Difference between revisions of "Tensorflow with gpu"

From ElphelWiki
Jump to: navigation, search
(20 intermediate revisions by the same user not shown)
Line 1: Line 1:
−
==Pre==
+
==Requirements==
−
* check device
+
* Kubuntu 16.04.4 LTS
−
  ~$ lspci | grep NVIDIA
+
==Setup==
 +
* Check device
 +
  <font size='2'><b>~$ lspci | grep NVIDIA</b>
 
  81:00.0 VGA compatible controller: NVIDIA Corporation GF119 [GeForce GT 610] (rev a1)
 
  81:00.0 VGA compatible controller: NVIDIA Corporation GF119 [GeForce GT 610] (rev a1)
−
  81:00.1 Audio device: NVIDIA Corporation GF119 HDMI Audio Controller (rev a1)
+
  81:00.1 Audio device: NVIDIA Corporation GF119 HDMI Audio Controller (rev a1)</font>
−
* check driver version
+
 
−
  ~$ cat /proc/driver/nvidia/version
+
* Check driver version:
 +
  <font size='2'><b>~$ cat /proc/driver/nvidia/version</b>
 
  NVRM version: NVIDIA UNIX x86_64 Kernel Module  387.26  Thu Nov  2 21:20:16 PDT 2017
 
  NVRM version: NVIDIA UNIX x86_64 Kernel Module  387.26  Thu Nov  2 21:20:16 PDT 2017
−
  GCC version:  gcc version 5.4.0 20160609 (Ubuntu 5.4.0-6ubuntu1~16.04.9)
+
  GCC version:  gcc version 5.4.0 20160609 (Ubuntu 5.4.0-6ubuntu1~16.04.9)</font>
−
* install cuda 9.0 with patches
 
−
https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1604&target_type=deblocal
 
−
* then
 
−
sudo apt-get install cuda-9-2 # included on that NVidia page
 
  
−
  https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#post-installation-actions
+
* Install cuda 9.2 with patch(es):
−
*
+
  <font size='2'># https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1604&target_type=deblocal:
−
  ~$ export PATH=/usr/local/cuda-9.2/bin${PATH:+:${PATH}}
+
<b>~$ sudo dpkg -i cuda-repo-ubuntu1604-9-2-local_9.2.88-1_amd64.deb
−
  ~$ export LD_LIBRARY_PATH=/usr/local/cuda-9.2/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
+
  ~$ sudo apt-key add /var/cuda-repo-9-2-local/7fa2af80.pub
−
*
+
~$ sudo apt-get update
−
  ~$ nvidia-smi  
+
  ~$ sudo apt-get install cuda</b>
−
  Thu Apr 26 12:39:25 2018       
+
<b># INSTALL THE PATCH(ES)</b></font>
 +
 
 +
* Might need to reboot PC. If cuda 9.2 got installed over other version, nvidia tools will be throwing errors about driver  versions mismatching, try
 +
  <font size='2'><b>~$ nvidia-smi</b></font>
 +
Good looking output:
 +
  <font size='1'>Wed Jun 13 15:55:44 2018       
 
  +-----------------------------------------------------------------------------+
 
  +-----------------------------------------------------------------------------+
−
  | NVIDIA-SMI 387.26                Driver Version: 387.26                    |
+
  | NVIDIA-SMI 396.26                Driver Version: 396.26                    |
 
  |-------------------------------+----------------------+----------------------+
 
  |-------------------------------+----------------------+----------------------+
 
  | GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
 
  | GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
 
  | Fan  Temp  Perf  Pwr:Usage/Cap|        Memory-Usage | GPU-Util  Compute M. |
 
  | Fan  Temp  Perf  Pwr:Usage/Cap|        Memory-Usage | GPU-Util  Compute M. |
 
  |===============================+======================+======================|
 
  |===============================+======================+======================|
−
  |  0  GeForce GT 610      Off  | 00000000:81:00.0 N/A |                  N/A |
+
  |  0  GeForce GTX 750 Ti  Off  | 00000000:01:00.0 On |                  N/A |
−
  | N/A   31C   P8   N/A /  N/A |    148MiB /   956MiB |     N/A     Default |
+
  | 33%   36C   P8     1W /  46W |    229MiB / 2000MiB |     0%     Default |
−
  +-------------------------------+----------------------+----------------------+
+
  +-------------------------------+----------------------+----------------------+                                                                            
−
                                                                               
 
 
