So last post I got a little hot over trying to install tensorflow properly. At this point, I’m going to once again try to get the GPU version functioning (I’ll keep it on my non-virtual environment for now).
pip uninstall tensorflow python -m pip install --upgrade pip pip install tensorflow-gpu python -c "import tensorflow as tf; tf.enable_eager_execution(); print(tf.reduce_sum(tf.random_normal([1000, 1000])))"
I got the same DLL error as before, but this time I’m just going to try and troubleshoot it because I know the GPU version will run faster. And I believe I have a CUDA compatible GPU (NVIDIA GeForce GTX 1060), so all I need to do is follow the instructions here.
Again I’ll just keep it to bullets of what I did:
- Updated my NVIDIA driver (CUDA 10.0 requires 410.x or higher)
- Installed NVIDIA CUDA Version 10.1
- Downloaded cuDNN SDK and then followed this installation guide to install (I got stuck on the last step in installing this, but discovered other people ran into this same issue and ignored it and everything still worked fine)
It looks like I should be set to continue on, but from now on I’m going to reference this article that explains how to install tensorflow using something called Anaconda. If you haven’t noticed, the main point of this blog is to keep me organized while I go through setting up things like this so my future self remembers what I’ve done after I inevitably rage quit.
conda create -n tensorflow python=3.5 activate tensorflow pip install tensorflow-gpu
Lastly I just had to validate the install by re-launching the Anaconda Prompt and typing in the following:
activate tensorflow python import tensorflow as tf
…AND I GOT THE SAME GD DLL ERROR AS BEFORE……..
THAT NEURO NETWORK TRAINING ARTICLE MADE IT LOOK LIKE THIS PART WAS EASY BY COMPLETELY GLOSSING OVER THIS NIGHTMARE.
Going back to the original gd instructions I went back to my trusty Command Prompt and entered the following just to see if maybe it was from a path environment setting error:
SET PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\bin;%PATH% SET PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\extras\CUPTI\libx64;%PATH% SET PATH=C:\tools\cuda\bin;%PATH%
And still no luck. So really, I’m done. When I return I’ll look into TensorFlow issues 22521 and 22794 since they both got the same “ImportError: DLL load failed: The specified module could not be found.”
This blog has become way too much about geeky stuff and not enough about cat pictures, so here’s some random ones on my hard drive.