I have recently installed tensorflow (Windows CPU version) and received the following message:
Successfully installed tensorflow-1.4.0 tensorflow-tensorboard-0.4.0rc2
Then when I tried to run
import tensorflow as tf
hello = tf.constant('Hello, TensorFlow!')
sess = tf.Session()
sess.run(hello)
'Hello, TensorFlow!'
a = tf.constant(10)
b = tf.constant(32)
sess.run(a + b)
42
sess.close()
(which I found through https://github.com/tensorflow/tensorflow)
I received the following message:
2017-11-02 01:56:21.698935: I C:\tf_jenkins\home\workspace\rel-win\M\windows\PY\36\tensorflow\core\platform\cpu_feature_guard.cc:137] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX AVX2
But when I ran
import tensorflow as tf
hello = tf.constant('Hello, TensorFlow!')
sess = tf.Session()
print(sess.run(hello))
it ran as it should and output Hello, TensorFlow!, which indicates that the installation was successful indeed but there is something else that is wrong.
Do you know what the problem is and how to fix it?
What is this warning about?
Modern CPUs provide a lot of low-level instructions, besides the usual arithmetic and logic, known as extensions, e.g. SSE2, SSE4, AVX, etc. From the Wikipedia:
Advanced Vector Extensions (AVX) are extensions to the x86 instruction
set architecture for microprocessors from Intel and AMD proposed by
Intel in March 2008 and first supported by Intel with the Sandy
Bridge processor shipping in Q1 2011 and later on by AMD with the
Bulldozer processor shipping in Q3 2011. AVX provides new features,
new instructions and a new coding scheme.
In particular, AVX introduces fused multiply-accumulate (FMA) operations, which speed up linear algebra computation, namely dot-product, matrix multiply, convolution, etc. Almost every machine-learning training involves a great deal of these operations, hence will be faster on a CPU that supports AVX and FMA (up to 300%). The warning states that your CPU does support AVX (hooray!).
I'd like to stress here: it's all about CPU only.
Why isn't it used then?
Because tensorflow default distribution is built without CPU extensions, such as SSE4.1, SSE4.2, AVX, AVX2, FMA, etc. The default builds (ones from pip install tensorflow) are intended to be compatible with as many CPUs as possible. Another argument is that even with these extensions CPU is a lot slower than a GPU, and it's expected for medium- and large-scale machine-learning training to be performed on a GPU.
What should you do?
If you have a GPU, you shouldn't care about AVX support, because most expensive ops will be dispatched on a GPU device (unless explicitly set not to). In this case, you can simply ignore this warning by
# Just disables the warning, doesn't take advantage of AVX/FMA to run faster
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
... or by setting export TF_CPP_MIN_LOG_LEVEL=2 if you're on Unix. Tensorflow is working fine anyway, but you won't see these annoying warnings.
If you don't have a GPU and want to utilize CPU as much as possible, you should build tensorflow from the source optimized for your CPU with AVX, AVX2, and FMA enabled if your CPU supports them. It's been discussed in this question and also this GitHub issue. Tensorflow uses an ad-hoc build system called bazel and building it is not that trivial, but is certainly doable. After this, not only will the warning disappear, tensorflow performance should also improve.
Update the tensorflow binary for your CPU & OS using this command
pip install --ignore-installed --upgrade "Download URL"
The download url of the whl file can be found here
https://github.com/lakshayg/tensorflow-build
CPU optimization with GPU
There are performance gains you can get by installing TensorFlow from the source even if you have a GPU and use it for training and inference. The reason is that some TF operations only have CPU implementation and cannot run on your GPU.
Also, there are some performance enhancement tips that makes good use of your CPU. TensorFlow's performance guide recommends the following:
Placing input pipeline operations on the CPU can significantly improve performance. Utilizing the CPU for the input pipeline frees the GPU to focus on training.
For best performance, you should write your code to utilize your CPU and GPU to work in tandem, and not dump it all on your GPU if you have one.
Having your TensorFlow binaries optimized for your CPU could pay off hours of saved running time and you have to do it once.
For Windows, you can check the official Intel MKL optimization for TensorFlow wheels that are compiled with AVX2. This solution speeds up my inference ~x3.
conda install tensorflow-mkl
For Windows (Thanks to the owner f040225), go to here: https://github.com/fo40225/tensorflow-windows-wheel to fetch the url for your environment based on the combination of "tf + python + cpu_instruction_extension". Then use this cmd to install:
pip install --ignore-installed --upgrade "URL"
If you encounter the "File is not a zip file" error, download the .whl to your local computer, and use this cmd to install:
pip install --ignore-installed --upgrade /path/target.whl
If you use the pip version of TensorFlow, it means it's already compiled and you are just installing it. Basically you install TensorFlow-GPU, but when you download it from the repository and trying to build it, you should build it with CPU AVX support. If you ignore it, you will get a warning every time when you run on the CPU. You can take a look at those too.
Proper way to compile Tensorflow with SSE4.2 and AVX
What is AVX Cpu support in tensorflow
The easiest way that I found to fix this is to uninstall everything then install a specific version of tensorflow-gpu:
uninstall tensorflow:
pip uninstall tensorflow
uninstall tensorflow-gpu: (make sure to run this even if you are not sure if you installed it)
pip uninstall tensorflow-gpu
Install specific tensorflow-gpu version:
pip install tensorflow-gpu==2.0.0
pip install tensorflow_hub
pip install tensorflow_datasets
You can check if this worked by adding the following code into a python file:
from __future__ import absolute_import, division, print_function, unicode_literals
import numpy as np
import tensorflow as tf
import tensorflow_hub as hub
import tensorflow_datasets as tfds
print("Version: ", tf.__version__)
print("Eager mode: ", tf.executing_eagerly())
print("Hub Version: ", hub.__version__)
print("GPU is", "available" if tf.config.experimental.list_physical_devices("GPU") else "NOT AVAILABLE")
Run the file and then the output should be something like this:
Version: 2.0.0
Eager mode: True
Hub Version: 0.7.0
GPU is available
Hope this helps
What worked for me tho is this library https://pypi.org/project/silence-tensorflow/
Install this library and do as instructed on the page, it works like a charm!
