3.5 Simulation Times
The simulation time of a model is dependent on several factors including:
- The area to be modelled
- The cell size of the model
- The number of cells that are wet
- The duration of the simulation
- The solution scheme (TUFLOW Classic or TUFLOW HPC)
- Computer Hardware (CPU or GPU)
Each of the factors listed above should be considered during the development of a model.
A simple spreadsheet has been created to estimate TUFLOW Classic simulation times based on the cell size, catchment area and model run time (i.e. event duration). The spreadsheet can be downloaded from the TUFLOW Wiki Estimating Runtimes. This spreadsheet is a basic guide, as it has several simplified assumptions, including the timestep, number of calculations per second the computer can run and fraction of wet cells (this will change throughout the simulation). Acknowledging these assumptions, it is still a useful guide. further consideration of the model design is recommended if the spreadsheet is estimating excessively long simulation times.
The speed at which TUFLOW simulations will run is dependent on the hardware used. Newer and higher CPU frequency computers will run faster than older lower CPU frequency computers. A hardware benchmark model is available from the TUFLOW Wiki Hardware Benchmarking. This page also contains the simulation times for the same model on a large range of computers and GPU devices.
There are several other secondary factors that can affect the speed performance of TUFLOW simulations, including:
- The latency to the location where results are being written. For fastest speed, write results directly to a local computer hard drive. Lower latency associated with writing to an external hard drive can slow a simulation down significantly.
- How frequently output is written, with more frequent output slowing the simulation.
- The output format(s) selected. XMDF and DAT formats are quicker to write than grid raster output types (TIF, FLT and ASC). Use XMDF format for temporal Map Output. Reserve grid raster output formats for maximum results.
- Whether the single or double precision version of TUFLOW is used - the single precision version will be faster. This is discussed in Section 13.3.2.
These factors are typically less significant than the primary factors listed at the start of this section (model size, cell count etc.).
As outlined in Section 12, the GPU Module may be utilised to increase simulation speed using TUFLOW HPC. The speed up ratio compared to CPU varies depending on model particulars and the hardware used. Nevertheless, the typical range in performance increase experienced using current technology (at the time of writing) is a 10x to 50x increase in simulation speed.