-
Notifications
You must be signed in to change notification settings - Fork 57
/
cuda.sh
executable file
·131 lines (100 loc) · 6.56 KB
/
cuda.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
#!/bin/bash
echo "########################"
echo "# CUDA and CUDNN Setup #"
echo "########################"
echo ""
echo "Manual steps are involved! you may need to update the"
echo "install paths and download some files yourse"
echo ""
echo "@author Andrew Hundt <[email protected]>"
echo ""
# source: https://gist.github.com/phatblat/1713458
# Save script's current directory
DIR=$(pwd)
# /bin/bash
set -e
set -u
set -x
echo ""
echo "###################################################"
echo "Go to the NVIDIA website and download CUDA + CUDNN"
echo "https://developer.nvidia.com/cuda-toolkit"
echo "https://developer.nvidia.com/cudnn"
echo "and put it in ~/Downloads"
echo "the new .deb versions are recommended on ubuntu"
echo ""
echo "IMPORTANT:"
echo "You may need to install the latest graphics driver first"
echo "CUDA must be installed with no graphics running"
echo "to disable graphics (from terminal):"
echo "sudo service lightdm stop"
echo "###################################################"
echo ""
# NVIDIA CUDA
sudo apt-get install -y linux-headers-$(uname -r)
if [ ! -f ~/Downloads/cuda-repo-ubuntu1604-9-1-local_9.1.85-1_amd64.deb ] ; then
# cuda itself
curl https://developer.nvidia.com/compute/cuda/9.1/Prod/local_installers/cuda-repo-ubuntu1604-9-1-local_9.1.85-1_amd64 --output ~/Downloads/cuda-repo-ubuntu1604-9-1-local_9.1.85-1_amd64.deb
# latest patch
curl https://developer.nvidia.com/compute/cuda/9.1/Prod/patches/1/cuda-repo-ubuntu1604-9-1-local-cublas-performance-update-1_1.0-1_amd64 --output ~/Downloads/cuda-repo-ubuntu1604-9-1-local-cublas-performance-update-1_1.0-1_amd64.deb
fi
sudo dpkg -i ~/Downloads/cuda-repo-ubuntu1604-9-1-local_9.1.85-1_amd64.deb
sudo dpkg -i ~/Downloads/cuda-repo-ubuntu1604-9-1-local-cublas-performance-update-1_1.0-1_amd64.deb
sudo apt-key add /var/cuda-repo-9-1-local/7fa2af80.pub
sudo apt-get update
sudo apt-get install -y cuda-libraries-9-1 cuda-repo-ubuntu1604-9-1-local-cublas-performance-update-1
# Note that if you are having problems with these steps, manually
# check the repository in the ubuntu software center:
# https://askubuntu.com/questions/650692/installing-cuda-7-0-on-ubuntu-14-04
# You will have to add the file to your software list. After running:
# sudo dpkg -i cuda-repo-ubuntu1404-7-0-local_7.0-28_amd64.deb
# open Softwares & Updates (search for it in the Dash)
# In the Other Software tab, enable the checkbox corresponding to your package. It will be of the form
# file:///var/cuda...
# Installation Instructions from :
# https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1604&target_type=deblocal:
# `sudo dpkg -i cuda-repo-ubuntu1604-9-1-local_9.1.85-1_amd64.deb`
# `sudo apt-key add /var/cuda-repo-<version>/7fa2af80.pub`
# `sudo apt-get update`
# `sudo apt-get install cuda`
# Other installation options are available in the form of meta-packages. For example, to install all the library packages, replace "cuda" with the "cuda-libraries-9-1" meta package. For more information on all the available meta packages click here.
