I think since Float16 type will be converted to Float32 during a 
calculation, there's no point to stock an array of Float16 (it will only 
slow down the computation), unless the main memory is highly demanding.

Reply via email to