Or you can flip your assumptions and run 10X faster... everywhere. It was a tough investment, but we went GPU-first via Apache Arrow & Nvidia RAPIDS. (In this case, I'd look at RAPIDS / NVStrings before this.)