【发布时间】:2020-12-02 20:39:34
【问题描述】:
我有一个 Windows 10 操作系统,我在 Jupyter Notebook 上使用 Tensorflow gpu 版本。
conda 4.9.2
python 3.7
tf '2.3.1'
cudatoolkit 11.0.221
tf.config.experimental.list_physical_devices()
返回:
[PhysicalDevice(name='/physical_device:CPU:0', device_type='CPU'),
PhysicalDevice(name='/physical_device:XLA_CPU:0', device_type='XLA_CPU'),
PhysicalDevice(name='/physical_device:XLA_GPU:0', device_type='XLA_GPU')]
和,
from tensorflow.python.client import device_lib
print(device_lib.list_local_devices())
返回:
[name: "/device:CPU:0"
device_type: "CPU"
memory_limit: 268435456
locality {
}
incarnation: 10428669251348137268
, name: "/device:XLA_CPU:0"
device_type: "XLA_CPU"
memory_limit: 17179869184
locality {
}
incarnation: 2248949917928228630
physical_device_desc: "device: XLA_CPU device"
, name: "/device:XLA_GPU:0"
device_type: "XLA_GPU"
memory_limit: 17179869184
locality {
}
incarnation: 17815507612308274706
physical_device_desc: "device: XLA_GPU device"
]
但是,当我想拟合模型时,它开始使用 100% 的 CPU 而不是 GPU。
以前有人遇到过这个问题吗?
【问题讨论】:
-
你需要 cudatoolkit 10.1.243 和 cudnn 7.6.5
标签: python tensorflow keras