tensorflow: TensorFlow 1st Test: “could not open file to read NUMA node” - what's wrong?
I went to StackOverflow with this and was pointed back to Github. 😉 see [http://stackoverflow.com/questions/37067297]
Environment info
Operating System: Gentoo Linux on Lenovo P50
Installed version of CUDA and cuDNN: I installed dev-util/nvidia-cuda-toolkit package, version 7.5.18-r2
# ll /opt/cuda/lib/libcud*
-rw-r--r-- 1 root root 189082 May 6 10:42 /opt/cuda/lib/libcudadevrt.a
lrwxrwxrwx 1 root root 16 Sep 19 2015 /opt/cuda/lib/libcudart.so -> libcudart.so.7.5
lrwxrwxrwx 1 root root 19 Sep 19 2015 /opt/cuda/lib/libcudart.so.7.5 -> libcudart.so.7.5.18
-rwxr-xr-x 1 root root 311596 Sep 19 2015 /opt/cuda/lib/libcudart.so.7.5.18
-rw-r--r-- 1 root root 557240 May 6 10:42 /opt/cuda/lib/libcudart_static.a
Plus I installed cuDNN 5 downloaded from Nvidia
# ll libcud*
lrwxrwxrwx 1 rj rj 13 Mar 22 08:44 libcudnn.so -> libcudnn.so.5
lrwxrwxrwx 1 rj rj 17 Mar 22 08:44 libcudnn.so.5 -> libcudnn.so.5.0.4
-rwxrwxr-x 1 rj rj 59823168 Mar 22 02:37 libcudnn.so.5.0.4
-rw-rw-r-- 1 rj rj 58734618 Mar 22 02:37 libcudnn_static.a
If installed from binary pip package, provide:
Which pip package you installed.
# pip3 -V
pip 8.1.1 from /usr/lib64/python3.4/site-packages (python 3.4)
The output from python -c “import tensorflow; print(tensorflow.version)”.
# python -c "import tensorflow; print(tensorflow.__version__)"
I tensorflow/stream_executor/dso_loader.cc:105] successfully opened CUDA library libcublas.so locally
I tensorflow/stream_executor/dso_loader.cc:105] successfully opened CUDA library libcudnn.so locally
I tensorflow/stream_executor/dso_loader.cc:105] successfully opened CUDA library libcufft.so locally
I tensorflow/stream_executor/dso_loader.cc:105] successfully opened CUDA library libcuda.so.1 locally
I tensorflow/stream_executor/dso_loader.cc:105] successfully opened CUDA library libcurand.so locally
0.8.0
Steps to reproduce
- import tensorflow as tf
- hello = tf.constant(‘Hello, TensorFlow!’)
- sess = tf.Session()
What have you tried?
- stackoverflow 😉
About this issue
- Original URL
- State: closed
- Created 8 years ago
- Comments: 15 (7 by maintainers)
Hi all, I am stuck in the same error. I am working with tensorflow. If you found any solution, kindly share. Thanks
successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zeroHm. I’m afraid the path template “/sys/bus/pci/devices/%s/numa_node” cannot be found on my system as /sys/bus/pci/devices looks like that:
no such file or directory
Which it does - of course, because the kernel has no NUMA support. 😃 So I compiled a new 4.5.3 kernel with NUMA support and …
So it seems my problem was simply due to the fact, that the kernel had no NUMA support. I will now switch back to Hybrid mode (activate the Intel GPU) and test if that works with NUMA enabled.
edit: it does. So the “fix” could be some warning sign somewhere “NUMA support is a must”.
Officially you need a GPU with compute level >= 3.5, but see https://github.com/tensorflow/tensorflow/issues/25
On Thu, Aug 18, 2016 at 9:30 AM, abhijayghildyal notifications@github.com wrote: