EDEM Particle Simulation: CUDA GPU and Precision Modes

Cesar Barreto
Cesar Barreto
Altair Employee
edited April 25 in Altair HyperWorks

EDEM CUDA precision modes

You can speed up your simulation with the CUDA GPU solver precision modes. All EDEM particle shapes (sphere, Multi-sphere, Polyhedral and Sphero-cylinder) are supported. 

The GPU CUDA solver has the option to select the precision of floating-point numbers used to represent simulation variables which affect simulation results. There are 3 options available:

  • Double Precision
  • Hybrid Precision
  • Single Precision

EDEM precision modes were introduced in EDEM 2021.2, however by using the latest EDEM 2022.0 you can run Multi-sphere, Polyhedral or Sphero-cylinder particle simulations using your Nvidia Graphics cards.  All you have to do is to install the latest EDEM version available on the Altair community Marketplace (altairone.com), and test your Nvidia card’s compatibility by running the GPU device test available in EDEM.

 

What are the main difference between precision modes?

  • Double precision has the highest numerical accuracy with the slowest solver speed.
  • Single precision has the fastest solver speed with the lowest numerical accuracy which is not suitable for all applications.
  • Hybrid precision uses a combination of single and double precision which provides a good balance between numerical accuracy and speed.

 

Hybrid precision not only provides faster simulations but also reduce memory usage. This is an advantage when running large scale simulations which contain millions of particles that might run out of memory. 

In conclusion hybrid precision gives the best balance of speed and accuracy.  If you’re interested in GPU CUDA precision modes, when to use them, and how to set a precision mode for all your EDEM simulations, check our Introduction to EDEM eLearning GPU CUDA - Lesson 15. 

 

Check out the links below for additional information on EDEM GPU:

10 Common Questions about EDEM GPU Answered

How to select a GPU (Graphics Processing Unit) for EDEM

Tagged: