I am very new to deep learning and recently tried to set up a tensorflow GPU environment with 1080 ti in Windows 7 64-bit. However, I am having issues running my tensorflow GPU setup. This is what I did:
(1) Installed latest Anaconda distribution with Python 3.6 in Windows 7 64-bit.
(2) Installed latest Nvidia driver.
(3) Installed CUDA toolkit version 8.0. Continued with setup when prompt with not recognised driver (1080 ti driver is newer but will still work under CUDA toolkit version 8.0). Also set path environment to CUDA toolkit directory.
(4) Installed cuDNN version 6.0. Copied the files into CUDA toolkit directory (Tensoflow 1.4 only supports CUDA toolkit 8.0 and cuDNN 6.0).
(5) Installed Tensorflow GPU and Keras. Used pip install for both.
(6) Tested Tensorflow in Python. import tensorflow as tf , then sess = tf.Session(config=tf.ConfigProto(log_device_placement=True)). Got message showing GPU is detected as Nvidia 1080 ti.
(7) Went into Spyder (my preferred Python editor). Tried to train a simple two layer ANN network as test. SLOW AS SNAIL. Sometimes even crashes. Same training took 30 s on CPU.
I have no idea what went wrong. Any of you guys have success running 1080ti with tensorflow in Windows? Would like to hear the setup process. Thank you!!!