tensorflow: Could not load dynamic library 'libcudart.so.11.0'

Please make sure that this is a build/installation issue. As per our GitHub Policy, we only address code/doc bugs, performance issues, feature requests and build/installation issues on GitHub. tag:build_template

System information

  • OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 20.04
  • TensorFlow installed from (source or binary): binary
  • TensorFlow version: 2.4.0
  • Python version: 3.7.9
  • Installed using virtualenv? pip? conda?: pip
  • Bazel version (if compiling from source):
  • GCC/Compiler version (if compiling from source):
  • CUDA/cuDNN version: libcudnn8_8.0.5.39-1+cuda11.0_amd64.deb (I also installed the dev version)
  • NVIDIA Driver Version: 450.80.02
  • GPU model and memory: GeForce RTX 2080 Super with Max-Q Design

Describe the problem

Provide the exact sequence of commands / steps that you executed before running into the problem I tried following the instructions as described in: https://www.tensorflow.org/install/gpu

  • I installed Ubuntu 20.04 (which installed the NVIDIA driver)
  • Installed python using pyenv
  • sudo apt install nvidia-cuda-toolkit
  • pip install tensorflow
  • Installed: libcudnn8_8.0.5.39-1+cuda11.0_amd64.deb
  • (saw that it failed to find ‘libcudart.so.11.0’)
  • Installed: libcudnn8-dev_8.0.5.39-1+cuda11.0_amd64.deb
  • (still failed to find ‘libcudart.so.11.0’)

Is there a way for me to check which part of the installation broke? Any ideas on what I can do to fix this? Thanks!

Any other info / logs Include any logs or source code that would be helpful to diagnose the problem. If including tracebacks, please include the full traceback. Large logs and files should be attached.

PyDev console: starting.
Python 3.7.9 (default, Dec 22 2020, 21:13:51) 
[GCC 9.3.0] on linux
>>> import tensorflow as tf
>>> print("Num GPUs Available: ", len(tf.config.experimental.list_physical_devices('GPU')))
2020-12-22 22:56:35.044676: 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
2020-12-22 22:56:35.044691: 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.
/home/ido/TeraResearch/venv/lib/python3.7/site-packages/pandas/compat/__init__.py:120: UserWarning: Could not import the lzma module. Your installed Python is incomplete. Attempting to use lzma compression will result in a RuntimeError.
  warnings.warn(msg)
2020-12-22 22:56:35.938373: I tensorflow/compiler/jit/xla_cpu_device.cc:41] Not creating XLA devices, tf_xla_enable_xla_devices not set
2020-12-22 22:56:35.938801: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcuda.so.1
2020-12-22 22:56:37.494184: E tensorflow/stream_executor/cuda/cuda_driver.cc:328] failed call to cuInit: CUDA_ERROR_UNKNOWN: unknown error
2020-12-22 22:56:37.494273: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:169] retrieving CUDA diagnostic information for host: ido-ml
2020-12-22 22:56:37.494291: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:176] hostname: ido-ml
2020-12-22 22:56:37.494480: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:200] libcuda reported version is: 450.80.2
2020-12-22 22:56:37.494544: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:204] kernel reported version is: 450.80.2
2020-12-22 22:56:37.494561: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:310] kernel version seems to match DSO: 450.80.2
Num GPUs Available:  0

About this issue

  • Original URL
  • State: closed
  • Created 4 years ago
  • Reactions: 98
  • Comments: 67 (6 by maintainers)

Most upvoted comments

@ido-tera-group

If you follow the guidelines as mentioned in above link , you should be able to install it. Glad to know you installed.Please, close this thread if your issue was resolved. Thanks!

+223
ravikyram on Dec 28, 2020

The fucking stupid bug still exist in the latest tf today.

+213
GF-Huang on Feb 10, 2021

Still having the same issue here.

+208
wijjj on Jan 22, 2021

Closing as stale. Please reopen if you’d like to work on this further.

+162
google-ml-butler[bot] on Jan 13, 2021

I had the same problem, it was solved by installing cudatoolkit conda install cudatoolkit

+138
aseedb on Feb 10, 2021

I has same problem ,and solved this problem by those step.

First, find out where the “libcudart.so.11.0” is

If you lost other at error stack, you can replace the “libcudart.so.11.0” by your word in below:

sudo find / -name 'libcudart.so.11.0'

Output in my system.This result shows where the “libcudart.so.11.0” is in my system:

/usr/local/cuda-11.1/targets/x86_64-linux/lib/libcudart.so.11.0

If the result shows nothing, please make sure you have install cuda or other staff that must install in your system.

