How to Choose Hardware for your Simulation
How to decide on the Number of Cores ?
The number of cores are generally decided based on the model size, model fidelity and expense.
- Model size: Cells, grid points, nodes, elements, degrees of freedom
- Model fidelity: Time accurate, Single vs Double precision and Solver
- Cost focus: Minimize dollar cost per solution
- Cost-performance focus: Good balance between time to solution and computing cost
For example, CFD generally start with 75K - 100K cells per core
A good rule of thumb is to initially utilize a smaller number of cores to test your model before running the full intended core count
List of Best Core Types
|Optimal cost vs performance||Onyx
| 60 GB
| 648 GB
|High Memory specialty Core types||Zinc
| 488 GB
| 1.3 TB
|Premium Specialty Core types||Topaz
| 2.0 TB
| 3.6 TB
More details on the number of cores per node, storage, memory and pricing can be found on Rescale platform's Hardware Summary page. For example, in the image below we can see that Onyx has 18 cores per node with 60.0 GB memory and 648 GB storage per node, while a specialized core type like Melanite has more memory, storage and cores per node. These details will help determine the best core type for your simulations.
Software <=> Core Type Recommendations=>
For CFD/FEA based simulations
For FEA codes with higher memory requirements
- Topaz (for higher memory requirements)