Mobile devices use fp16 often yes. The Switch chip was actually the first time Nvidia did this too, and claimed it was the first 1TFLOPs mobile chip, but this is only fp16, and with Switch clocks it is 384gflops and 768gflops in full fp16 mode (which couldn't be used for games)
Realistically though, these GPUs are able to accomplish a mixed precision and developers have reported as much as 70% of their code being capable of running in fp16, meaning that you'd get a 30% FP32 gflops number and then a 70% FP16 number, so the 196gflops portable Switch would max out around 333gflops of Maxwell+ performance, while it's docked performance would be twice that. These are important numbers as engines start utilizing this Technic more and more, because PS4 and XB1 base consoles and XB1X do not support FP16 modes, making this a push for parity, with mixed precision the Switch comes in ~25% of the PS4's base performance in handheld for best case scenarios, while docked performance is ~50%, and it's even better when compared to XB1. Nintendo and Nvidia could have done a better job, but it's certainly better than Nintendo and AMD did with Wii U.