python - 为什么tensorflow在“成功打开动态库libcudart.so.10.1”后会暂停3分钟？

Question

为什么tensorflow在“成功打开动态库libcudart.so.10.1”和“设备互连StreamExecutor与强度1边缘矩阵”之间有3分钟的停顿？

相关代码（不完整）：

from keras.models import Sequential

cnn = Sequential()
cnn.add(Convolution1D(64, 3, padding="same",activation="relu",input_shape=(43, 1)))
cnn.add(Convolution1D(64, 3, padding="same", activation="relu"))
cnn.add(MaxPooling1D(pool_size=(2)))
cnn.add(Convolution1D(128, 3, padding="same", activation="relu"))
cnn.add(Convolution1D(128, 3, padding="same", activation="relu"))
cnn.add(MaxPooling1D(pool_size=(2)))
cnn.add(LSTM(lstm_output_size))
cnn.add(Dropout(0.1))
cnn.add(Dense(1, activation="sigmoid"))

cnn.compile(loss="binary_crossentropy", optimizer="adam",metrics=['accuracy'])

输出结果：

2021-04-20 17:03:43.689454: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:941] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-04-20 17:03:43.690115: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:941] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-04-20 17:03:43.690774: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1862] Adding visible gpu devices: 0
2021-04-20 17:03:43.690807: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.10.1
2021-04-20 17:06:41.297456: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1261] Device interconnect StreamExecutor with strength 1 edge matrix:
2021-04-20 17:06:41.297566: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1267]      0 
2021-04-20 17:06:41.297574: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1280] 0:   N 

康达环境：

• 蟒蛇=3.7.10

• cudatoolkit=10.1.243

• cudnn=7.6.5

• 张量流-GPU=2.4.1

硬件：

• 英特尔® 酷睿™ i5-7200U

• GeForce 940MX

标签： pythontensorflowkerasconv-neural-network