keras specifies that the program trains instances on a particular card

Scenario: There are three cards on a particular machine, and you want to open three programs at the same time and put them on three cards for training.

Strategy: CUDA_VISIBLE_DEVICES=1 python Then you can specify that the program is trained on a certain card.

Additional knowledge:keras specifies GPU and graphics memory usage

Specify GPU

import os
["CUDA_VISIBLE_DEVICES"] = "0"

Specify GPU and video memory usage

import os
from .tensorflow_backend import set_session

["CUDA_VISIBLE_DEVICES"] = "0"
config = ()
config.gpu_options.per_process_gpu_memory_fraction = 0.3
set_session((config=config))

Specify GPU memory usage on demand

import .tensorflow_backend as KTF
import os

["CUDA_VISIBLE_DEVICES"] = "0"
config = ()
config.gpu_options.allow_growth=True 
sess = (config=config)
KTF.set_session(sess)

The above example of this keras designation program to train on a certain card is all that I have shared with you.