Closed lebedov closed 6 years ago
Your environment is a mixture of packages from conda-forge
and defaults
and some pip installed packages. These two channels are not always compatible. Can you try creating an environment using just the packages in defaults
and see if the issue persists? conda create -n tflow_test -c defaults python=3.6 tensorflow-gpu=1.8.0
should create such an environment.
Yes - it appears that the conda-forge and default packages are clashing and the former probably don't support GPUs because this issue is still open; using the default channel only, I can see my GPU with TensorFlow 1.8.0; in contrast to what I posted earlier, the environment created with default-only tensorflow packages contains
conda list t.*flow
# packages in environment at /home/lebedov/miniconda3/envs/tflow_test:
#
# Name Version Build Channel
_tflow_180_select 1.0 gpu
tensorflow 1.8.0 hb11d968_0
tensorflow-base 1.8.0 py36hc1a7637_0
tensorflow-gpu 1.8.0 h7b35bdc_0
Actual Behavior
I tried accessing an NVIDIA Quadro M2200 GPU I have in my system with tensorflow-gpu 1.8.0-h7b35bdc_0; however, it doesn't seem to be detected when I run the following code:
Expected Behavior
The above Python code should return something similar to the following (which is what I get when tensorflow-gpu 1.7.0 was installed):
Steps to Reproduce
Anaconda or Miniconda version:
Miniconda3
Operating System:
Ubuntu 16.04.4
conda info
conda list --show-channel-urls