Failed to get convolution algorithm error ~ tensorflow-gpu on ubuntu 20.04
Failed to get convolution algorithm error ~ tensorflow-gpu on ubuntu 20.04
我有 NVIDIA 2070 RTX GPU，我的 OS 是 Ubuntu20.04。
我已经用 conda 安装了 tensorflow-gpu 包。我 没有 安装了 CUDA-toolkit 我相信它还会从 CUDA-toolkit 安装所需的库以使用 gpu-acceleration，因为 conda install tensorflow-gpu 给出了以下软件包列表将安装：
Collecting package metadata (current_repodata.json): done
Solving environment: done
## Package Plan ##
environment location: /home/psychotechnopath/anaconda3/envs/DeepLearning3.6
added / updated specs:
- tensorflow-gpu
The following packages will be downloaded:
package | build
---------------------------|-----------------
_tflow_select-2.1.0 | gpu 2 KB
absl-py-0.9.0 | py36_0 167 KB
asn1crypto-1.3.0 | py36_0 164 KB
astor-0.8.0 | py36_0 46 KB
blinker-1.4 | py36_0 22 KB
c-ares-1.15.0 | h7b6447c_1001 89 KB
cachetools-3.1.1 | py_0 14 KB
cffi-1.14.0 | py36h2e261b9_0 223 KB
chardet-3.0.4 | py36_1003 180 KB
click-7.1.1 | py_0 71 KB
cryptography-2.8 | py36h1ba5d50_0 552 KB
cudatoolkit-10.1.243 | h6bb024c_0 347.4 MB
cudnn-7.6.5 | cuda10.1_0 179.9 MB
cupti-10.1.168 | 0 1.4 MB
gast-0.2.2 | py36_0 155 KB
google-auth-1.13.1 | py_0 57 KB
google-auth-oauthlib-0.4.1 | py_2 20 KB
google-pasta-0.2.0 | py_0 44 KB
grpcio-1.27.2 | py36hf8bcb03_0 1.3 MB
h5py-2.10.0 | py36h7918eee_0 1.0 MB
idna-2.9 | py_1 49 KB
keras-applications-1.0.8 | py_0 33 KB
keras-preprocessing-1.1.0 | py_1 36 KB
libprotobuf-3.11.4 | hd408876_0 2.9 MB
markdown-3.1.1 | py36_0 116 KB
mkl-service-2.3.0 | py36he904b0f_0 219 KB
mkl_fft-1.0.15 | py36ha843d7b_0 155 KB
mkl_random-1.1.0 | py36hd6b4f25_0 324 KB
numpy-1.18.1 | py36h4f9e942_0 5 KB
numpy-base-1.18.1 | py36hde5b4d6_1 4.2 MB
oauthlib-3.1.0 | py_0 88 KB
opt_einsum-3.1.0 | py_0 54 KB
protobuf-3.11.4 | py36he6710b0_0 635 KB
pyasn1-0.4.8 | py_0 58 KB
pyasn1-modules-0.2.7 | py_0 63 KB
pycparser-2.20 | py_0 92 KB
pyjwt-1.7.1 | py36_0 33 KB
pyopenssl-19.1.0 | py36_0 87 KB
pysocks-1.7.1 | py36_0 30 KB
requests-2.23.0 | py36_0 91 KB
requests-oauthlib-1.3.0 | py_0 22 KB
rsa-4.0 | py_0 29 KB
scipy-1.4.1 | py36h0b6359f_0 14.6 MB
six-1.14.0 | py36_0 27 KB
tensorboard-2.1.0 | py3_0 3.3 MB
tensorflow-2.1.0 |gpu_py36h2e5cdaa_0 4 KB
tensorflow-base-2.1.0 |gpu_py36h6c5654b_0 155.9 MB
tensorflow-estimator-2.1.0 | pyhd54b08b_0 251 KB
tensorflow-gpu-2.1.0 | h0d30ee6_0 3 KB
termcolor-1.1.0 | py36_1 8 KB
urllib3-1.25.8 | py36_0 169 KB
werkzeug-1.0.1 | py_0 240 KB
wrapt-1.12.1 | py36h7b6447c_1 49 KB
------------------------------------------------------------
Total: 716.6 MB
当我检查是否检测到我的 GPU 时，使用：
import tensorflow as tf
print(tf.__version__)
print("Num GPUs Available: ", tf.config.experimental.list_physical_devices('GPU'))
它检测到我的 GPU，但它似乎有一些（我不知道的）NUMA 错误。
2020-05-01 11:39:26.778829: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcuda.so.1
2020-05-01 11:39:26.799789: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:39:26.800132: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties:
pciBusID: 0000:08:00.0 name: GeForce RTX 2070 computeCapability: 7.5
coreClock: 1.62GHz coreCount: 36 deviceMemorySize: 7.79GiB deviceMemoryBandwidth: 417.29GiB/s
2020-05-01 11:39:26.800234: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
2020-05-01 11:39:26.801035: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-05-01 11:39:26.801981: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2020-05-01 11:39:26.802098: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2020-05-01 11:39:26.802926: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2020-05-01 11:39:26.803409: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2020-05-01 11:39:26.805224: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-05-01 11:39:26.805297: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:39:26.805669: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:39:26.805974: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1697] Adding visible gpu devices: 0
这是打印语句：
Num GPUs Available: [PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
随后，当我尝试 运行 一个卷积神经网络时，我得到以下 output/error（我决定包括完整的输出，因为我不知道哪个部分是相关的，哪个部分是相关的不是；对于那里的所有 tensorflow 专家：请随意编辑输出中不相关的部分）
2020-05-01 11:41:53.682279: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcuda.so.1
2020-05-01 11:41:53.703168: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:53.703512: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties:
