tensorflow is failing to find the GPU on a new system:
>>> import tensorflow
2021-02-09 12:26:05.718456: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/lib/x86_64-linux-gnu
2021-02-09 12:26:05.718530: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
tensorflow version
Looks to be 2.4.1
(venv) root@thebeast:~/Documents/DGRU/Code/ml# pip freeze | grep tensorflow
tensorflow==2.4.1
tensorflow-estimator==2.4.0
Cuda version
Appears to be 11.2
(venv) root@thebeast:/usr/lib/x86_64-linux-gnu# nvidia-smi
Tue Feb 9 12:17:41 2021
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 460.32.03 Driver Version: 460.32.03 CUDA Version: 11.2 |
|-------------------------------+----------------------+----------------------+
however, nvcc seems to disagree?
(venv) root@thebeast:/usr/lib/x86_64-linux-gnu# nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2018 NVIDIA Corporation
Built on Tue_Jun_12_23:07:04_CDT_2018
Cuda compilation tools, release 9.2, V9.2.148
My error is the same as the Feb 3 post on this very recent github issue
Has anyone installed TF for CUDA11.1 yet?
I just tried to install tf-nightly-2.5.0.dev20210202 from pip3 but got this error. It seems that tf-nightly still look for CUDA11.0 rather than 11.1.
import tensorflow as tf
2021-02-03 01:04:35.803100: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-11.1/lib64:/usr/local/cuda-11.1/extras/CUPTI/lib64
2021-02-03 01:04:35.803125: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
The accepted answer on this question seems to give a working solution, but is very old.
System information
Nvidia RTX 2060
Debian 10
Question
What is the latest stable cuda that I can use with tensorflow 2.4.2, and how do I downgrade to it? Or, if I don’t need the latest version for an RTX 2060, what version should I use? Thanks!