Learning Forum

[marvinator]

Hi all,nafter searching for a reason why my Ryzen-based System is way slower than an older Intel Platform, I stumbled across the seemingly well known issue of the Intel MKL crippling AMD Hardware.nI tried to use the often described workaround of setting an environment variable with the name and value MKL_DEBUG_CPU_TYPE=5 . I directly set it via windows Powershell and used Mechanical for computation of the result, but there was no difference in speed. I also set it via windows cmd before starting Mechanical ADPL in batch-mode. That also didn't make any difference in compute time. nThe MKL Library used by Mechanical 2020R2 is Intel(R) Math Kernel Library Version 2020.0.0 Product Build 20191125 . The tweak should still work with this version of the MKL, right?nHas anyone an idea, why the tweak doesn't work (anymore)?nCheers,nMarvinn

[mattCED]

I'd also like to know this! This stuff is over my head a bit but I wish Ansys would be clearer on Zen limitations in their literature.nI see there might be some movement from Intel + AMD on future integration but not sure when that'll be.nn

[marvinator]

Hi Matt,nI kept trying different strategies and finally found a solution that works for me. At first I downloaded Ansys Mechanical 2019R1 and launched the APDL via batch-file again using the input-file I created in Mechanical 2020R2. I achieved the same way too slow compute rate as before.nAfter replicating the project in Ansys workbench 2019R1 and calculating the solution via the workflow, the MKL tweak finally worked! Even opening the 2019R1 project with 2020R2 and calculating it via workbench 2020R2, showed the same immense speedup by a factor of approx. 3, outperforming the XEON by about 30%!!nI'm pretty sure that the issue is buried somewhere in the solver input file...n