how can use some(numbers) cores of cpu from all cores, with tensorflow attributes

user_14235213 0
Viewed 257

I'm trying to use the university server for my deep code, all CPU's core on the server is 64 but I have to use just 24 cores to everybody can use the server too. I try to limit my CPU resource. I search all StackOverflow to find a solution but all suggestion doesn't work for me for example downgrade tensorflow and use

config = tf.ConfigProto(allow_soft_placement=True, 
                        intra_op_parallelism_threads=ncpu,
                        inter_op_parallelism_threads=ncpu)

and some others solutions by using

import tensorflow as tf

tf.config.threading.set_intra_op_parallelism_threads(numb)
tf.config.threading.set_inter_op_parallelism_threads(numb)

I have to use TensorFlow version 2 or higher because I use 'kerastuner' package in my code

tensorflow deep-learning python
1 Answers

If you have Admin rights on the server and its running a Version of Windows, you can simply restrict the resources via the task-manager.

If you want to do it by code... It looks like its a bug in Tensorflow, which might be fixed, regarding to the github issue.

You might want to try:

export OMP_NUM_THREADS=2
tf.config.threading.set_intra_op_parallelism_threads(2)
tf.config.threading.set_inter_op_parallelism_threads(1)

As this was reported working by Leslie-Fang. If this does not work for you, I guess your only option is to join the github discussion, until its fixed.

user_138044430
Related