Radioss Solver Run Issue on AMD threadripper pro processor.
Hi Team,
We are trying to run a Radioss simulation, which is taking longer time on AMD processor compared to Intel. And also, when the same run is carried only on AMD processor with multicore usage, 32 cores is taking lesser time than 64 cores whereas to happen vice-versa.
For example, 32 cores AMD processor is taking 0.5 hr to run a simulation where 64Cores is taking almost 1 hr to solve the same solution.
we found some issue with the scalability.
Does anyone has come across this kind of issue ?
Any solution would help me.
Thanks
Nikhilesh
Find more posts tagged with
Hi Team,
We are using window 10 enterprise OS and and using compute console to solve the analysis. we have also disabled the KMP_AFFINITY. we have used every possibility mentioned in the Community Page.
The Elements are more than the minimum range.
Also please find the Image on details of AMD processor and also computation time for the model :
np | nt | Computation time |
---|---|---|
16 | 1 | 05:54:20 |
32 | 1 | 06:42:20 |
64 | 1 | 08:45:48 |
Hi @NIkhilesh
Could you please provide a screenshot of the Cumulative CPU Time Summary (information at the end of the 1.out file)?
Also, how does the runtime in SMP mode (-np 1 and - nt 16 (or 32/64)) compare to SPMD? And HMPP (-np 4 - nt16)?
Hello @NIkhilesh
Does your processor have 64 physical cores, or does it include logical cores as well? Could you provide details about your AMD processor?
Additionally, which operating system are you using, and how are you submitting the job (e.g. Compute Console)? If you are using Windows 10 but no the latest version, you migh need to set the KMP_AFFINITY environment variable to disabled.
In general, for parallelism settings, I would recommend checking this article: Altair Radioss Parallel Computing Methods.