You're thinking of Nvidia's GP100 chip, which does run FP16 twice as fast as FP32. GP100 is only used on the Tesla P100 and Quadro GP100 cards though. The Tita

You're thinking of Nvidia's GP100 chip, which does run FP16 twice as fast as FP32. GP100 is only used on the Tesla P100 and Quadro GP100 cards though. The Titan X Pascal (and 1080ti) uses the smaller GP102 chip, which as znfi pointed out is practically useless for FP16.