I am writing to ask you for advice in configuring a workstation with Nvidia GPU for running Abaqus.
We want to use the GPGPU acceleration feature of Abaqus with a workstation with the following configuration:
- CPU: Dual Intel Xeon Gold 6242R 3.1GHz,(4.1GHz Turbo, 20C, 10.4GT/s 2UPI,35.75MB Cache,HT
(205W) DDR4-2933 2nd)
- RAM: 192GB 12x16GB DDR4 2933MHz RDIMM ECC Memory
- GPU: Dual NVLink Nvidia RTX A5000, 24GB, 4DP
We ran some analysis with about 3 Millions of DOFs, but we cannot see any improvement in the speed of analysis. We ran an analysis with softening plasticity model without a user subroutine, We also ran phase-field fracture using the UMAT subroutine, and the results are the same. We do not see any acceleration performance with GPU.
The Intel oneAPI Toolkit has already been installed with all components. Also, CUDA Version 11.7 was installed. When an analysis is run with GPU, it is possible to see the message “GPU SOLVER ACCELERATION ENABLED” in the message file. Also, it can be seen that Dedicated GPU memory usage starts to increase to 90% of GPU memory, but CUDA will be increased a bit at the first analysis and then will be zero.
We have already add below command to custom_v6.env:
- os.environ[“ABA_ACCELERATOR_TYPE”]=“PLATFORM_CUDA” # Nvidia
However, we cannot see any performance by CUDA.
We appreciate it if you advise us on how we can benefit GPGPU computing with this machine. We are also happy to provide more information if needed.