“Why can’t I run PreonLab on my GPU?” We often heard this question and didn’t have a good answer. But behind the scenes, we have been experimenting with GPUs for quite some time. We are excited to announce that we are finally ready. PreonLab 6.0 is here, and it changes everything. This is how:
This is just a selection of new features and improvements. Check out the changelog to learn about all the changes.
PreonLab, PreonCLI and PreonNode now support GPU-accelerated simulations using Nvidia CUDA. The new CUDA-based implementation is not a simple port but instead uses data structures that are specifically optimized to get the most out of GPU architectures. At the same time, we took great care to ensure that the solver math matches the CPU version. This means that both implementations also produce the same simulation results.
The figure above shows a case study in which a 6x performance improvement is achieved over a traditional multi-socket system equipped with 64 physical cores (128 threads). What makes this speedup even more impressive is that the GPU is rated for a Thermal Design Power (TDP) of 300 Watts, while the two CPUs combined require 360 Watts. This is a powerful example demonstrating the clear advantage in performance and energy efficiency that GPUs currently enjoy over CPUs. Most importantly, it means that a task that previously required close to a week now runs within a day. Will you use this time to iterate faster, or will you use it for even more accurate and longer simulations? In any case, a world of possibilities suddenly opens that was not there before. We think you will love PreonLab on GPU just as much as we do!
Read more about the simulation performance boost with GPU here.
Please note that PreonLab 6.0 does not support all features on GPU yet. We plan to deliver most missing features such as CPS with the 6.1 release later this year. To learn more, read the Manual or contact our Support Team. PreonLab will also check for unsupported features when starting a simulation.
The new experimental lateral adhesion model enables you to model the effect of the contact angle hysteresis. The feature was designed to keep droplets on inclined surfaces by means of a static resistance force. A good example of this is the windshield wiper simulation below. In general, the new model can greatly increase realism in rainwater management scenarios such as tailgate runoff simulations.
GPU support has finally arrived, and it is here to stay. But PreonLab 6.0 also has plenty of improvements to offer if you are not using GPUs. The most notable additions include the new welcome screen, a CPS-optimized fill mode for the Volume Source and Pathline mesh export.
Check out the changelog for a full list of changes. To learn more about the new features, have a look at the updated manual. We hope you will enjoy working with PreonLab 6.0 and are looking forward to your feedback.
Want to know more about what you can expect from PreonLab?
Have a look at these clips!