I recently purchased a 2080 Ti and am excited to use the fp16 tensor cores. I’ve installed a version of Tensorflow 1.12 built to use CUDA 10 and CuDNN 7.3.1. The performance is much better than my previous Titan X, but I want to confirm TF is using the fp16 tensor cores. Is there a way to check?
Just to clarify, I am using tf.float16 for all the placeholders and weights in my network.