  +-----------------------------------------------------------------------------+
 
  +-----------------------------------------------------------------------------+
 
  | Processes:                                                      GPU Memory |
 
  | Processes:                                                      GPU Memory |
 
  |  GPU      PID  Type  Process name                            Usage      |
 
  |  GPU      PID  Type  Process name                            Usage      |
 
  |=============================================================================|
 
  |=============================================================================|
−
  |    0                   Not Supported                                      |
+
  |    0     1305      G  /usr/lib/xorg/Xorg                          136MiB |
−
  +-----------------------------------------------------------------------------+
+
|    0      3587      G  /usr/bin/krunner                              1MiB |
 +
|    0      3590      G  /usr/bin/plasmashell                          67MiB |
 +
|    0      3693      G  /usr/bin/plasma-discover                      20MiB |
 +
  +-----------------------------------------------------------------------------+</font>
 +
 
 +
* Check out post installation docs:
 +
<font size='2'>https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#post-installation-actions:
 +
# Export paths
 +
<b>~$ export PATH=/usr/local/cuda-9.2/bin${PATH:+:${PATH}}
 +
~$ export LD_LIBRARY_PATH=/usr/local/cuda-9.2/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}</b></font>
 +
 
 +
* Install TensorFlow (build from sources for cuda 9.2):
 +
<font size='2'>link 1: (preferrable guide): http://www.python36.com/install-tensorflow141-gpu/
 +
link 2: https://www.tensorflow.org/install/install_sources</font>
 +
 
 +
* '''[Optional]''' Install TensorFlow (prebuilt for cuda 9.0?):
 +
<font size='2'># docs:
 +
# - https://www.tensorflow.org/install/install_linux
 +
# some instructions:
 +
# - install cuDNN
 +
<b>~$ sudo apt-get install python3-pip # if it is not already installed</b>
 +
<b>~$ sudo pip3 install --ignore-installed --upgrade https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.7.0-cp35-cp35m-linux_x86_64.whl</b>
  
−
* install tensorflow
+
==Testing setup==
−
~$ sudo apt-get install python3-pip # if it is not already installed
 
  
−
  ~$ sudo pip3 install --ignore-installed --upgrade https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.7.0-cp35-cp35m-linux_x86_64.whl
+
* Supported card '''GeForce GTX 750 Ti''' (<b>list of [https://developer.nvidia.com/cuda-gpus supported graphic cards]</b>):
 +
  <font size='2'><b>~$ python3</b>
 +
Python 3.5.2 (default, Nov 23 2017, 16:37:01)
 +
[GCC 5.4.0 20160609] on linux
 +
Type "help", "copyright", "credits" or "license" for more information.
 +
'''>>> import tensorflow as tf'''
 +
'''>>> hello = tf.constant('Hello, World!')''' 
 +
'''>>> sess = tf.Session()'''
 +
2018-04-26 18:14:05.427668: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:898] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning
 +
NUMA node zero
 +
2018-04-26 18:14:05.428033: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1344] Found device 0 with properties:
 +
name: GeForce GTX 750 Ti major: 5 minor: 0 memoryClockRate(GHz): 1.1105
 +
pciBusID: 0000:01:00.0
 +
totalMemory: 1.95GiB freeMemory: 1.53GiB
 +
2018-04-26 18:14:05.428061: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1423] Adding visible gpu devices: 0
 +
2018-04-26 18:14:05.927106: I tensorflow/core/common_runtime/gpu/gpu_device.cc:911] Device interconnect StreamExecutor with strength 1 edge matrix:
 +
2018-04-26 18:14:05.927149: I tensorflow/core/common_runtime/gpu/gpu_device.cc:917]      0
 +
2018-04-26 18:14:05.927163: I tensorflow/core/common_runtime/gpu/gpu_device.cc:930] 0:  N
 +
2018-04-26 18:14:05.927313: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1041] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 1289 MB memory) -> physical GPU (device: 0,
 +
name: GeForce GTX 750 Ti, pci bus id: 0000:01:00.0, compute capability: 5.0)
 +
'''>>> print(sess.run(hello))'''
 +
b'Hello, World!'</font>
  
−
* testing: unsupported card '''GeForce GT 610'''
+
* Unsupported card '''GeForce GT 610'''
 
  ~$ python3
 
  ~$ python3
 
  Python 3.5.2 (default, Nov 23 2017, 16:37:01)  
 
  Python 3.5.2 (default, Nov 23 2017, 16:37:01)  
Line 62: Line 106:
 
  b'Hello, World!'
 
  b'Hello, World!'
  
