New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Unable to locate package cuda-command-line-tools #16214

Closed
Queequeg92 opened this Issue Jan 18, 2018 · 25 comments

Comments

Projects
None yet
@Queequeg92

Queequeg92 commented Jan 18, 2018

System information

  • Have I written custom code (as opposed to using a stock example script provided in TensorFlow):No
  • OS Platform and Distribution (e.g., Linux Ubuntu 16.04):Linux Ubuntu 16.04
  • TensorFlow installed from (source or binary):binary
  • TensorFlow version (use command below):r1.5
  • Python version: 3.6
  • Bazel version (if compiling from source):
  • GCC/Compiler version (if compiling from source):
  • CUDA/cuDNN version:9.0
  • GPU model and memory:
  • Exact command to reproduce:

Describe the problem

In the official installing guide of r1.5, the libcupti-dev library is required to run tensorflow with GPU support. When issue the following command line for CUDA Toolkit >= 8.0:
$ sudo apt-get install cuda-command-line-tools
I got this error:
$ E: Unable to locate package cuda-command-line-tools
It can't be solved after updating source list.
I have tried on my desktop and a VM instance on Google Cloud Platform, both with Linux Ubuntu 16.04.

Source code / logs

@PatWie

This comment has been minimized.

PatWie commented Jan 18, 2018

As part of an answer to your email @Queequeg92
I found it important to explicitly set the ENV vars to the correct version like

export CUPIT_LIB_PATH=${OPT_PATH}/cuda/toolkit_9.0/cuda/extras/CUPTI/lib64
export LD_LIBRARY_PATH=${CUPIT_LIB_PATH}:$LD_LIBRARY_PATH

(same for Toolkit 8.0)
in my .bashrc. Note there might be some old libcupti.so* in /usr/lib/x86_64-linux-gnu/libcupti.so:

user@host $ ll ${OPT_PATH}/cuda/toolkit_9.0/cuda/extras/CUPTI/lib64
libcupti.so
libcupti.so.9.0
libcupti.so.9.0.176*
user@host $ locate libcupti.so
/usr/lib/x86_64-linux-gnu/libcupti.so
/usr/lib/x86_64-linux-gnu/libcupti.so.7.5
/usr/lib/x86_64-linux-gnu/libcupti.so.7.5.18

The correct package for apt-get would be sudo apt-get install libcupti-dev

@cyrilzh

This comment has been minimized.

cyrilzh commented Feb 3, 2018

Tensorflow's Linux installation guide is misleading. Try a search and here is the result.

$ sudo apt-cache search cuda-command-line-tool
cuda-command-line-tools-8-0 - CUDA command-line tools
cuda-command-line-tools-9-0 - CUDA command-line tools
cuda-command-line-tools-9-1 - CUDA command-line tools

What you should run is something like "sudo apt install cuda-command-line-tools-9-1"

@tensorflowbutler

This comment has been minimized.

Member

tensorflowbutler commented Feb 17, 2018

Nagging Awaiting Response: It has been 14 days with no activityand the awaiting response label was assigned. Is this still an issue?

@manbharae

This comment has been minimized.

manbharae commented Feb 18, 2018

@tensorflowbutler yup still an issue. @cyrilzh solution works. tensorflow documentation here has to be updated : sudo apt-cache search cuda-command-line-tools-9-0

@AIGyan

This comment has been minimized.

AIGyan commented Feb 22, 2018

Hello Team,
I am still getting the same error after using @cyrilzh solution.
$sudo apt-cache search cuda-command-line-tool
$ sudo apt install cuda-command-line-tools-9-1
Reading package lists... Done
Building dependency tree
Reading state information... Done
E: Unable to locate package cuda-command-line-tools-9-1

@bjenkinsgit

This comment has been minimized.

bjenkinsgit commented Feb 22, 2018

I am in the same boat as AIGyan. Unable to locate command line tools package.

@manbharae

This comment has been minimized.

manbharae commented Feb 22, 2018

@AIGyan @bjenkinsgit
https://www.tensorflow.org/install/install_linux is a some parts accurate some parts misleading and someparts inaccurate.

Like @cyrilzh suggested do 👍
:~$ sudo apt-cache search cuda-command-line-tool
[sudo] password for manbharae:
cuda-command-line-tools-8-0 - CUDA command-line tools
cuda-command-line-tools-9-0 - CUDA command-line tools
cuda-command-line-tools-9-1 - CUDA command-line tools

but don't do use cuda-command-line-tools-9-1,
Tensorflow supports only 9-0 at the moment.

uninstall cuda 9-1 and install cuda-9-0.
then install cuda-command-line-tools-9-0

hope it helps.
This worked for me.

@lejafar

This comment has been minimized.

lejafar commented Feb 26, 2018

@bjenkinsgit @AIGyan I'm using Ubuntu 16.04 and I resolved the issue by downloading the package from http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/

curl -O http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/cuda-9-0_9.0.176-1_amd64.deb
dpkg -i ./cuda-9-0_9.0.176-1_amd64.deb
apt-get update
apt-get install cuda-9-0

This will also install cuda-command-line-tools-9-0

@themightyoarfish

This comment has been minimized.

