From the Lab to Research
Learning objectives
- Estimate the cost of a modelling run
- See how grid size and steps drive compute and memory
- Watch 2D-to-3D and acoustic-to-elastic explode the cost
- Match a run to a browser, a workstation, or a cluster
Sizing the Jump
The browser Lab runs small, cheap models in an instant. A research run is a different animal: a large grid, thousands of time steps, and an engine that touches every cell many times over. This calculator makes that difference concrete. Choose a grid size, a step count, and an engine, and it estimates the cell count, the memory, the total floating-point work, and the wall-clock time on three platforms, a browser, a workstation, and a cluster.
The Cost of Dimensions and Physics
Two moves dominate the numbers. Going from 2D to 3D does not double the cell count, it cubes the side length, so a grid that was a million cells becomes a billion. And going from acoustic to elastic multiplies the work per cell and the fields in memory, because there are more wave modes and more stiffness terms to update. Together they turn a run that is instant in the browser into one that needs a cluster and gigabytes of memory. Watch the memory line in particular: a full-scale 3D elastic model can exceed what a workstation holds, and that memory wall, as much as the compute, is what forces the jump to dedicated hardware.
This is fit-for-purpose one last time, now measured in compute. It is why the course built the cheap engines first, so you could learn the physics interactively at a scale that runs in milliseconds, and why it hands the heavy engines to Python and the HPC codes for the runs that truly need them. Model at the smallest scale that answers your question; scale up only when the answer demands it. The final section closes the course.