−
* testing: supported card '''GeForce GTX 750 Ti'''
+
* As a quickfix had to install CuDNN 7.0.5 instead of latest:
−
~$ python3
+
  https://stackoverflow.com/questions/49960132/cudnn-library-compatibility-error-after-loading-model-weights
−
Python 3.5.2 (default, Nov 23 2017, 16:37:01)
 
−
[GCC 5.4.0 20160609] on linux
 
−
Type "help", "copyright", "credits" or "license" for more information.
 
−
>>> import tensorflow as tf
 
−
>>> hello = tf.constant('Hello, World!') 
 
−
>>> sess = tf.Session()
 
−
2018-04-26 18:14:05.427668: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:898] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning
 
−
NUMA node zero
 
−
2018-04-26 18:14:05.428033: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1344] Found device 0 with properties:
 
−
name: GeForce GTX 750 Ti major: 5 minor: 0 memoryClockRate(GHz): 1.1105
 
−
  pciBusID: 0000:01:00.0
 
−
totalMemory: 1.95GiB freeMemory: 1.53GiB
 
−
2018-04-26 18:14:05.428061: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1423] Adding visible gpu devices: 0
 
−
2018-04-26 18:14:05.927106: I tensorflow/core/common_runtime/gpu/gpu_device.cc:911] Device interconnect StreamExecutor with strength 1 edge matrix:
 
−
2018-04-26 18:14:05.927149: I tensorflow/core/common_runtime/gpu/gpu_device.cc:917]      0
 
−
2018-04-26 18:14:05.927163: I tensorflow/core/common_runtime/gpu/gpu_device.cc:930] 0:  N
 
−
2018-04-26 18:14:05.927313: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1041] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 1289 MB memory) -> physical GPU (device: 0,
 
−
name: GeForce GTX 750 Ti, pci bus id: 0000:01:00.0, compute capability: 5.0)
 
−
>>> print(sess.run(hello))
 
−
b'Hello, World!'
 
  
−
* Supported cards:
+
* Print tensorflow version
−
  https://developer.nvidia.com/cuda-gpus
+
  <font size='2'>'''>>> print(tf.__version__)'''</font>
  
−
* As a quickfix had to install CuDNN 7.0.5 instead of latest:
+
==Problems==
−
  https://stackoverflow.com/questions/49960132/cudnn-library-compatibility-error-after-loading-model-weights
+
* <font color='green'>[SOLVED]</font> <b>AttributeError: '_NamespacePath' object has no attribute 'sort'</b>
 +
<font size='2'># Notes:
 +
  After updating some packages probably. python3?
 +
# How to reproduce:
 +
  1:
 +
<b>~$ python3
 +
>>> import tensorflow</b>
 +
2:
 +
<b>~$ virtualenv --system-site-packages -p python3</b>
 +
# Solution:
 +
<b>~$ sudo pip3 install setuptools --upgrade</b></font>

Revision as of 15:58, 13 June 2018

Requirements

  • Kubuntu 16.04.4 LTS

Setup

  • Check device
~$ lspci | grep NVIDIA
81:00.0 VGA compatible controller: NVIDIA Corporation GF119 [GeForce GT 610] (rev a1)
81:00.1 Audio device: NVIDIA Corporation GF119 HDMI Audio Controller (rev a1)
  • Check driver version:
~$ cat /proc/driver/nvidia/version
NVRM version: NVIDIA UNIX x86_64 Kernel Module  387.26  Thu Nov  2 21:20:16 PDT 2017
GCC version:  gcc version 5.4.0 20160609 (Ubuntu 5.4.0-6ubuntu1~16.04.9)
  • Install cuda 9.2 with patch(es):
# 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-ubuntu1604-9-2-local_9.2.88-1_amd64.deb
~$ sudo apt-key add /var/cuda-repo-9-2-local/7fa2af80.pub
~$ sudo apt-get update
~$ sudo apt-get install cuda
# INSTALL THE PATCH(ES)
  • Might need to reboot PC. If cuda 9.2 got installed over other version, nvidia tools will be throwing errors about driver versions mismatching, try
~$ nvidia-smi

Good looking output: 

