I forgot to add why the separate FP32 and FP64 floating point units are "bad" (from our PoV)... having the separate units just increase complexity, and NVIDIA G

I forgot to add why the separate FP32 and FP64 floating point units are "bad" (from our PoV)... having the separate units just increase complexity, and NVIDIA GPUs, for instance, have 4x the number of FP32 units compared to FP64 units, thus making their single precision numbers (which are what they typically advertise) 4x that of their double precision numbers.