[jonas21]

Yeah, I wouldn't be surprised if the majority of code performing large matrix multiplications these days was written in Python and executed on GPUs by libraries like Tensorflow and PyTorch. With the right abstractions, programmers can be "lazy" and still get great performance.

[srg0]

Matrix multiplication is usually done by a platform-specific BLAS library (BLAS is an API, there are multiple implementations, e.g. Intel MKL, OpenBLAS, cuBLAS). There are some other linear algebra APIs/libraries, but this is what's used the most.

Most of the numerical code that cares about performance for linear algebra uses this API and links an appropriate implementation.

[otabdeveloper2]

The 'written in Python' you speak of is actually Fortran under the hood.