It’s been almost four years since I first published a blog on run times for 2D simulations conducted on GPU cards back in December 2012. Since that time, NVIDIA have continually evolved their GPU technology. By comparison, over the same period of time, the CPU world has gone rather flat in terms of outright speed improvements.
InfoWorks ICM is able to leverage technological improvements as new GPU cards come onto the market. The tables below show runtimes for two of our standard 2D testing models. Each model was run on NVidia GeForce 970 and 1080 cards, on an NVidia Quadro K5200 card and on a TESLA K20c card.
The test models contain 1.6 million and 3.3 million 2D elements and are 2D only models (i.e. no pipes or manholes). Each run is a 2hr simulation using a M5-60 design storm. Rainfall is applied directly to 100% of the mesh, so all elements in the 2D zone are wet throughout the simulation (this is important, because it’s the number of wet elements that most influences the runtime, not the overall number of 2D elements in the model).
My thanks go to the Modelling Team at Atkins (UK), who kindly offered to conduct test runs using our standard GPU Testing Database on their GeForce GTX970 and GTX1080 cards to provide the data shown in the tables above.