My CPU doesn't support Tensorflow AVX instructions
bertagknowles opened this issue · comments
I was able to install all the requirements. However while generating samples, getting the following error. I have an Intel i3 First gen Processor and running Ubuntu 18.
2019-02-16 03:12:49.453982: F tensorflow/core/platform/cpu_feature_guard.cc:37] The TensorFlow library was compiled to use AVX instructions, but these aren't available on your machine.
Aborted (core dumped)
I then installed Tensorflow 1.5 (pip3 install tensorflow==1.5). The sample was generated, however another warning popped up as shown below. Will this affect the quality? Do I need to compile TensorFlow on my system?
2019-02-16 03:22:19.785441: I tensorflow/core/platform/cpu_feature_guard.cc:137] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2
This piece of documentation by Microsoft may help you with AVX, at least if you are using Windows 7. On Linux it depends on whether your kernel supports the instructions or not (I believe, not sure), you may need to check that.
Regarding SSE4.1 and SSE4.2, if you compile the library yourself and you are using GCC, you may need to use some flags (depending on the version of the compiler) to enable the instructions. The contents of this issue can help you in this case. This instructions started to be introduced on the Core 2 Duo processors, so if yours is older than that (it is not), you may be out of luck.
Thx for the suggestions. Can I use a precompiled wheel from https://github.com/mind/wheels ?
I suppose you can, I don't see any reason why you wouldn't.
The mind wheels may not work, because they are a bit old. Maybe one of these is useful: https://github.com/yaroslavvb/tensorflow-community-wheels/issues
And its no problem if the warning pops up that your cpu supports instructions that are not in the build. It only means that tensorflow could run faster when using the wright wheel.
I ran into the exact same problem with Ubuntu Server running on top of Hyper-V. I'm going to switch to running on Ubuntu Server on hardware to see if it makes a difference.
Switched to Ubuntu Server running on bare hardware (Dell R710) and got this message: 2019-02-16 17:07:14.788642: F tensorflow/core/platform/cpu_feature_guard.cc:37] The TensorFlow library was compiled to use AVX instructions, but these aren't available on your machine.
Anyone found a workaround?
OK -- switched to an Ubuntu Server virtual machine on Hyper-V on Windows 10 running on an Intel i7 -- everything is working and producing fascinating text. :)
I get this same error but it ends up working if I wait it out
I get this same error but it ends up working if I wait it out
Yes, even on my virtual machine, the error is there but after some time the output is shown. Just want to know if resolving this warning will affect the quality of the output. I don't know how to compile the tensorflow binary and which flags to use.
It should not affect the quality, but only the speed. Those instructions are optimizations to run faster floating point operations.
Has anyone figured out how to prompt the system with sentences, as in the examples?
Has anyone figured out how to prompt the system with sentences, as in the examples?
This is off topic, but yeah. if you start the model in conditional mode, it will give you a prompt to enter your sentences.
python3 src/interactive_conditional_samples.py --top_k 40
closing, feel free to reopen if there's an actual issue to resolve
Try Anaconda. It has TensorFlow distribution for old CPU.
The CPU its self also needs to support the instructions sets. My Legacy Intel(R) Xeon(R) CPU E5520 in my r710 does not. Stumbled upon this: https://www.tensorflow.org/install/pip#hardware-requirements The link takes you to compatible CPUs.