I just got my first AMD Threadripper system (was using Intel before) and am trying to port efficient C++ parallelization fro INtel to AMD CPUs. Before I used either OpenMP or INtel THread Building Blocks, both work 3-4 times slower on AMD vs Intel (defeats teh purpose of getting AMD CPUs). Both Microsoft Visual Studio (Visual C++) and Linux. What C++ parallelizers are AMD professionals using for massive CPU-intensive computations?
@hyperstrings2003Thanks for reaching out. We understand you have run on Intel/AMD CPUs C++ code that is parallelized using either OpenMP or Intel TBB and are observing slower runtime on Threadrippers. Performance depends on various compile-time and run-time configurations.
Various parallelizing APIs can be used to deliver performance on AMD CPUs, including OpenMP, MPI etc. Kindly refer to User Guide provided at https://developer.amd.com/amd-aocc/ for various aocc-specific compile-time optimization options.
If you can share a reproducible code and relevant details that is showing degraded performance, we will be able to help you better.
Note that TBB is a proprietary technology of Intel codes with TBB need to be built with icc/icpc compilers.