pciBusID: 0000:08:00.0 name: GeForce RTX 2070 computeCapability: 7.5
coreClock: 1.62GHz coreCount: 36 deviceMemorySize: 7.79GiB deviceMemoryBandwidth: 417.29GiB/s
2020-05-01 11:41:53.703618: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
2020-05-01 11:41:53.704375: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-05-01 11:41:53.705278: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2020-05-01 11:41:53.705394: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2020-05-01 11:41:53.706237: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2020-05-01 11:41:53.706725: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2020-05-01 11:41:53.708557: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-05-01 11:41:53.708630: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:53.708994: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:53.709299: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1697] Adding visible gpu devices: 0
2020-05-01 11:41:53.709511: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
2020-05-01 11:41:53.733654: I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 3792915000 Hz
2020-05-01 11:41:53.734418: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x55ad4b26e7d0 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-05-01 11:41:53.734434: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): Host, Default Version
2020-05-01 11:41:53.734576: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:53.735123: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties:
pciBusID: 0000:08:00.0 name: GeForce RTX 2070 computeCapability: 7.5
coreClock: 1.62GHz coreCount: 36 deviceMemorySize: 7.79GiB deviceMemoryBandwidth: 417.29GiB/s
2020-05-01 11:41:53.735146: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
2020-05-01 11:41:53.735157: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-05-01 11:41:53.735167: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2020-05-01 11:41:53.735176: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2020-05-01 11:41:53.735186: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2020-05-01 11:41:53.735195: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2020-05-01 11:41:53.735204: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-05-01 11:41:53.735259: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:53.735820: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:53.736333: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1697] Adding visible gpu devices: 0
2020-05-01 11:41:53.736360: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
2020-05-01 11:41:54.012838: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1096] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-05-01 11:41:54.012856: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102] 0
2020-05-01 11:41:54.012861: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] 0: N
2020-05-01 11:41:54.012980: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:54.013316: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:54.013643: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:54.013951: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1241] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 7011 MB memory) -> physical GPU (device: 0, name: GeForce RTX 2070, pci bus id: 0000:08:00.0, compute capability: 7.5)
2020-05-01 11:41:54.015048: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x55ad4ef1fe00 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
2020-05-01 11:41:54.015055: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): GeForce RTX 2070, Compute Capability 7.5
2020-05-01 11:41:54.619977: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-05-01 11:41:54.765976: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-05-01 11:41:55.109936: E tensorflow/stream_executor/cuda/cuda_dnn.cc:329] Could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR
2020-05-01 11:41:55.123585: E tensorflow/stream_executor/cuda/cuda_dnn.cc:329] Could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR
2020-05-01 11:41:55.123654: W tensorflow/core/common_runtime/base_collective_executor.cc:217] BaseCollectiveExecutor::StartAbort Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
[[{{node sequential/conv2d/Conv2D}}]]
Traceback (most recent call last):
File "/home/psychotechnopath/MEGA/Machine Learning/11. Deep learning for Python/5. Convolutional neural networks/CH19_Digits.py", line 66, in <module>