Wed Jun 13 15:55:44 2018       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 396.26                 Driver Version: 396.26                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 750 Ti  Off  | 00000000:01:00.0  On |                  N/A |
| 33%   36C    P8     1W /  46W |    229MiB /  2000MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+                                                                             
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0      1305      G   /usr/lib/xorg/Xorg                           136MiB |
|    0      3587      G   /usr/bin/krunner                               1MiB |
|    0      3590      G   /usr/bin/plasmashell                          67MiB |
|    0      3693      G   /usr/bin/plasma-discover                      20MiB |
+-----------------------------------------------------------------------------+
  • Check out post installation docs:
https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#post-installation-actions:
# Export paths
~$ export PATH=/usr/local/cuda-9.2/bin${PATH:+:${PATH}}
~$ export LD_LIBRARY_PATH=/usr/local/cuda-9.2/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
  • Install TensorFlow (build from sources for cuda 9.2):
link 1: (preferrable guide): http://www.python36.com/install-tensorflow141-gpu/
link 2: https://www.tensorflow.org/install/install_sources
  • [Optional] Install TensorFlow (prebuilt for cuda 9.0?):
# docs: 
# - https://www.tensorflow.org/install/install_linux
# some instructions:
# - install cuDNN
~$ sudo apt-get install python3-pip # if it is not already installed
~$ sudo pip3 install --ignore-installed --upgrade https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.7.0-cp35-cp35m-linux_x86_64.whl

Testing setup

~$ python3
Python 3.5.2 (default, Nov 23 2017, 16:37:01) 
[GCC 5.4.0 20160609] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow as tf
>>> hello = tf.constant('Hello, World!')  
>>> sess = tf.Session()
2018-04-26 18:14:05.427668: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:898] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning 
NUMA node zero
2018-04-26 18:14:05.428033: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1344] Found device 0 with properties: 
name: GeForce GTX 750 Ti major: 5 minor: 0 memoryClockRate(GHz): 1.1105
pciBusID: 0000:01:00.0
totalMemory: 1.95GiB freeMemory: 1.53GiB
2018-04-26 18:14:05.428061: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1423] Adding visible gpu devices: 0
2018-04-26 18:14:05.927106: I tensorflow/core/common_runtime/gpu/gpu_device.cc:911] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-04-26 18:14:05.927149: I tensorflow/core/common_runtime/gpu/gpu_device.cc:917]      0 
2018-04-26 18:14:05.927163: I tensorflow/core/common_runtime/gpu/gpu_device.cc:930] 0:   N 
2018-04-26 18:14:05.927313: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1041] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 1289 MB memory) -> physical GPU (device: 0, 
name: GeForce GTX 750 Ti, pci bus id: 0000:01:00.0, compute capability: 5.0)
>>> print(sess.run(hello))
b'Hello, World!'
  • Unsupported card GeForce GT 610
~$ python3
Python 3.5.2 (default, Nov 23 2017, 16:37:01) 
[GCC 5.4.0 20160609] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow as tf
>>> hello = tf.constant('Hello, World!')                                                                                                                                                                            
>>> sess = tf.Session()                                                                                                                                                                                                 
2018-04-26 13:00:19.050625: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA                                           
2018-04-26 13:00:19.181581: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1344] Found device 0 with properties:                                                                                                             
name: GeForce GT 610 major: 2 minor: 1 memoryClockRate(GHz): 1.62                                                                                                                                                                
pciBusID: 0000:81:00.0                                                                                                                                                                                                           
totalMemory: 956.50MiB freeMemory: 631.69MiB                                                                                                                                                                                              
2018-04-26 13:00:19.181648: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1394] Ignoring visible gpu device (device: 0, name: GeForce GT 610, pci bus id: 0000:81:00.0, compute capability: 2.1) with Cuda compute capability 2.1. The minimum required Cuda capability is 3.5.                                                                                                                                                                                                              
2018-04-26 13:00:19.181669: I tensorflow/core/common_runtime/gpu/gpu_device.cc:911] Device interconnect StreamExecutor with strength 1 edge matrix:                                                                                       
2018-04-26 13:00:19.181683: I tensorflow/core/common_runtime/gpu/gpu_device.cc:917]      0                                                                                                                                                       
2018-04-26 13:00:19.181695: I tensorflow/core/common_runtime/gpu/gpu_device.cc:930] 0:   N                                                                                                                                                       
>>> print(sess.run(hello))                                                                                                                                                                                                                             
b'Hello, World!'
  • As a quickfix had to install CuDNN 7.0.5 instead of latest:
https://stackoverflow.com/questions/49960132/cudnn-library-compatibility-error-after-loading-model-weights
  • Print tensorflow version
>>> print(tf.__version__)

Problems

  • [SOLVED] AttributeError: '_NamespacePath' object has no attribute 'sort'
# Notes:
 After updating some packages probably. python3?
# How to reproduce:
1:
~$ python3
>>> import tensorflow
2:
~$ virtualenv --system-site-packages -p python3
# Solution:
~$ sudo pip3 install setuptools --upgrade
Retrieved from "https://wiki.elphel.com/index.php?title=Tensorflow_with_gpu&oldid=16042"