Hello all
I tried to perform a benchmark test for CPU performance. I have Intel(R) Xeon(R) Gold 6134 CPU @ 3.20GHz 3.19 GHz (2 Processors) and 256 GB Ram
Below is the model I used for the test
The result I obtained were pretty strange.
Please let me know if you guys also obtain the same results.
Regards
Rishabh
This model has 197456 nodes. Your CPU seems to have 8 physical cores so the benchmark should basically end at that point. Hyperthreading is not used here.
Kerne = cores, Logische Prozessoren = logical processors
I have 16 cores
there’s no information about the solver used, i guessed is default PaStiX. Different set up may lead to speed inconsistently, maybe it can try by setting OPENBLAS_NUM_THREADS in environment.
Oh sorry I forgot to write that I used Pardiso Solver.
Should I set something like “OPENBLAS_NUM_THREADS=4” ? or just the same text as you wrote?
it seems Pardiso solver have different setting, using OMP_NUM_THREADS instead. In PaStiX the optimum number setting is well known not the number of threads or processor, can be one or half number. Maybe this condition apply on Pardiso also.
I can say that for 14900k with E-Cores the best is to active E-Core in BIOS but to use only 24cores in the settings for Pardiso solver.
https://www.intel.de/content/www/de/de/products/sku/236773/intel-core-i9-processor-14900k-36m-cache-up-to-6-00-ghz/specifications.html
Thank you for the suggestion. I used the environment variable in following way.
Should I set anything for number of processor or would it be overwritten by the environment variable. As I have set 12 as number of cores I also observed that solver was using 12 cores
it seems number of cpu is set to equal to threads by CalculiX, try using NUMBER_OF_CPUS whenever needed and fixed number of threads to make it consistent, probably.