Selecting the Ideal Core Type for Your Job
Rescale offers users the flexibility to select from a range of core types to power their cloud simulations. However, for newcomers, determining the most appropriate core type for their specific requirements can often be a challenging task. For guidance, recommended hardware types for Computational Fluid Dynamics (CFD) simulations are outlined in the table below and different factors affecting hardware needs are discussed later on.
Follow these five steps to find the right core type for you:
-
Determine the typical parallel scalability limit using following equation
- Refer to the table below and identify the coretype with the closest match to the required number of cores. Note that Emerald offer configurations with 1, 2, 4, 8 and 18 cores per node and Luna 24 cores per node. Choose the coretype with the most cores per node instead of using multiple smaller nodes.
- Run your job using the selected coretype.
- Check if the hardware features of the selected coretype met your needs. Learn how to monitor Rescale hardware performance and costs for detailed instructions.
- For repeated simulations, run benchmark tests to find the most suitable core type for your needs.
Typical Coretypes for CFD simulations
| Coretype | Processor | Cores/node | Clock speed | Memory/node | Storage/node |
|---|---|---|---|---|---|
| Emerald | Intel Xeon Platinum P-8124 (Skylake) | 36 | 3.0 GHz | 144.00 GB | 1296 GiB |
| Carbon | Intel Xeon Platinum 8168 (Skylake) | 44 | 2.7 GHz | 352.00 GB | 700 GiB |
| Luna | 2nd Generation Intel Xeon Scalable Processors (Cascade Lake) | 48 | 3.0 GHz | 192.00 GB | 1728 GiB |
| Hematite | AMD EPYC 7V73X (Milan-X) | 64 | 2.2 GHz | 448.00 GB | 1920 GiB |
| Amber V2 | AMD EPYC 7742 (Rome) | 120 | 2.5 GHz | 480.00 GB | 960 GiB |
| Rozenite | AMD EPYC 7V73X (Milan-X) | 120 | 2.2 GHz | 448.00 GB | 1920 GiB |
Other considerations
Choosing the right coretype for cloud simulations is important for achieving optimal performance and cost-effectiveness. It’s not always about opting for the cheapest core hours; faster processors deliver results more quickly. This means you’ll need to purchase fewer hardware hours, and more importantly, it saves your employees’ time.
ANSYS Fluent exhibits excellent parallel scalability, meaning that more cores generally accelerate your simulations up to a certain limit. This limit typically falls in the range of 30,000 to 50,000 cells per core. However, the mesh size is just one of several factors influencing parallel scalability and solution time. Consider the following factors when choosing the appropriate core type for your simulations:
-
Solver Type and Methods: The simulation’s complexity varies depending on the chosen numerical behavior.
-
Physics Models and Boundary Conditions: More intricate physics require higher computational power.
-
Steady-state or Transient: Transient simulations often require more iterations due to time-dependent calculations and the solution time might be much longer
-
Mesh Adaption: The mesh size may differ significantly during the simulation.
-
UDFs, Calculation activities or other computationally heavy additions
-
Mesh Complexity: Alongside mesh size, the mesh type also impacts computational demands.
-
Memory Capacity: Ensure your chosen coretype provides sufficient RAM to avoid memory-related bottlenecks.
-
Consider GPU: ANSYS Fluent native GPU solver support some of the simulation models. If applicable, consider GPU solver and coretypes with GPU options for enhanced performance.
-
On-Demand Priority or On-Demand Economy: On-Demand Economy offers a lower unit price but run time and costs may increase if there are multiple stops during the calculation.