Second, add the path to environment file.

  1. edit /etc/profile
sudo vim /etc/profile
  1. append path to “LD_LIBRARY_PATH” in profile file
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-11.1/targets/x86_64-linux/lib
  1. make environment file work
source /etc/profile

Last, when it all done, you can try your python code again.

Hope it help for you!

At the end

I do every thing on guide list to install tensorflow. But still occurs the same problem my environment: ubuntu: 20.04LTS tensorflow: 2.4.1 python3: 3.8 pip3: 21 CUDA: 11.1 cuDNN: 8.1.0 GPU: GeForce GTX 860M GPU Driver: 460.39

My error stack before solve this problem:

2021-01-31 15:04:10.925736: 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/extras/CUPTI/lib64
2021-01-31 15:04:10.925774: 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.
2021-01-31 15:04:11.738577: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcuda.so.1
2021-01-31 15:04:11.771425: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:941] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-01-31 15:04:11.771969: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1770] Found device 0 with properties: 
pciBusID: 0000:01:00.0 name: GeForce GTX 860M computeCapability: 5.0
coreClock: 1.0195GHz coreCount: 5 deviceMemorySize: 1.96GiB deviceMemoryBandwidth: 74.65GiB/s
2021-01-31 15:04:11.772084: 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/extras/CUPTI/lib64
2021-01-31 15:04:11.772174: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcublas.so.11'; dlerror: libcublas.so.11: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: :/usr/local/cuda/extras/CUPTI/lib64
2021-01-31 15:04:11.772255: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcublasLt.so.11'; dlerror: libcublasLt.so.11: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: :/usr/local/cuda/extras/CUPTI/lib64
2021-01-31 15:04:11.772333: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcufft.so.10'; dlerror: libcufft.so.10: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: :/usr/local/cuda/extras/CUPTI/lib64
2021-01-31 15:04:11.772413: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcurand.so.10'; dlerror: libcurand.so.10: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: :/usr/local/cuda/extras/CUPTI/lib64
2021-01-31 15:04:11.772490: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcusolver.so.11'; dlerror: libcusolver.so.11: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: :/usr/local/cuda/extras/CUPTI/lib64
2021-01-31 15:04:11.772567: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcusparse.so.11'; dlerror: libcusparse.so.11: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: :/usr/local/cuda/extras/CUPTI/lib64
2021-01-31 15:04:11.772745: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudnn.so.8
2021-01-31 15:04:11.772763: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1803] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...

After I fix it , the python code works fine

+136
Wanxp on Jan 31, 2021

this helped me:

conda install pytorch torchvision cudatoolkit=10.1 -c pytorch

+45
unnir on Apr 2, 2021

I had the same problem, it was solved by installing cudatoolkit conda install cudatoolkit

My computer no GPU, why I need cudatoolkit?

+39
GF-Huang on Feb 10, 2021

I see, thank you! After tinkering a bit, it worked… But I changed the instructions to work for 20.04:

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-ubuntu2004.pin
sudo mv cuda-ubuntu2004.pin /etc/apt/preferences.d/cuda-repository-pin-600
sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/7fa2af80.pub
sudo add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/ /"
sudo apt-get update

Perhaps the docs should be updated?

Thanks again!

+31
ido-tera-group on Dec 24, 2020

Ubuntu 20.04 doesn’t upgrade to CUDA-11.x. Please follow the instructions at the CUDA download page to install CUDA-11.x.

After all, don’t forget joint the CUDA library path with environment of LD_LIBRARY_PATH. echo 'export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/cuda/targets/x86_64-linux/lib' >> ~/.bashrc

+15
vortune on Feb 14, 2021

saw same error:

2021-01-06 17:05:11.751597: 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/Qt/5.14.1/lib
2021-01-06 17:05:11.751625: 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.

but it’s harmless, I don’t want to use a GPU.

+11
thijstriemstra on Jan 6, 2021

Can you refer me to clear instructions on how to install? I see a bunch of blogs each with different ideas… Is there an official (or semi-offical) guide that should actually work? Thanks!

+11
ido-tera-group on Dec 23, 2020

A short quick fix that worked for me

Running Ubuntu 20.04, Python 3.8.5, cuDNN 8.1.0, on an RTX 2080 Super.

I downloaded this run file from NVIDIA, then only opted in to install the CUDA 11 toolkit. After it finished downloading, I added /usr/local/cuda-11.0/bin to my PATH and /usr/local/cuda-11.0/lib64 to my LD_LIBRARY_PATH.

