EDEM 2021.2 GPU benchmark

phamad
phamad Altair Community Member
edited March 2022 in Community Q&A

Hello EDEM team,

I am currently conducting a survey of hardware (CPU, GPU) for simulations of a large number of particles. I have a chance to test EDEM 2021.2 on a system with GPU RTX A6000, and here is the result:

 

image

 

As you can see, If I select CUDA solver single precision, the performance is the highest. It is much higher than that of OpenCL.

My questions are:

1. Does this result reflect the current performance of the EDEM code?

2. It seems like the CUDA solver has not supported multi-GPU yet, so if I use OpenCL solver and want to gain the performance as the case of CUDA FP32, I have to use a multi-GPU setup, right?

 

Thank you very much for your time!

Regards,

Tagged:

Best Answer

  • RWood
    RWood
    Altair Employee
    edited March 2022 Answer ✓

    Hi,

    1. I would say it reflects some of what we already know, in terms of the ordering you found. We currently find the CUDA solver to be much faster than the OpenCL solver, and the ordering of single, double and hybrid also agrees with what I'd expect. 

    2. What you say is correct but CUDA multi-GPU should be available in the next release (EDEM 2022).

    Cheers,

    Richard

Answers

  • RWood
    RWood
    Altair Employee
    edited March 2022 Answer ✓

    Hi,

    1. I would say it reflects some of what we already know, in terms of the ordering you found. We currently find the CUDA solver to be much faster than the OpenCL solver, and the ordering of single, double and hybrid also agrees with what I'd expect. 

    2. What you say is correct but CUDA multi-GPU should be available in the next release (EDEM 2022).

    Cheers,

    Richard

Categories