Contributor

themightyoarfish commented Feb 27, 2018

@lejafar I had installed Cuda 9.0 from Nvidia + cudnn 7 and when I tried your suggestion...

sudo dpkg -i ./cuda-9-0_9.0.176-1_amd64.deb
Selecting previously unselected package cuda-9-0.
(Reading database ... 127348 files and directories currently installed.)
Preparing to unpack ./cuda-9-0_9.0.176-1_amd64.deb ...
Unpacking cuda-9-0 (9.0.176-1) ...
dpkg: dependency problems prevent configuration of cuda-9-0:
 cuda-9-0 depends on cuda-toolkit-9-0 (>= 9.0.176); however:
  Package cuda-toolkit-9-0 is not installed.
 cuda-9-0 depends on cuda-runtime-9-0 (>= 9.0.176); however:
  Package cuda-runtime-9-0 is not installed.
 cuda-9-0 depends on cuda-demo-suite-9-0 (>= 9.0.176); however:
  Package cuda-demo-suite-9-0 is not installed.

dpkg: error processing package cuda-9-0 (--install):
 dependency problems - leaving unconfigured
Errors were encountered while processing:
 cuda-9-0
@bjenkinsgit

This comment has been minimized.

bjenkinsgit commented Feb 27, 2018

In the end I downloaded the Tensorflow 9.1 source, installed bazel and other tools and compiled the source for my GPU. I now have Tensorflow 9.1, cudnn 7.0.5 running on Ubuntu 17.10.

@zxr12748

This comment has been minimized.

zxr12748 commented Mar 2, 2018

I tried
$ sudo apt-cache search cuda-command-line-tool
but got nothing.
I have cuda 9.0,cudnn7.05 running on Ubuntu 16.04.
Is there any other solution?
THX:-D

@Netroman

This comment has been minimized.

Netroman commented Mar 5, 2018

@zxr12748

I tried
$ sudo apt-cache search cuda-command-line-tool
but got nothing.

Same here... Did you perhaps find a solution?

@max0x

This comment has been minimized.

max0x commented Mar 9, 2018

Based on @lejafar 's suggestion, I downloaded http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/cuda-command-line-tools-9-1_9.1.85-1_amd64.deb and then ran

sudo dpkg -i cuda-command-line-tools-9-1_9.1.85-1_amd64.deb

However, it will remind you to install some dependency packages first, such as cuda-nvprof-9-1. Thus I installed all dependency packages and then installed cuda-command-line-tools-9-1 successfully.

@NataliaDiaz

This comment has been minimized.

NataliaDiaz commented Mar 13, 2018

Yes, this is still an issue, @tensorflowbutler

@WormCoder

This comment has been minimized.

WormCoder commented Mar 21, 2018

This is not necessary. CUPTI has been installed along with CUDA.
Just set the LD_LIBRARY_PATH, and tf works fine.

@punitvara

This comment has been minimized.

punitvara commented Mar 23, 2018

Do following things. Hope it helps :
$ sudo apt-get install cuda-9.0
$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-9.0/lib64

Download cuDNN for 9.0 (You need to register before downloading)
https://developer.nvidia.com/rdp/form/cudnn-download-survey

$ sudo dpkg -i libcudnn7_7.1.2.21-1+cuda9.0_amd64.deb

Close all terminal and open new
$ source activate tensorflow
$ python
#> > import tensorflow as tf

You should not get any error now. @NataliaDiaz @Netroman @zxr12748

@jiyuhan

This comment has been minimized.

jiyuhan commented Mar 30, 2018

importing tensorflow doesn't necessarily check if cuda is installed, I may be wrong.

@AllBecomesGood

This comment has been minimized.

AllBecomesGood commented Apr 12, 2018

@cyrilzh solution worked (sudo apt install cuda-command-line-tools-9-0), however it apparently got installed already somewhere along the way (i installed nvidia gpu driver, cuda 9.0, cudnn 7.0.5 runtime and cudnn 7.0.5 dev lib).

@vissac

This comment has been minimized.

vissac commented Apr 14, 2018

It should be installed already, just set the env to path is fine.

@jrajagopal

This comment has been minimized.

jrajagopal commented Apr 17, 2018

As pointed out by @vissac and @manbharae, cuda-command-line-tools-9-0 is already installed when you load cuda. (I had to install the CUDA package using the local deb)

@tensorflowbutler

This comment has been minimized.

Member

tensorflowbutler commented May 2, 2018

It has been 14 days with no activity and the awaiting response label was assigned. Is this still an issue?

@bjenkinsgit

This comment has been minimized.

bjenkinsgit commented May 2, 2018

I solved the problem and got 9.1 installed.

@chenzhekl

This comment has been minimized.

chenzhekl commented Jun 16, 2018

The documentation is still incorrect at the moment.

@slothkong

This comment has been minimized.

slothkong commented Jun 19, 2018

@lejafar Prior to installing that package, you installed and cuda package using dpkg? I started with ony the vnidia drivers install, the I followed your comment but had the same error as @themightyoarfish . After that whenever I use sudo apt-get intall I get a messave of unmenment dependencies that includes cuda-toolking-9.0. How did you guuys solve the issue?

@ijoseph

This comment has been minimized.

ijoseph commented Jul 19, 2018

After about literally an hour of dependency hell, I managed to find all the .debs I needed for command-line-tools for Ubuntu 16.04 compatible with cuda 9.2 on this page. (i.e., without having to downgrade to cuda 9.0 or cuda 9.1). Similar to @max0x 's solution, I guess.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment