We encourage all our users to optimize their code. Optimizing the compilation of your code may dramatically increase code performance.
When optimizing your compilation, not only will you get your computation results faster and potentially reduce the cost of your research, but you will also reduce the energy footprint of your project.
Information on how to optimally compile your code can be found in our documentation. See for instance:
- Our articles on building scientific software
- Using Intel MKL on AMD
- Efficient memory utilization on Betzy
- Efficient processor/network utilization on Betzy
A more in-depth read on processor performance can be found in the PRACE best practice guide for modern processors.
We especially encourage users who are migrating from our Intel-based machines (Saga and Fram) to our AMD-based machine (Betzy) to have a look at the documentation, as there might be some performance differences using the various compilation flags.
Note for instance that compiler flags involving on Betzy involving -xavx, -xavx2 and -xcore-avx2 can cause problems, the -x part of the flag generates code exclusively for the processors that support these instruction set and cause Intel to insert a test for just this particular set of Intel processors.
As an example, an S3D user who recently migrated to Betzy saw a 30% increase in performance by optimizing the compilation of their code.