model.fit(X_train, y_train, validation_data=(X_test, y_test), batch_size=200, epochs=10, verbose=2)
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training.py", line 819, in fit
use_multiprocessing=use_multiprocessing)
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_v2.py", line 342, in fit
total_epochs=epochs)
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_v2.py", line 128, in run_one_epoch
batch_outs = execution_function(iterator)
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_v2_utils.py", line 98, in execution_function
distributed_function(input_fn))
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/eager/def_function.py", line 568, in __call__
result = self._call(*args, **kwds)
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/eager/def_function.py", line 632, in _call
return self._stateless_fn(*args, **kwds)
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/eager/function.py", line 2363, in __call__
return graph_function._filtered_call(args, kwargs) # pylint: disable=protected-access
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/eager/function.py", line 1611, in _filtered_call
self.captured_inputs)
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/eager/function.py", line 1692, in _call_flat
ctx, args, cancellation_manager=cancellation_manager))
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/eager/function.py", line 545, in call
ctx=ctx)
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/eager/execute.py", line 67, in quick_execute
six.raise_from(core._status_to_exception(e.code, message), None)
File "<string>", line 3, in raise_from
tensorflow.python.framework.errors_impl.UnknownError: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
[[node sequential/conv2d/Conv2D (defined at /MEGA/Machine Learning/11. Deep learning for Python/5. Convolutional neural networks/CH19_Digits.py:66) ]] [Op:__inference_distributed_function_1027]
Function call stack:
distributed_function
这似乎是tensorflow中的一个已知错误，它与tensorflow在20XX卡中进行的内存分配有关。在此处查看详细线程：
https://github.com/tensorflow/tensorflow/issues/24496
解决我问题的方法是在我的脚本顶部添加以下代码：
tf.config.experimental.set_memory_growth(tf.config.list_physical_devices('GPU')[0], True)
我有 NVIDIA 2070 RTX GPU，我的 OS 是 Ubuntu20.04。
我已经用 conda 安装了 tensorflow-gpu 包。我 没有 安装了 CUDA-toolkit 我相信它还会从 CUDA-toolkit 安装所需的库以使用 gpu-acceleration，因为 conda install tensorflow-gpu 给出了以下软件包列表将安装：
Collecting package metadata (current_repodata.json): done
Solving environment: done
## Package Plan ##
environment location: /home/psychotechnopath/anaconda3/envs/DeepLearning3.6
added / updated specs:
- tensorflow-gpu
The following packages will be downloaded:
package | build
---------------------------|-----------------
_tflow_select-2.1.0 | gpu 2 KB
absl-py-0.9.0 | py36_0 167 KB
asn1crypto-1.3.0 | py36_0 164 KB
astor-0.8.0 | py36_0 46 KB
blinker-1.4 | py36_0 22 KB
c-ares-1.15.0 | h7b6447c_1001 89 KB
cachetools-3.1.1 | py_0 14 KB
cffi-1.14.0 | py36h2e261b9_0 223 KB
chardet-3.0.4 | py36_1003 180 KB
click-7.1.1 | py_0 71 KB
cryptography-2.8 | py36h1ba5d50_0 552 KB
cudatoolkit-10.1.243 | h6bb024c_0 347.4 MB
cudnn-7.6.5 | cuda10.1_0 179.9 MB
cupti-10.1.168 | 0 1.4 MB
gast-0.2.2 | py36_0 155 KB
google-auth-1.13.1 | py_0 57 KB
google-auth-oauthlib-0.4.1 | py_2 20 KB
google-pasta-0.2.0 | py_0 44 KB
grpcio-1.27.2 | py36hf8bcb03_0 1.3 MB
h5py-2.10.0 | py36h7918eee_0 1.0 MB
idna-2.9 | py_1 49 KB
keras-applications-1.0.8 | py_0 33 KB
keras-preprocessing-1.1.0 | py_1 36 KB
libprotobuf-3.11.4 | hd408876_0 2.9 MB
markdown-3.1.1 | py36_0 116 KB
mkl-service-2.3.0 | py36he904b0f_0 219 KB
mkl_fft-1.0.15 | py36ha843d7b_0 155 KB
mkl_random-1.1.0 | py36hd6b4f25_0 324 KB
numpy-1.18.1 | py36h4f9e942_0 5 KB
numpy-base-1.18.1 | py36hde5b4d6_1 4.2 MB
oauthlib-3.1.0 | py_0 88 KB
opt_einsum-3.1.0 | py_0 54 KB
protobuf-3.11.4 | py36he6710b0_0 635 KB
pyasn1-0.4.8 | py_0 58 KB
pyasn1-modules-0.2.7 | py_0 63 KB
pycparser-2.20 | py_0 92 KB
pyjwt-1.7.1 | py36_0 33 KB
pyopenssl-19.1.0 | py36_0 87 KB
pysocks-1.7.1 | py36_0 30 KB
requests-2.23.0 | py36_0 91 KB
requests-oauthlib-1.3.0 | py_0 22 KB
rsa-4.0 | py_0 29 KB
scipy-1.4.1 | py36h0b6359f_0 14.6 MB
six-1.14.0 | py36_0 27 KB
tensorboard-2.1.0 | py3_0 3.3 MB
tensorflow-2.1.0 |gpu_py36h2e5cdaa_0 4 KB
tensorflow-base-2.1.0 |gpu_py36h6c5654b_0 155.9 MB
tensorflow-estimator-2.1.0 | pyhd54b08b_0 251 KB
