0

I am just starting to play around with tensorflow (GPU) on Ubuntu 16.04 and have followed the installation instructions here Installing Tensorflow. After running the example, where you create the session (sess = tf.Session())I get the following output:

2018-04-23 14:51:29.548236: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2018-04-23 14:51:29.659644: 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
2018-04-23 14:51:29.659965: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1344] Found device 0 with properties: 
name: GeForce GTX 1060 6GB major: 6 minor: 1 memoryClockRate(GHz): 1.7085
pciBusID: 0000:01:00.0
totalMemory: 5.93GiB freeMemory: 5.04GiB
2018-04-23 14:51:29.659980: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1423] Adding visible gpu devices: 0
2018-04-23 14:51:29.840482: I tensorflow/core/common_runtime/gpu/gpu_device.cc:911] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-04-23 14:51:29.840526: I tensorflow/core/common_runtime/gpu/gpu_device.cc:917]      0 
2018-04-23 14:51:29.840532: I tensorflow/core/common_runtime/gpu/gpu_device.cc:930] 0:   N 
2018-04-23 14:51:29.840728: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1041] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 4802 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1060 6GB, pci bus id: 0000:01:00.0, compute capability: 6.1)

which is extremely overwhelming and confusing. Simple models run fine, so I don't think these are errors. I can't seem to find any examples of where this happens and wanted to know is this normal? Have I installed something incorrectly from what you can see in this output?

BartoszKP
  • 34,786
  • 15
  • 102
  • 130
Jakey
  • 160
  • 1
  • 2
  • 10
  • What exactly is overwhelming and confusing? – BartoszKP Apr 23 '18 at 07:31
  • The lack of any (from what I can find) documentation about what this means or that you should even expect this. Also a note saying "All good" or something would be nice.. haha – Jakey Apr 23 '18 at 07:32
  • What this means? Which part don't you understand? I'm sorry but it's hard to see what's the actual question here. This looks like a typical set of information logs. There are no errors, not even warnings - what is the problem? – BartoszKP Apr 23 '18 at 07:41
  • No need to be so aggressive buddy. I "wanted to know is this normal" and if "I installed something incorrectly from what you can see in this output" as the question asks. I'm guessing that this is normal from your comment, but I have seen no mention of these information logs in the installation instructions. The only example I have seen looks completely different with logs ending in " successfully opened CUDA library libcurand.so locally" – Jakey Apr 23 '18 at 08:02
  • Sorry I sounded aggressive, that's not my intent. What you list are two different questions, none of which seems appropriate for SO. But maybe not completely inappropriate. What I'm saying that if you focus on one thing exactly and clearly formulate what is your problem you have greater chance of someone answering. In this form your question seems to state that you are confused and it's hard to decide what to do about it. – BartoszKP Apr 23 '18 at 08:09
  • I get that, but in this case I feel it's pretty clear what I am asking, which is what I asked in my original question and restated in my last comment. I see @AlexanderHarnisch has answered my question, so thanks for your input and good day, sir. – Jakey Apr 23 '18 at 08:16
  • Good for you :) Unfortunately it happens that some people jump in and answer loosely formulated questions. But maybe it's just me, and indeed your question is a good one :) Time will tell from votes (I'll refrain from voting). Cheers! – BartoszKP Apr 23 '18 at 08:21

2 Answers2

1

No you have done nothing wrong, this is expected behavior and often times actually really helpful when running TensorFlow code on a bunch of different systems, especially cluster nodes.

If you are just annoyed by this while testing on your local system you can supress the debugging output by doing something like this: 

import os
import tensorflow as tf
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
Alexander Harnisch
  • 624
  • 3
  • 8
  • Ah, great. That's a relief! Thanks for the tip. – Jakey Apr 23 '18 at 08:19
  • 1
    I am glad to help. But also keep in mind that this might suppress some error messages you need for debugging. Read [this](https://stackoverflow.com/questions/35911252/disable-tensorflow-debugging-information?utm_medium=organic&utm_source=google_rich_qa&utm_campaign=google_rich_qa) or directly the [source](https://github.com/tensorflow/tensorflow/issues/1258). Setting the log level to 1 should be enough, that way only info messages are being suppressed and you still get warnings and errors. – Alexander Harnisch Apr 23 '18 at 08:32
0

This issue is seconded. I just got this same result doing the x64 install with an i3-4160 and GTX1060.

This warning/information seems to happen almost every single line I run using TensorFlow. Since those are likely not far from the most common install hardware, I was surprised it was so hard to find the response for this.

Also, the TensorFlow installation guide could allude to the version specificity a bit better. I found that tidbit elsewhere on StackOverflow.

Daniel Lewis
  • 1