I am running various Abaqus simulations with a model size of about 1.5 million degrees of freedom in total. To speed up the calculations I am trying to decide what number of CPUs would be optimal and what are influencing factors (like model size, steps, time steps, outputs, hardware etc.). I'm interested in the question at what number the writing and merging of partial results and data between the different cores outweighs the benefit of using multiple CPUs.
Dear Jens,
To speed up the calculations, you may use the mass scaling factor also in Abaqus explicit. I mostly use 10 core to speed up my simulations and it gives appropriate results. Now, there are multiple influencing factors like - model size (the number of degrees of freedom), size of the solution step, etc. You may play with the time step to reduce the size of the solution step.
Hi Jens,
It really depends on multiple factors including, but not limited to, the nature of simulation, element types used, and of course, hardware! Assuming you have got the model and solution technique already optimized, if we just focus on hardware for computational capability: I'd try with about 8 to 12 core CPU (providing 16 to 24 logical cores when multi-threading is enabled) w/ 2.0+GHz base clock-speed. I'd also make sure to have at least 32GB of RAM (memory), NVMe M.2 SSD (for faster data transfer during analysis), and enable GPGPU computation w/ dedicated GPU.
- Farhan
Hi,
Please check out this website for any HW/SW/Workload standards.
http://spec.org/
Regards.
" I'm interested in the question at what number the writing and merging of partial results and data between the different cores outweighs the benefit of using multiple CPUs"
A classic scaling question that can apply in many environments. The answer will depend partly on the hardware ( how fast are the cpu interconnects/ shared memory/ semaphores) but mostly on how good is the software at scaling up. Matureware such as Abaqus likely to be very good at using what ever resources you throw at it as it has been doing multitasking since the early '80s on Cray machines.
Your problem and solution may vary.
Have a look here https://downloads.dell.com/manuals/all-products/esuprt_solutions_int/esuprt_solutions_int_solutions_resources/high-computing-solution-resources_white-papers102_en-us.pdf
Surprisingly though the HPC Advisory Council studies seem to be a bit out of date
https://www.hpcadvisorycouncil.com/best_practices.php#Abaqus
Abaqus Standard tends to have benefit of number of cores start to level off significantly after 16 cores. More is always better though for larger problems. Using the MP_HOST_SPLIT parameter and setting it to an integer that can be evenly divided into the number of cores or sockets will help immensely, more so than even GPU computing. GPU computing can help in larger problems, but you need number of GPUs installed in a machine equal to MP_HOST_SPLIT parameter. Set No of GPU's always equal to 1, and have no of GPUs installed equal to MP_HOST_SPLIT. If MP_HOST_SPLIT = 1 you'll get poor performance, and a single GPU can help a lot then. But with no GPU and MP_HOST_SPLIT set to the proper number (say 2 or 4 for 16 cores) you'll get a good speed up, better than an single GPU and MP_HOST_SPLIT=1. All the hype on GPU computing is totally overblown, believe me I know. I have 3 GPUs installed and I still don't get much improvement, if at all. Spent many an hour with the Abaqus people on this. All of the benchmarks are false because they use MP_HOST_SPLIT=1 for those benchmarks, which no one would ever do cause it's too slow. Just set MP_HOST_SPLIT equal to the proper number and you'll do better than most GPU computing. Did I mention they don't support or document this parameter much? Silly. For Abaqus Explicit, that is a different animal, and more cores is always way better for larger problems, and run times still scale nicely with number of cores.
Good luck,
Steve
- Badges
- Science topic
- Similar topics
- Filing