tensorflow-gpu-2.1.0 | h0d30ee6_0 3 KB
termcolor-1.1.0 | py36_1 8 KB
urllib3-1.25.8 | py36_0 169 KB
werkzeug-1.0.1 | py_0 240 KB
wrapt-1.12.1 | py36h7b6447c_1 49 KB
------------------------------------------------------------
Total: 716.6 MB
当我检查是否检测到我的 GPU 时，使用：
import tensorflow as tf
print(tf.__version__)
print("Num GPUs Available: ", tf.config.experimental.list_physical_devices('GPU'))
它检测到我的 GPU，但它似乎有一些（我不知道的）NUMA 错误。
2020-05-01 11:39:26.778829: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcuda.so.1
2020-05-01 11:39:26.799789: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:39:26.800132: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties:
pciBusID: 0000:08:00.0 name: GeForce RTX 2070 computeCapability: 7.5
coreClock: 1.62GHz coreCount: 36 deviceMemorySize: 7.79GiB deviceMemoryBandwidth: 417.29GiB/s
2020-05-01 11:39:26.800234: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
2020-05-01 11:39:26.801035: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-05-01 11:39:26.801981: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2020-05-01 11:39:26.802098: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2020-05-01 11:39:26.802926: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2020-05-01 11:39:26.803409: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2020-05-01 11:39:26.805224: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-05-01 11:39:26.805297: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:39:26.805669: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:39:26.805974: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1697] Adding visible gpu devices: 0
这是打印语句：
Num GPUs Available: [PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
随后，当我尝试 运行 一个卷积神经网络时，我得到以下 output/error（我决定包括完整的输出，因为我不知道哪个部分是相关的，哪个部分是相关的不是；对于那里的所有 tensorflow 专家：请随意编辑输出中不相关的部分）
2020-05-01 11:41:53.682279: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcuda.so.1
2020-05-01 11:41:53.703168: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:53.703512: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties:
pciBusID: 0000:08:00.0 name: GeForce RTX 2070 computeCapability: 7.5
coreClock: 1.62GHz coreCount: 36 deviceMemorySize: 7.79GiB deviceMemoryBandwidth: 417.29GiB/s
2020-05-01 11:41:53.703618: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
2020-05-01 11:41:53.704375: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-05-01 11:41:53.705278: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2020-05-01 11:41:53.705394: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2020-05-01 11:41:53.706237: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2020-05-01 11:41:53.706725: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2020-05-01 11:41:53.708557: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-05-01 11:41:53.708630: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:53.708994: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:53.709299: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1697] Adding visible gpu devices: 0
2020-05-01 11:41:53.709511: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
2020-05-01 11:41:53.733654: I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 3792915000 Hz
2020-05-01 11:41:53.734418: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x55ad4b26e7d0 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-05-01 11:41:53.734434: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): Host, Default Version
2020-05-01 11:41:53.734576: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:53.735123: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties:
pciBusID: 0000:08:00.0 name: GeForce RTX 2070 computeCapability: 7.5
coreClock: 1.62GHz coreCount: 36 deviceMemorySize: 7.79GiB deviceMemoryBandwidth: 417.29GiB/s
2020-05-01 11:41:53.735146: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
2020-05-01 11:41:53.735157: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-05-01 11:41:53.735167: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2020-05-01 11:41:53.735176: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2020-05-01 11:41:53.735186: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2020-05-01 11:41:53.735195: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2020-05-01 11:41:53.735204: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-05-01 11:41:53.735259: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:53.735820: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:53.736333: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1697] Adding visible gpu devices: 0
