https://hgpu.org/?p=17870
Investigating Half Precision Arithmetic to Accelerate Dense Linear System Solvers