GPU PROCESSING: HOW MUCH FASTER IS IT?

The engineers who develop Rocky DEM are continually striving for more efficient methods to accelerate the calculation speed. They know that the faster a simulation completes, the more productive Rocky DEM clients can be in their overall design process.

With the latest release of Rocky DEM, it is now possible to accelerate the computational workload by making use of the Graphics Processing Unit (GPU). This capability allows the calculations to be executed on the GPU while control and data transmission are processed on the CPU. For typical cases that include large amounts of particles, it is possible to obtain simulation results much faster than ever before. This is because the advanced capabilities of a GPU with 100+ cores are very well suited to DEM simulations.



PUTTING GPU AND CPU PROCESSING TO THE TEST

To find out just how much faster GPU processing could be, our engineers ran a test simulation through Rocky DEM three times on different types of processors—one CPU and two GPU—and then analyzed the difference in simulation time.

For this comparison, they used a 16-Core CPU simulation on two Intel E5-2690 processors, a GPU simulation on a NVIDIA Tesla K40, and another GPU simulation on a NVIDIA Titan Black. The simulation they ran featured one million particles passing through a transfer chute system over 10 seconds and with a time-step of 1.e-4 seconds. The set of particles included spheres as well as rounded polygons with three, five, and seven corners—a mixture particularly hard for the GPU due to divergence in the execution.

The baseline CPU simulation took just under 26 hours to complete. The results of the test show that the GPU simulations took between a third and a fourth of the CPU simulation time, depending upon the type of GPU.

Realistic Results
Relative Speed-Up of GPU Compared to CPU Simulations




Test simulation used in the benchmark, which included one million particles