NVIDIA Took 20 Years to Ship 25 Lines of Python

20 lines of Python. 90% of cuBLAS. Zero thread management. The catch: NVIDIA owns the compiler.