I would actually like to see a benchmark that there solving multiple linear systems can be efficiently parallelized. And I don’t think it’s numpys fault that notation for tensor operations is hard.