Try using anaconda. I had the same error. One lone option was to build tensorflow from source which took long time. I tried using conda and it worked.
Create a new environment in anaconda.
conda install -c conda-forge tensorflow
Then, it worked.
He provided once the list, deleted by someone but see the answer is
Download packages list
Output:
F:\temp\Python>python test_tf_logics_.py
[0, 0, 26, 12, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 0]
[ 0 0 0 26 12 0 0 0 2 0 0 0 0 0 0 0 0]
[ 0 0 26 12 0 0 0 2 0 0 0 0 0 0 0 0 0]
2022-03-23 15:47:05.516025: I tensorflow/core/platform/cpu_feature_guard.cc:151] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX AVX2
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2022-03-23 15:47:06.161476: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1525] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 10 MB memory: -> device: 0, name: NVIDIA GeForce GTX 1060 6GB, pci bus id: 0000:01:00.0, compute capability: 6.1
[0 0 2 0 0 0 0 7 0 0 0 0 0 0 0 0 0]
...
As the message says, your CPU supports instructions that TensorFlow binary was not compiled to use. This should not be an issue with CPU version of TensorFlow since it does not perform AVX (Advanced Vector Extensions) instructions.
However, it seems that TensorFlow uses AVX instructions in some parts of the code and the message is just a warning, you can safely ignore it.
You can compile your own version of TensorFlow with AVX instructions.
My system is ubuntu 16.04 version my laptop is dell Inspiron-5521 and it has intel graphic card but tensorflow needs nvidia graphics for cuda support.
Is there any way where i can run tensorflow with GPU(with CPU is working) on intel graphics.
During installation of tensorflow-gpu i have no error when i import i get
"
Failed to load the native TensorFlow runtime
."
Did some digging then found to install cuda downloaded the "cuda_9.1.85_387.26_linux.run" file but faces issues while running it
"Detected 4 CPUs online; setting concurrency level to 4.
The file '/tmp/.X0-lock' exists and appears to contain the process ID
'1033' of a runnning X server.
It appears that an X server is running. Please exit X before
installation. If you're sure that X is not running, but are getting
this error, please delete any X lock files in /tmp."
Deleted files from tmp folder and tried still same issue.
To run tensorflow-gpu you need nvidia card. You'll need to stick to running normal tensorflow on CPU.
Is Intel based graphic card compatible with tensorflow/GPU?
Tensorflow does not support OpenCL API that you can use with Intel or AMD, only CUDA. CUDA is a proprietary NVidia technology that only works with NVidia GPUs.
You may like to search for machine learning frameworks that utilise OpenCL, but I only find some niche projects at the moment.
I had to switch from AMD to NVidia to be able to run Tensorflow calculations on GPU.
My tests show that Tensorflow GPU operations are ~6% slower on Python 3 compared to Python 2. Does anyone have any insight on this?
Platform:
Ubuntu 16.04.2 LTS
Virtualenv 15.0.1
Python 2.7.12
Python 3.6.1
TensorFlow 1.1
CUDA Toolkit 8.0.44
CUDNN 5.1
GPU: GTX 980Ti
CPU: i7 4 GHz
RAM: 32 GB
When operating Tensorflow from python most code to feed the computational engine with data resides in python domain. There are known differences between python 2/3 when it comes to performance on various tasks. Therefore, I'd guess that the python code you use to feed the net (or TF python layer, which is quite thick) makes heavy use of python features that are (by design) a bit slower in python 3.
I have problems setting up tensorflow with python 3.5.3 on Windows 7. The CUDA version is 8.0 and the GPU driver is updated to the latest version.
On the machine I used, there are one Geforce titan z (two cores) and one 750ti installed on the machine. I kept getting error msgs saying the "peer access is not supported" between the two cores on the titan z card and between the titan z and 750ti.
I am wondering:
1. Is it possible to use only one GPU for the tensorflow, and let it work?
2. Is it possible to use both cores in the titan z card for tensorflow?
3. Any suggestions on the python version and anaconda version for tensorflow? Will a Linux environment solve these problems?
Thank you!
It is possible to use a single or multiple GPUs and specify which one to use for each operation (link). TensorFlow has a requirement of compute capability 3.5 for an eligible GPU. Python/linux environment doesn't affect how GPU is used.
I was setting up python and theano for use with gpu on;
ubuntu 14.04,
GeForce GTX 1080
already installed NVIDIA driver (367.27) and CUDA toolkit (7.5) successfully for the system,
but on testing with theano gpu implementation I get the above error (for example; when importing theano with gpu enabled)
I have tried to look for possible solutions but didn't succeed.
I'm a little new to ubuntu and gpu programming, so I would appreciate any insight into how I can solve this problem.
Thanks
As Robert Crovella said, SM 6.1 (sm_61) is only supported in CUDA 8.0 and above, and thus you should download CUDA 8.0 Release Candidate from https://developer.nvidia.com/cuda-toolkit
Ubuntu 14.04 is supported, and the instructions on the website on how to setup should be straightforward (copy and paste lines to the console).
I would also recommend downloading CUDA 8.0 when it comes out, since the RC is not the final version.
I was able to find a solution to this problem (since I still want to use CUDA 7.5) by including the following line in the .theanorc file
flags = -arch=sm_52
no more nvcc fatal error