0

I'm using Win 10 Pro 64bit. I have already:

  • installed latest graphics card drivers for GTX 1650 SUPER (w/ CUDA 11.2.109 driver)
  • installed CUDA drivers
  • installed Anaconda and activated environment
  • installed Python 3.8.0
  • installed Tensorflow 2.3.0
  • installed cudatoolkit 10.2.89, (then 11.0.221 later)
  • installed cudnn 7.6.5.32 to relevant folders
  • Checked PATH entries

nvidia-smi shows the card.

Yet, Python import tensorflow as tf print(tf.reduce_sum(tf.random.normal([1000, 1000]))) returns tf.Tensor(1348.6893, shape=(), dtype=float32)

tf.config.list_physical_devices('GPU') returns []

tf.test.is_built_with_cuda() returns False

What am I missing?

I have this working on my friends' 980ti seemingly with the same settings.

Braiam
  • 1
  • 11
  • 47
  • 78
Michael Lanzon
  • 1

1 Answers1

0

Nevermind...

installed Python 3.8.0 & installed Pillow (Don't know why. Thought I installed it last time I guess.) installed "tensorflow-gpu=2.3 tensorflow=2.3=mkl_py38h1fcfbd6_0" as recommended at Tensorflow-GPU not using GPU with CUDA,CUDNN

and it worked.

Michael Lanzon
  • 1