Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
This question does not appear to be about a specific programming problem, a software algorithm, or software tools primarily used by programmers. If you believe the question would be on-topic on another Stack Exchange site, you can leave a comment to explain where the question may be able to be answered.
Closed 2 years ago.
Improve this question
Tensorflow official recommendation
So, I'm using Ubuntu 20.4 and I want to use Tensorflow with version 2.3. The offcial Tf sources say that 10.1 is supported, but I couldn't find the installation of CUDA 10.1 for Ubuntu 20.4.
Is it possible to use CUDA 10.1 on Ubuntu and if not, how can I install CUDA 11, so I can make TF 2.3 work?
Yes you can! The usual way would be to build TF from source, which can take many hours (thats atleast what I read). This is required, as tensorflow is compiled with a specific cuda version, thats why they have to match.
After some research I found out, that davidenunes compiled different TF version with different cuda version, so you dont need to do that!
Have a look at his github and pull the version you need. With this I got my 2.3 tf working on Ubuntu 20.4 with cuda 11.
If you have a working cuda 11 installation ready, or you need cuda 11, you can do it this way.
However, you should too be able to install cuda 10.1 on Ubuntu 20.04, but I recommend you to use cuda 11, as with it you can use cudnn 8 which speeds learning up alot.
Related
Closed. This question is not about programming or software development. It is not currently accepting answers.
This question does not appear to be about a specific programming problem, a software algorithm, or software tools primarily used by programmers. If you believe the question would be on-topic on another Stack Exchange site, you can leave a comment to explain where the question may be able to be answered.
Closed 6 days ago.
Improve this question
I have some experience with Linux systems and finally switched to Arch Linux. The setup went fine and everything seemed to work well until this morning. Since then I encounter two problems that could have the same source, here I will briefly describe theme:
When trying to clone repositories from github (or other sources) using ssl I will get the following message on my arch setup:
fatal: unable to access 'https://github.com/random/repository/to/clone.git': Insufficient randomness
When trying to generate a key pair using ssh-keygen I receive the following error:
PRNG is not seeded
So my guess was, that this has something to do with random generators of the system and researched a lot there.
I did already recreate both
mknod /dev/random c 1 8
and
mknod /dev/urandom c 1 9
using mknod.
I installed the following packages:
rng-tools 6.16.1
jitterentropy 3.4.1
rtl-sdr 10.8.0
I installed Arch Linux last Friday and everything seemed to work fine. The Kernel is 6.1.10, it should be up to date. Due to the problems with the ssl connection I cannot not directly use pacman to install new packages or update the installed versions, but I is possible to download them from a mirror and install them by using pacman -U.
The entropy available seems to be stable at 256, which older pages tell me is way to low, but with the newer kernel versions is fine. I use a laptop from DELL, if the specs are relevant I can provide them. For all I read, there are a lot of old solutions but I found no matching problem and not quite relevant for more modern kernels (like using haveged, could be, but should not be necessary how I understood it).
To use root privileged to create keys or clone a git does not change a thing.
I hope anyone has an idea that will help me and I will provide any further information that could be helpful for solving the problem.
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
This question does not appear to be about a specific programming problem, a software algorithm, or software tools primarily used by programmers. If you believe the question would be on-topic on another Stack Exchange site, you can leave a comment to explain where the question may be able to be answered.
Closed 1 year ago.
Improve this question
I'm working on a computer on which Nvidia drivers and Cuda were installed by someone else so I don't know the method they used to install them.
In the /usr/local/ there were two directories cuda and cuda.10.0. Running nvidia-smi would output:
CUDA Version: 11.0
which made me believe two cuda versions were installed on the system which were causing some errors.
following this question I removed cuda by first doing:
sudo apt-get --purge remove "*cublas*" "cuda*" "nsight*"
and then doing
sudo rm -rf /usr/local/cuda*
(I did not uninstsall nvidia-drivers and Driver Version: 450.80.02 is installed).
Running nvidia-smi still outputs:
CUDA Version: 11.0
How do I uninstall cuda 11? I prefer to have cuda 10 and I can't find where cuda 11 is installed.
Do I need to uninstall nvidia-drivers as well?
The nvidia-smi command does not show which version of CUDA is installed, it shows which CUDA version the installed nVidia driver supports, so there is no problem here, just the incorrect interpretation of the output of this command.
Even if you remove all CUDA installations, nvidia-smi would still show the maximum CUDA version that you can use with this driver.