2020-05-01 11:41:53.736360: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
2020-05-01 11:41:54.012838: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1096] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-05-01 11:41:54.012856: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102] 0
2020-05-01 11:41:54.012861: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] 0: N
2020-05-01 11:41:54.012980: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:54.013316: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:54.013643: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-05-01 11:41:54.013951: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1241] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 7011 MB memory) -> physical GPU (device: 0, name: GeForce RTX 2070, pci bus id: 0000:08:00.0, compute capability: 7.5)
2020-05-01 11:41:54.015048: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x55ad4ef1fe00 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
2020-05-01 11:41:54.015055: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): GeForce RTX 2070, Compute Capability 7.5
2020-05-01 11:41:54.619977: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-05-01 11:41:54.765976: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-05-01 11:41:55.109936: E tensorflow/stream_executor/cuda/cuda_dnn.cc:329] Could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR
2020-05-01 11:41:55.123585: E tensorflow/stream_executor/cuda/cuda_dnn.cc:329] Could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR
2020-05-01 11:41:55.123654: W tensorflow/core/common_runtime/base_collective_executor.cc:217] BaseCollectiveExecutor::StartAbort Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
[[{{node sequential/conv2d/Conv2D}}]]
Traceback (most recent call last):
File "/home/psychotechnopath/MEGA/Machine Learning/11. Deep learning for Python/5. Convolutional neural networks/CH19_Digits.py", line 66, in <module>
model.fit(X_train, y_train, validation_data=(X_test, y_test), batch_size=200, epochs=10, verbose=2)
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training.py", line 819, in fit
use_multiprocessing=use_multiprocessing)
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_v2.py", line 342, in fit
total_epochs=epochs)
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_v2.py", line 128, in run_one_epoch
batch_outs = execution_function(iterator)
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_v2_utils.py", line 98, in execution_function
distributed_function(input_fn))
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/eager/def_function.py", line 568, in __call__
result = self._call(*args, **kwds)
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/eager/def_function.py", line 632, in _call
return self._stateless_fn(*args, **kwds)
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/eager/function.py", line 2363, in __call__
return graph_function._filtered_call(args, kwargs) # pylint: disable=protected-access
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/eager/function.py", line 1611, in _filtered_call
self.captured_inputs)
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/eager/function.py", line 1692, in _call_flat
ctx, args, cancellation_manager=cancellation_manager))
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/eager/function.py", line 545, in call
ctx=ctx)
File "/home/psychotechnopath/anaconda3/envs/DeepLearning3.6/lib/python3.6/site-packages/tensorflow_core/python/eager/execute.py", line 67, in quick_execute
six.raise_from(core._status_to_exception(e.code, message), None)
File "<string>", line 3, in raise_from
tensorflow.python.framework.errors_impl.UnknownError: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
[[node sequential/conv2d/Conv2D (defined at /MEGA/Machine Learning/11. Deep learning for Python/5. Convolutional neural networks/CH19_Digits.py:66) ]] [Op:__inference_distributed_function_1027]
Function call stack:
distributed_function
这似乎是tensorflow中的一个已知错误，它与tensorflow在20XX卡中进行的内存分配有关。在此处查看详细线程：
https://github.com/tensorflow/tensorflow/issues/24496
解决我问题的方法是在我的脚本顶部添加以下代码：
tf.config.experimental.set_memory_growth(tf.config.list_physical_devices('GPU')[0], True)