Tensorflow 找不到 GPU

Question

pip install tensorflow==1.14.0
pip install tensorflow-gpu==1.14

import tensorflow as tf

tf.test.is_gpu_available(
cuda_only=False,
min_cuda_compute_capability=None
)

假

GPUs = GPU.getGPUs()
gpu = GPUs[0]
def printm():
process = psutil.Process(os.getpid())
print("Gen RAM Free: " + humanize.naturalsize( psutil.virtual_memory().available ), " | Proc size: " + humanize.naturalsize( process.memory_info().rss))
print("# of CPU: {0}".format(psutil.cpu_count()))
print("CPU type: {0}".format(platform.uname()))
print("GPU Type: {0}".format(gpu.name))
print("GPU RAM Free: {0:.0f}MB | Used: {1:.0f}MB | Util {2:3.0f}% | Total {3:.0f}MB".format(gpu.memoryFree, gpu.memoryUsed, gpu.memoryUtil*100, gpu.memoryTotal))
printm()
Gen RAM Free: 65.8 GB  | Proc size: 294.4 MB
# of CPU: 8
CPU type: uname_result(system='Linux', node='lian-2', release='5.3.0-53-generic', version='#47~18.04.1-Ubuntu SMP Thu May 7 13:10:50 UTC 2020', machine='x86_64', processor='x86_64')
GPU Type: GeForce RTX 2080 Ti
GPU RAM Free: 10992MB | Used: 26MB | Util   0% | Total 11018MB

我重新安装了 cuda 10

Mon Jun  8 15:48:44 2020       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.82       Driver Version: 440.82       CUDA Version: 10.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce RTX 208...  Off  | 00000000:01:00.0 Off |                  N/A |
| 24%   28C    P8     2W / 260W |     26MiB / 11018MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0      1805      G   /usr/lib/xorg/Xorg                             9MiB |
|    0      2192      G   /usr/bin/gnome-shell                          14MiB |
+-----------------------------------------------------------------------------+
define CUDNN_MAJOR 7
#define CUDNN_MINOR 4
#define CUDNN_PATCHLEVEL 2
--
#define CUDNN_VERSION (CUDNN_MAJOR * 1000 + CUDNN_MINOR * 100 + CUDNN_PATCHLEVEL)

我正在尝试从 jupyter 笔记本 (ubuntu 18 ssh( 训练它：

model = Sequential()
model.add(layers.Conv1D(16, 9, activation='relu', input_shape=(800, X_train.shape[-1])))
model.add(layers.MaxPooling1D(2))
model.add(layers.Bidirectional(layers.CuDNNGRU(8, return_sequences=True)))
model.add(layers.Bidirectional(layers.CuDNNGRU(8)))
model.add(layers.Dense(8, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))
model.compile(optimizer= 'Adam', loss='binary_crossentropy', metrics=['acc'])
model.summary()
train_to_epoch = 150
start_epoch = 3
t1 = datetime.datetime.now()
print('Training start time = %s' %  t1)
history = model.fit(X_train, y_train,
batch_size=128,  epochs=train_to_epoch, verbose=0,
validation_data=(X_val, y_val))
print('nTraining Duration = %s' % (datetime.datetime.now()-t1))

InvalidArgumentError： 没有 OpKernel 被注册以支持 {{node bidirectional_1/CudnnRNN}} 使用的 Op 'CudnnRNN'，具有以下属性： [input_mode="linear_input"， T=DT_FLOAT， direction="unidirectional"， rnn_mode="gru"， seed2=0， is_training=true， seed=87654321， dropout=0] 已注册设备：[CPU、XLA_CPU] 注册内核：

[[bidirectional_1/CudnnRNN]]

也许需要另一个版本的 CUDA？

Answer 1

是的，它是 CUDA 10.1 版本。