はじめに
Tensorflow2.1.0で遊ぼうと思ったらCUDNN_STATUS_NOT_INITIALIZEDのエラーが出たので、解決方法を書いていきます。
※Tensorflow2.0.0でも同様のエラーが出るかもしれません。
環境
- OS :: Windows10 home
- GPU :: NVIDIA RTX2070
- cuDNN :: 7.6
- CUDA :: 10.1
- Python 3.7.6 :: Anaconda custom (64-bit)
- tensorflow :: tensorflow-gpu==2.1.0
環境は以下を参考にして入れました。
https://www.tensorflow.org/install/source#tested_build_configurations
エラー内容
2020-03-21 00:29:53.137414: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2020-03-21 00:29:55.399548: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library nvcuda.dll
2020-03-21 00:29:55.423219: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce RTX 2070 computeCapability: 7.5
coreClock: 1.71GHz coreCount: 36 deviceMemorySize: 8.00GiB deviceMemoryBandwidth: 417.29GiB/s
2020-03-21 00:29:55.430144: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2020-03-21 00:29:55.438727: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2020-03-21 00:29:55.445024: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cufft64_10.dll
2020-03-21 00:29:55.449169: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library curand64_10.dll
2020-03-21 00:29:55.456919: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusolver64_10.dll
2020-03-21 00:29:55.462110: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusparse64_10.dll
2020-03-21 00:29:55.473068: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2020-03-21 00:29:55.476609: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1697] Adding visible gpu devices: 0
2020-03-21 00:29:55.479345: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2020-03-21 00:29:55.484766: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce RTX 2070 computeCapability: 7.5
coreClock: 1.71GHz coreCount: 36 deviceMemorySize: 8.00GiB deviceMemoryBandwidth: 417.29GiB/s
2020-03-21 00:29:55.493764: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2020-03-21 00:29:55.497168: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2020-03-21 00:29:55.502105: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cufft64_10.dll
2020-03-21 00:29:55.505886: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library curand64_10.dll
2020-03-21 00:29:55.510337: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusolver64_10.dll
2020-03-21 00:29:55.513831: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusparse64_10.dll
2020-03-21 00:29:55.518574: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2020-03-21 00:29:55.522030: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1697] Adding visible gpu devices: 0
2020-03-21 00:29:56.067998: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1096] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-03-21 00:29:56.071564: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102] 0
2020-03-21 00:29:56.075105: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] 0: N
2020-03-21 00:29:56.078938: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1241] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 6304 MB memory) -> physical GPU (device: 0, name: GeForce RTX 2070, pci bus id: 0000:01:00.0, compute capability: 7.5)
2020-03-21 00:29:56.930030: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2020-03-21 00:29:57.220935: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2020-03-21 00:29:58.451976: E tensorflow/stream_executor/cuda/cuda_dnn.cc:329] Could not create cudnn handle: CUDNN_STATUS_NOT_INITIALIZED
2020-03-21 00:29:58.455735: E tensorflow/stream_executor/cuda/cuda_dnn.cc:333] Error retrieving driver version: Unimplemented: kernel reported driver version not implemented on Windows
2020-03-21 00:29:58.461795: E tensorflow/stream_executor/cuda/cuda_dnn.cc:329] Could not create cudnn handle: CUDNN_STATUS_NOT_INITIALIZED
2020-03-21 00:29:58.466492: E tensorflow/stream_executor/cuda/cuda_dnn.cc:333] Error retrieving driver version: Unimplemented: kernel reported driver version not implemented on Windows
2020-03-21 00:29:58.472369: F tensorflow/core/kernels/conv_grad_input_ops.cc:1163] Check failed: stream->parent()->GetConvolveBackwardDataAlgorithms( conv_parameters.ShouldIncludeWinogradNonfusedAlgo<T>(stream->parent()), &algorithms)
解決方法
ソースコード
以下のどちらかをソースコードに入れる。
from tensorflow.compat.v1 import ConfigProto
from tensorflow.compat.v1 import InteractiveSession
config = ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.2
config.gpu_options.allow_growth = True
session = InteractiveSession(config=config)
import tensorflow as tf
gpu_devices = tf.config.experimental.list_physical_devices('GPU')
for device in gpu_devices: tf.config.experimental.set_memory_growth(device, True)
参考
一番下に記載してありました。
https://github.com/tensorflow/tensorflow/issues/24496
おわりに
githubのissueを見る限り、RTX2070もしくはRTX2080だと発生するエラーのようです。