# nccl which is useful for horovod support http://github.com/uber/horovod/
if [ ! -f ~/Downloads/nccl-repo-ubuntu1604-2.1.4-ga-cuda9.1_1-1_amd64.deb ] ; then
curl https://developer.nvidia.com/compute/machine-learning/nccl/secure/v2.1/prod/nccl-repo-ubuntu1604-2.1.4-ga-cuda9.1_1-1_amd64 --output nccl-repo-ubuntu1604-2.1.4-ga-cuda9.1_1-1_amd64.deb
fi
sudo dpkg -i ~/Downloads/nccl-repo-ubuntu1604-2.1.4-ga-cuda9.1_1-1_amd64.deb
sudo apt-get update
sudo apt-get install -y libnccl2 libnccl-dev libnccl-dev nccl-repo-ubuntu1604-2.1.4-ga-cuda9.1
# The steps below are outdated, remove if there have been no problems
# # install the CUDA profiler for cuda 8.0
# # Here is how to profile tensorflow: https://towardsdatascience.com/howto-profile-tensorflow-1a49fb18073d
# # also see https://github.com/tensorflow/tensorflow/issues/9341#issuecomment-324041125
# if [ ! -f ~/Downloads/libcupti8.0_8.0.44-3_amd64.deb ] ; then
# curl http://cz.archive.ubuntu.com/ubuntu/pool/multiverse/n/nvidia-cuda-toolkit/libcupti8.0_8.0.44-3_amd64.deb --output ~/Downloads/libcupti8.0_8.0.44-3_amd64.deb
# fi
# sudo dpkg -i ~/Downloads/libcupti8.0_8.0.44-3_amd64.deb
# if [ ! -f ~/Downloads/libcupti-dev_8.0.44-3_amd64.deb ] ; then
# curl http://cz.archive.ubuntu.com/ubuntu/pool/multiverse/n/nvidia-cuda-toolkit/libcupti-dev_8.0.44-3_amd64.deb --output ~/Downloads/libcupti-dev_8.0.44-3_amd64.deb
# fi
# sudo dpkg -i ~/Downloads/libcupti-dev_8.0.44-3_amd64.deb
# install CUDNN for deep learning
sh cudnn.sh
# OLD ADVICE FROM 2016
# --------------------
# echo "GTX1080 instructions that may help: https://github.com/fchollet/keras/issues/3043#issuecomment-233480326"
# for anyone finding this: just spent a day trying to get a gtx 1070 to run on ubuntu 16.04 w/ CUDA 8 rc and Theano. Here are some guidelines:
# gtx 1070 and 1080 require a nvidia 367 driver. You can't currently get this from apt, so you need to download a runfile and execute it w/ the ubuntu gui shut down. here's a tutorial: http://www.yourownlinux.com/2016/06/how-to-install-nvidia-367-27-stable-graphics-drivers-in-linux.html
# cuda 7.5 did not seem to want to work w/ the 367 driver. Potentially I could have fixed this. Instead I decided to install cuda 8.0 rc. However, using the deb file installation approach automatically installed the nvidia 361 drivers, which caused a conflict that was very difficult to remove. This is the source of the "could not insert 'nvidia_361_uvm'" error.
# Also, cuda 8.0 rc requires you to reinstall nvidia drivers anyway, which I did not realize.
# To repair:
# sudo apt-get purge nvidia* to remove 361 and its assorted packages.
# get to pure shell w/ ctrl+alt+f2, `sudo service lightdm stop`, and use the 367 runfile w/ the --uninstall argument to uninstall the 367 drivers. i.e. sudo sh nvidia367.xx.run --uninstall
# w/ lightdm shutdown, reinstall 367 drivers with the runfile. i.e. sudo sh nvidia367.xx.run
# install cuda-8.0 rc w/ its runfile, which gives you the option to install the 361 drivers. When presented with it, choose no. i.e. sudo sh cuda-8.0.x.x.run
# following this I needed to update some PATH stuff that was pointing to the cuda 7.5 folder. Also I had copied cuDNN files into 7.5 include and lib64 folders, so those files needed to be copied to the 8.0 equivalents.
# echo "Also open /usr/local/cuda/include/host_config.h"
# echo "and comment out the following line with two backslashes //"
# echo "#error -- unsupported GNU version! gcc versions later than 5.3 are not supported!"
# deep learning setup with GTX 1080
# http://yangcha.github.io/GTX-1080/
cd $DIR