Billions of dollars of product decisions use JS benchmark speed as one of the standard benchmarks to base buying decision on (for a good reason).

For machine learning speed compiling to the right CUDA / OpenCL kernel is much more crucial, so there's where the money goes.