News for hackers. Stuff that matters.
Re:Train TensorFlow models faster and at lower cost on Cloud... by AlexCoventry on Wednesday December 12, 2018 @06:11PM (#18668532)
I thought TPUs were mostly used for inference, because they're low- precision. Has that changed?
Re:Train TensorFlow models faster and at lower cost on Cloud... by gradys on Wednesday December 12, 2018 @06:13PM (#18668546)
Yep. That was the first generation that was only for inference. Subsequent generations can do inference and training.