2D GPU Solver Benchmarking
Flood Modeller's GPU-enabled 2D solver is amazingly fast, offering users significant time savings when running larger models.
We have undertaken extensive testing and benchmarking to make sure the 2D GPU solver meets the highest possible standards and is optimised for speed.
GPU hardware benchmarking
Using Flood Modeller's GPU solver and a NVIDIA H100 graphics card, to run an 8.9M cell 2D model with a 5m grid, model run times were 90 times faster when compared to running it on the CPU.
Flood Modeller's GPU solver doesn’t require high-specification graphics cards and provides significantly faster runtimes using a standard laptop GPU.
Using Flood Modeller's GPU solver and a NVIDIA H100 graphics card to run an 8.9M cell 2D model with a 5m grid, model run times were 90 times faster when compared to running it on the CPU.
GPU vs CPU
The data below shows the results from a range of different tests that we have completed, all using a standard specification computer that we use for modelling projects. There are some amazing time reductions to be achieved.
Results shown in graph are taken from a selection of different models (1D-2D, 2D only, event durations, boundary conditions, and timesteps) run by Jacobs, using Intel(R) Xeon(R) CPU E5-1620 v3 @ 3.50GHz 3.50 GHz (RAM 32.0 GB) and GeForce RTX 2080 Ti (Memory 11.0 GB). The data in the table is an indicative assessment of the simulation time reductions that may be achieved for the TVD solver based on these results.
The first dataset (Jacobs_BMTest_001) is for a small very steep urban catchment with both surface and urban drainage network components. The TVD solver is specifically designed to deal with the trans-critical flows that are likely to occur due to the very steep topography.
The second dataset (Jacobs_BMTest_002) is for a large relatively flat catchment which is modelling just the 2D domain. A 5m grid resolution has been used resulting in almost nine million cells. Either the ADI or TVD solver is well suited to this problem.
As you can see from our testing, the GPU significantly reduces the simulation time when compared to running either the ADI or TVD solver on the CPU.