GPU Acceleration in Optistruct Solver

Noah Ray
Noah Ray Altair Community Member
edited January 2021 in Community Q&A

Hi Everyone, 

I am writing this post to inquire about GPU acceleration in the Optistruct solver. 

For context:

Currently running OS 2019.3 (this could be updated if need be). 

CPU specs:

AMD EYPC 7742 (64 core @ 2.25 GHz)

1024 GB RAM

NVIDIA TESLA V100 (32GB) [QTY:2]

 

Over the last two weeks, I have been benchmarking this PC to determine the effects of GPU acceleration. So far, I have been running simple static structural analysis (MBB) as the benchmarking model. Particularly, I am interested in solving jobs as quickly as possible. The jobs I typically used to benchmark are 1.4M element, 2.0M element, and 3.5M element. These job sizes roughly equate to 10%, 20%, and 40% RAM demand without GPU acceleration. 

My question is regarding properly setting up the GPU acceleration for Optistruct. So far, the GPU acceleration has actually slowed down the larger jobs. For example, running the 3.5M element model with only CPU + RAM will take 2 hours.... Running the 3.5M element model with GPUs will take 4 hours. 

Possible solutions I have tried:

- downloaded CUDA (dont know if anything specific needs to be configured)

- tried a variety of GPU commands (ie. -gpu -gpuid -ngpu)

 

Does anybody have any experiences using GPU acceleration in Optistruct with positive results, have any configuration recommendations, and/or see anything I may be doing wrong?

 

Thanks!

 

Tagged:

Answers

  • Q.Nguyen-Dai
    Q.Nguyen-Dai Altair Community Member
    edited January 2021

    I know nothing about GPU.

    But I just have a question: to comparer CPU vs GPU do you run with the sam memory parameter?

    If 64-core system is not enough for you, go for HPC Cluster 128 or 256 core.

     

Categories