expensive, and slow. So rare that most industries learned to live without it.
What drives the transition from single runs to high-throughput data generation?
But something has changed.
We are no longer running one simulation for one design.
We are now running thousands, millions of simulations to generate data for:
- Physical AI
- Robotics
- World models
- Digital twins
- AI for Science
So the question is no longer:
Can we run one very large simulation?
The real question is:
Can we run a very large simulation a million times?
This is why billion-cell scale matters. Not for academic benchmarking. Not for pretty CFD pictures.
But because at that scale, you are no longer simulating a part. You are simulating the world.
And when simulation becomes world-scale and high-throughput, it stops being just “CAE”.
It becomes data infrastructure for the physical world.
This is the shift we are seeing now: From accuracy → speed → scale → volume.
This is also what we are building at EXDYNA.
— Data infrastructure for science, engineering, AI and robotics.
探索十亿级单元，城市级风场模拟演示