tensorflow设置gpu及gpu显存使用

1.查看GPU

     终端： nvidia-smi

2.在终端执行程序时指定GPU   

CUDA_VISIBLE_DEVICES=1   python  your_file.py

1:  表示用哪块GPU

CUDA_VISIBLE_DEVICES=1           Only device 1 will be seen
CUDA_VISIBLE_DEVICES=0,1         Devices 0 and 1 will be visible
CUDA_VISIBLE_DEVICES="0,1"       Same as above, quotation marks are optional
CUDA_VISIBLE_DEVICES=0,2,3       Devices 0, 2, 3 will be visible; device 1 is masked
CUDA_VISIBLE_DEVICES=""          No GPU will be visible

3.在Python代码中指定GPU

import os
os.environ["CUDA_VISIBLE_DEVICES"] = "0"

4.设置定量的GPU使用量

config = tf.ConfigProto() 
config.gpu_options.per_process_gpu_memory_fraction = 0.9 # 占用GPU90%的显存 
session = tf.Session(config=config)

5. 设置最小的GPU使用量

config = tf.ConfigProto() 
config.gpu_options.allow_growth = True 
session = tf.Session(config=config)

转载：https://blog.csdn.net/guvcolie/article/details/77164230