After that, my GPU was functional again.

+10
chriscardillo on Feb 4, 2021

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you.

+10
google-ml-butler[bot] on Jan 4, 2021

I’m facing the same problem. I installed tensorflow-gpu 2.2.0 using Anaconda and it will automatically download cudnn 7.6 and cudatoolkit 10.1. anyone using anaconda also faces the same problem? ( i installed tensorflow-gpu by running this line conda create -n tf-gpu tensorflow-gpu==2.2.0 python==3.8)

+9
songssssss on Dec 6, 2021

I celebrated too soon… tensorflow did import without that immediate message, but when I tried building a model, I got:

2020-12-24 15:31:52.325975: I tensorflow/compiler/jit/xla_cpu_device.cc:41] Not creating XLA devices, tf_xla_enable_xla_devices not set
2020-12-24 15:31:52.326576: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcuda.so.1
2020-12-24 15:31:52.341852: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:941] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-12-24 15:31:52.342269: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1720] Found device 0 with properties: 
pciBusID: 0000:01:00.0 name: GeForce RTX 2080 Super with Max-Q Design computeCapability: 7.5
coreClock: 1.23GHz coreCount: 48 deviceMemorySize: 7.79GiB deviceMemoryBandwidth: 327.88GiB/s
2020-12-24 15:31:52.342442: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.11.0
2020-12-24 15:31:52.343750: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcublas.so.11
2020-12-24 15:31:52.343787: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcublasLt.so.11
2020-12-24 15:31:52.343824: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcufft.so.10'; dlerror: libcufft.so.10: cannot open shared object file: No such file or directory
2020-12-24 15:31:52.343872: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcurand.so.10'; dlerror: libcurand.so.10: cannot open shared object file: No such file or directory
2020-12-24 15:31:52.343906: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcusolver.so.10'; dlerror: libcusolver.so.10: cannot open shared object file: No such file or directory
2020-12-24 15:31:52.343938: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcusparse.so.11'; dlerror: libcusparse.so.11: cannot open shared object file: No such file or directory
2020-12-24 15:31:52.344019: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudnn.so.8
2020-12-24 15:31:52.344026: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1757] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...

Any clues?

+7
ido-tera-group on Dec 24, 2020

LONG LIVE THE BUG

+6
celikmustafa89 on Jun 1, 2023

This worked for me:

# dependecies 
sudo apt install cuda-cudart-11-0
sudo apt install libcufft10
sudo apt install libcurand10
sudo apt install libcusolver-11-0 # installs so.10, see below 
sudo apt install libcudnn8
sudo apt install libcublas-11-0
sudo apt install libcusparse-11-0

# add this path to LD_LIBRARY_PATH in .bashrc as in previous posts
/usr/local/cuda-11.0/targets/x86_64-linux/lib/

# in the /usr/... path, run:
sudo ln -s libcusolver.so.10.6.0.245 libcusolver.so.11
+5
ferroao on Aug 19, 2021

I had the same problem, it was solved by installing cudatoolkit conda install cudatoolkit

This worked for me as well.

+5
hamidpalangi on Feb 11, 2021

you can try to add these code to load dynamic library

from ctypes import * lib8 = cdll.LoadLibrary(‘/data/users/CHDHPC/2017902628/cuda/lib64/libcublas.so.11’) lib1 = cdll.LoadLibrary(‘/data/users/CHDHPC/2017902628/cuda/lib64/libcudart.so.11.0’) lib2 = cdll.LoadLibrary(‘/data/users/CHDHPC/2017902628/cuda/lib64/libcublasLt.so.11’) lib3 = cdll.LoadLibrary(‘/data/users/CHDHPC/2017902628/cuda/lib64/libcufft.so.10’) lib4 = cdll.LoadLibrary(‘/data/users/CHDHPC/2017902628/cuda/lib64/libcurand.so.10’) lib5 = cdll.LoadLibrary(‘/data/users/CHDHPC/2017902628/cuda/lib64/libcusolver.so.10’) lib6 = cdll.LoadLibrary(‘/data/users/CHDHPC/2017902628/cuda/lib64/libcusparse.so.11’) lib7 = cdll.LoadLibrary(‘/data/users/CHDHPC/2017902628/cuda/lib64/libcudnn.so.8’)

+5
Yeah21 on Feb 7, 2021

Good to know! Another solution that worked for me is to find the correct version of Tensorflow that matches my OS, CUDA version, and python version here: https://www.tensorflow.org/install/source#gpu, and just pip install that version of tensorflow.

