For years, Intel’s compiler, math library MKL and their profiler, VTune, really only worked well with their own CPUs. There was in fact code that decreased performance if it detected a non-Intel CPU in place:
That later became part of a larger lawsuit, but since Intel is not discriminating against AMD directly, but rather against all other non-Intel CPUs, the result of the lawsuit was underwhelming. In fact, it’s still a problem today:
Given that the MKL is a widely used library, people also indirectly suffer from this if they buy an AMD CPU and utilize software that links against that library.
As someone working in low-level optimization, that was/is a shitty situation. I still bought an AMD CPU after the latest fiasco a couple of weeks ago.
For years, Intel’s compiler, math library MKL and their profiler, VTune, really only worked well with their own CPUs. There was in fact code that decreased performance if it detected a non-Intel CPU in place:
https://www.agner.org/optimize/blog/read.php?i=49&v=f
That later became part of a larger lawsuit, but since Intel is not discriminating against AMD directly, but rather against all other non-Intel CPUs, the result of the lawsuit was underwhelming. In fact, it’s still a problem today:
https://www.extremetech.com/computing/302650-how-to-bypass-matlab-cripple-amd-ryzen-threadripper-cpus
https://medium.com/codex/fixing-intel-compilers-unfair-cpu-dispatcher-part-1-2-4a4a367c8919
Given that the MKL is a widely used library, people also indirectly suffer from this if they buy an AMD CPU and utilize software that links against that library.
As someone working in low-level optimization, that was/is a shitty situation. I still bought an AMD CPU after the latest fiasco a couple of weeks ago.