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
This question does not appear to be about a specific programming problem, a software algorithm, or software tools primarily used by programmers. If you believe the question would be on-topic on another Stack Exchange site, you can leave a comment to explain where the question may be able to be answered.
Closed 3 years ago.
Improve this question
I am trying to set up my system for gpu computing for training deep learning models. The tensorflow version required is 1.15 gpu. I would like to know which version of CUDA and CUDnn i have to install in my system?
From the official TF documentation.
For TF >=1.13, CUDA 10. Here
For TF>= 2.1, CUDA 10.1. Here
And, CuDNN will be same for both, CuDNN >= 7.6. Here
I found "TF 1.15 was compiled against CUDA 10.0" here. cuDNN 7.6.4 seems to fit this following this
It is now Oct 29, 2018
After much googling, I have not found a definitive answer or any examples of people using the latest cuda10 for tensorflow on ubuntu 14.04.
My dilemma is whether to upgrade my OS (currently at 14.04) in order to run cuda9 so I can use the latest tensorflow version or use CUDA10 on my existing 14.04 install.
Note cuda9 does not support 14.04, however, Nvidia has indicated that 14.04 will be supported for cuda10.
So, any examples/experiences of people using tensorflow with cuda10 on ubuntu14.04 are keenly sought after!
Also note cuda10 is not specifically supported by tensorflow...yet...they say "soon". But TF can be built from source with cuda10.
This is a link for cuda10+tensorflow on ubuntu16.04:
https://github.com/tensorflow/tensorflow/issues/22706
The short answer, I realize, is "try building it myself". Before I do that, I thought I'd ask around. Thanks.
I don't know whether CUDA 10 can work well on Ubuntu 14.04, but I was managed to build TensorFlow with CUDA 10 on Ubuntu 18.04 with using NVIDIA released docker image.
You can pull the 'TensorFlow Release 18.09' and try it on your current system.
If the previous step does not work, consider upgrading your OS to 18.04.
I wrote down my installation experience on this page, you could read it for some detail if you need.
Closed. This question is off-topic. It is not currently accepting answers.
Want to improve this question? Update the question so it's on-topic for Stack Overflow.
Closed 10 years ago.
Improve this question
I would be interested to try a GPU emulator, but I have tried to use Multi2Sim, GPGPU-sim, and Ocelot, and for each of these three emulators I get a problem for which it seems hard to find a solution on the internet. I will describe the problem I have with each emulator and maybe you can help. First of all, to give you some detailed context, I am using Ubuntu 12.04 LTS.
Multi2Sim says that it is not compatible with 64-bit and so you should compile for 32-bit. If I compile CUDA code for 32-bit, then when I run the compiled executable, I get the error message "CUDA driver version is insufficient for CUDA runtime version." If I compile OpenCL code for 32-bit, then when I run the compiled executable, I find that the function clGetPlatformIDs does not give me the Nvidia OpenCL platform that I get when I compile for 64-bit.
The documentation for GPGPU-sim says:
We have tested OpenCL on GPGPU-Sim using NVIDIA driver version 256.40
http://developer.download.nvidia.com/compute/cuda/3_1/drivers/devdriver_3.1_linux_64_256.40.run
Note the most recent version of the NVIDIA driver produces PTX that is incompatible with this version of GPGPU-Sim.
I have NVIDIA Driver Version 295.49. When I look in "Additional Drivers" from "System Settings" I see two things listed: "NVIDIA accelerated graphics driver (version current) [Recommended]" and "NVIDIA accelerated graphics driver (post-release updates) (version current-updates)". The first one was activated, so I clicked Remove and then the second one automatically became activated. So I decided to just try installing version 256.40 and I got this error message which simply intimidates me:
ERROR: If you are using a Linux 2.4 kernel, please make sure
you either have configured kernel sources matching your
kernel or the correct set of kernel headers installed
on your system.
If you are using a Linux 2.6 kernel, please make sure
you have configured kernel sources matching your kernel
installed on your system. If you specified a separate
output directory using either the "KBUILD_OUTPUT" or
the "O" KBUILD parameter, make sure to specify this
directory with the SYSOUT environment variable or with
the equivalent nvidia-installer command line option.
Depending on where and how the kernel sources (or the
kernel headers) were installed, you may need to specify
their location with the SYSSRC environment variable or
the equivalent nvidia-installer command line option.
When I try to build Ocelot, I get the following, even though I followed the instructions "To pull from the LLVM SVN and build":
ocelot/ocelot/ir/implementation/ExternalFunctionSet.cpp:27:36: fatal error: llvm/Target/TargetData.h: No such file or directory