+4
shaynemei on Feb 16, 2022

I had the same problem, it was solved by installing cudatoolkit conda install cudatoolkit

this worked for me

+4
JapinderSandhu on Apr 11, 2021

@ido-tera-group

Please, refer the tutorial from here.Thanks!

+4
ravikyram on Dec 23, 2020

I has same problem ,and solved this problem by those step.

First, find out where the “libcudart.so.11.0” is

If you lost other at error stack, you can replace the “libcudart.so.11.0” by your word in below:

sudo find / -name 'libcudart.so.11.0'

Output in my system.This result shows where the “libcudart.so.11.0” is in my system:

/usr/local/cuda-11.1/targets/x86_64-linux/lib/libcudart.so.11.0

If the result shows nothing, please make sure you have install cuda or other staff that must install in your system.

Second, add the path to environment file.

  1. edit /etc/profile
sudo vim /etc/profile
  1. append path to “LD_LIBRARY_PATH” in profile file
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-11.1/targets/x86_64-linux/lib
  1. make environment file work
source /etc/profile

Last, when it all done, you can try your python code again.

Hope it help for you!

At the end

I do every thing on guide list to install tensorflow. But still occurs the same problem my environment: ubuntu: 20.04LTS tensorflow: 2.4.1 python3: 3.8 pip3: 21 CUDA: 11.1 cuDNN: 8.1.0 GPU: GeForce GTX 860M GPU Driver: 460.39

My error stack before solve this problem:

2021-01-31 15:04:10.925736: 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/extras/CUPTI/lib64
2021-01-31 15:04:10.925774: 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.
2021-01-31 15:04:11.738577: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcuda.so.1
2021-01-31 15:04:11.771425: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:941] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-01-31 15:04:11.771969: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1770] Found device 0 with properties: 
pciBusID: 0000:01:00.0 name: GeForce GTX 860M computeCapability: 5.0
coreClock: 1.0195GHz coreCount: 5 deviceMemorySize: 1.96GiB deviceMemoryBandwidth: 74.65GiB/s
2021-01-31 15:04:11.772084: 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/extras/CUPTI/lib64
2021-01-31 15:04:11.772174: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcublas.so.11'; dlerror: libcublas.so.11: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: :/usr/local/cuda/extras/CUPTI/lib64
2021-01-31 15:04:11.772255: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcublasLt.so.11'; dlerror: libcublasLt.so.11: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: :/usr/local/cuda/extras/CUPTI/lib64
2021-01-31 15:04:11.772333: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcufft.so.10'; dlerror: libcufft.so.10: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: :/usr/local/cuda/extras/CUPTI/lib64
2021-01-31 15:04:11.772413: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcurand.so.10'; dlerror: libcurand.so.10: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: :/usr/local/cuda/extras/CUPTI/lib64
2021-01-31 15:04:11.772490: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcusolver.so.11'; dlerror: libcusolver.so.11: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: :/usr/local/cuda/extras/CUPTI/lib64
2021-01-31 15:04:11.772567: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcusparse.so.11'; dlerror: libcusparse.so.11: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: :/usr/local/cuda/extras/CUPTI/lib64
2021-01-31 15:04:11.772745: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudnn.so.8
2021-01-31 15:04:11.772763: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1803] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...

After I fix it , the python code works fine

Hi, in this how do you do Step #2?

  1. Do you open your virtual python environment, then do these changes to add it?
  2. Do you simply open the etc/profile file and add to the end?
+3
cyriltw on Nov 22, 2021

Please run these command. It will solve the issue

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-ubuntu1804.pin sudo mv cuda-ubuntu1804.pin /etc/apt/preferences.d/cuda-repository-pin-600 sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/7fa2af80.pub sudo add-apt-repository “deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/ /” sudo apt-get update sudo apt-get -y install cuda

+3
NarenZen on Jun 23, 2021

What solved the problems for me with cuda12.2 on Ubuntu 22.04 LTS was setting below PATHs according to Nvidia CUDA Installation instructions

Source: https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#post-installation-actions

To do this I added the below lines to my ~/.bashrc file

export PATH=/usr/local/cuda-12.2/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-12.2/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

The fix is then running the following commands:

sudo cp /usr/local/cuda-12.2/lib64/libcudart.so.12 /usr/local/cuda-12.2/lib64/libcudart.so.11.0
sudo cp /usr/lib/x86_64-linux-gnu/libnvinfer.so.8.6.1 /usr/local/cuda-12.2/lib64/libnvinfer.so.7
sudo cp /usr/lib/x86_64-linux-gnu/libnvinfer_plugin.so.8.6.1 /usr/local/cuda-12.2/lib64/libnvinfer_plugin.so.7
sudo cp /usr/local/cuda-12.2/lib64/libcublas.so.12 /usr/local/cuda-12.2/lib64/libcublas.so.11
sudo cp /usr/local/cuda-12.2/lib64/libcublasLt.so.12 /usr/local/cuda-12.2/lib64/libcublasLt.so.11
sudo cp /usr/local/cuda-12.2/lib64/libcufft.so.11 /usr/local/cuda-12.2/lib64/libcufft.so.10
sudo cp /usr/local/cuda-12.2/targets/x86_64-linux/lib/libcusparse.so.12 /usr/local/cuda-12.2/lib64/libcusparse.so.11

Your file names and locations will vary so please adjust accordingly. As someone suggested above use command below to locate above file locations (if they exist):

sudo find / -name 'whateverfileyouwant'

It seems recent TensorFlow versions have not updated the names, versions, and locations of these files for more recent CUDA versions so we are just copying the new files to the old location and renaming them to their old names while keeping the recent versions with their original names in their original locations.

+2
dsanmiguel on Jul 21, 2023

Oh, I guess those two packages are now combined into one: https://www.tensorflow.org/install image

it’s really annoying to users to show this warning though if they actually don’t want to use the GPU

Edit: Oh, I guess now you can actively install the CPU version only with the new versions of tensorflow:

A smaller CPU-only package is also available:

$ pip install tensorflow-cpu

I guess that may fix my warning.

+2
imatiach-msft on Jan 5, 2022

this helped me:

conda install pytorch torchvision cudatoolkit=10.1 -c pytorch

What the hell, it seriously worked for me! Can’t believe it, what a lol sol.

+2
braindotai on Jun 23, 2021

I see, thank you! After tinkering a bit, it worked… But I changed the instructions to work for 20.04:

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-ubuntu2004.pin
sudo mv cuda-ubuntu2004.pin /etc/apt/preferences.d/cuda-repository-pin-600
sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/7fa2af80.pub
sudo add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/ /"
sudo apt-get update

Perhaps the docs should be updated?

Thanks again!

Thanks for this. What package did you subsequently install?

+2
dbolser on Mar 16, 2021

@ido-tera-group

As per the build configurations from here it should work with this configuration. Can you please create a fresh environment and try to install from scratch and see if the issue still persists.Thanks!

+2
ravikyram on Dec 23, 2020

ONE BUG TO RULE THEM ALL

still exists, unbelievable.

+1
celikmustafa89 on Jun 1, 2023

I’ve resolved the issue by performing following-

  1. Removed currently installed cuda and tensorflow versions.
  2. Installed cuda-toolkit using the command sudo apt install nvidia-cuda-toolkit
  3. upgraded to NVIDIA Driver Version: 510.54
  4. Installed Tensorflow==2.7.0

and everything is just amazing now:) thanks for the help.

+1
Kamlesh364 on Feb 16, 2022

For anyone having this issue (ping @Kamlesh364), I strongly recommend checking out Docker and using the official images: https://hub.docker.com/r/tensorflow/tensorflow/. It’s a new technology to learn, but then saves a lot of time when solving these dependency-hell problems, since the Docker image has all the dependencies (including CUDA) pre-installed. Docker also integrates nicely with VSCode (check out devcontainers).

You just need to install a CUDA-capable driver and enable GPU passthrough to Docker: Linux tutorial, Windows tutorial.

+1
JanPokorny on Jan 20, 2022

I had the same problem, it was solved by installing cudatoolkit conda install cudatoolkit

Thanks! This worked for me .

+1
KatameRonin on Sep 3, 2021

normally don’t post but go to the following link below, do exactly as the instructions say, then reboot your linux box(server) and you should be fine.

https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&Distribution=WSL-Ubuntu&target_version=2.0&target_type=deb_network

PS, other people wrote the correct steps but in the case that the download URL changes or etc… I have provided the downloads link from NVIDIA

+1
ithllc on Jun 26, 2021
Read more comments on GitHub
← tensorflow: "Could not load dynamic library 'cudart64_110.dll'; dlerror: cudart64_110.dll not found"
tensorflow: Could not load dynamic library 'libcudnn.so.8' →