POSTED BY
2023-01-11

Google TensorFlowバージョン2 の公式チュートリアルをすすめるメモ（１）
Google TensorFlowバージョン2 の公式チュートリアルをすすめるメモ（２）
【Windows10】Python統合開発環境Anaconda3メモ
【Anaconda3】指定した仮想環境でJupyter Notebookを動かす
【Windows10】NVIDIAグラフィックカードのGPUをTensorFlowに認識させて動かす

続きです。TensorFlowでGPUを認識した集大成として、TensorFlow公式チュートリアルサンプル

●テンソルと演算
https://www.tensorflow.org/tutorials/customization/basics?hl=ja

をすすめてみます。Anaconda3 64-bit Jupyter Notebook(64bit)を起動し、前回のGPU仮想環境カーネルＮew→Environment「conda_tf-gpu」にて新規作成します。

★出現したエラー：なし

無事エラー無く実行できました。Jupyter Notebook(Anaconda3)コンソールログではGPU利用の痕跡が見えます。本記事の最後に貼付します。

上記チュートリアルの「GPUによる高速化」「明示的デバイス配置」にて、GPUのチェックと同一の処理をCPUとGPUで明示的に指定して行ってテストできています。

GPUのチェック

x = tf.random.uniform([3, 3])

print("利用できるGPUはあるか: "),
print(tf.config.experimental.list_physical_devices("GPU"))

print("テンソルはGPU #0にあるか:  "),
print(x.device.endswith('GPU:0'))

結果

利用できるGPUはあるか: 
[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
テンソルはGPU #0にあるか:  
True

CPUとGPUの比較

import time

def time_matmul(x):
  start = time.time()
  for loop in range(10):
    tf.matmul(x, x)

  result = time.time()-start

  print("10 loops: {:0.2f}ms".format(1000*result))

# CPUでの実行を強制
print("On CPU:")
with tf.device("CPU:0"):
  x = tf.random.uniform([1000, 1000])
  assert x.device.endswith("CPU:0")
  time_matmul(x)

# GPU #0があればその上での実行を強制
if tf.config.experimental.list_physical_devices("GPU"):
  print("On GPU:")
  with tf.device("GPU:0"): # 2番めのGPUなら GPU:1, 3番目なら GPU:2 など
    x = tf.random.uniform([1000, 1000])
    assert x.device.endswith("GPU:0")
    time_matmul(x)

結果

On CPU:
10 loops: 49.87ms
On GPU:
10 loops: 1.00ms

初回の実行こそライブラリのロードなどでGPUは時間がかかりCPUより遅いようでしたが、2度目以降はCPUよりはるかに速いという結果が得られました。

以下、Jupyter Notebookコンソールのログです。

[I 13:58:00.211 NotebookApp] Creating new notebook in
[I 13:58:03.410 NotebookApp] Kernel started: e0dfc74e-0ef6-4c37-b3a7-ffbf06a3d37d, name: conda_tf-gpu
[I 14:00:02.993 NotebookApp] Saving file at /Untitled4.ipynb
2021-01-12 14:54:44.132264: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2021-01-12 14:54:58.697013: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library nvcuda.dll
2021-01-12 14:54:58.725958: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce GTX 1050 Ti computeCapability: 6.1
coreClock: 1.455GHz coreCount: 6 deviceMemorySize: 4.00GiB deviceMemoryBandwidth: 104.43GiB/s
2021-01-12 14:54:58.726091: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2021-01-12 14:54:59.476740: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2021-01-12 14:54:59.956291: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cufft64_10.dll
2021-01-12 14:55:00.030618: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library curand64_10.dll
2021-01-12 14:55:00.345922: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusolver64_10.dll
2021-01-12 14:55:00.654046: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusparse64_10.dll
2021-01-12 14:55:01.215301: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2021-01-12 14:55:01.215529: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1703] Adding visible gpu devices: 0
2021-01-12 14:55:01.217412: I tensorflow/core/platform/cpu_feature_guard.cc:143] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2021-01-12 14:55:01.224931: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x1dd92508e20 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2021-01-12 14:55:01.225021: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version
2021-01-12 14:55:01.225609: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce GTX 1050 Ti computeCapability: 6.1
coreClock: 1.455GHz coreCount: 6 deviceMemorySize: 4.00GiB deviceMemoryBandwidth: 104.43GiB/s
2021-01-12 14:55:01.226249: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2021-01-12 14:55:01.226679: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2021-01-12 14:55:01.227056: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cufft64_10.dll
2021-01-12 14:55:01.227449: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library curand64_10.dll
2021-01-12 14:55:01.227813: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusolver64_10.dll
2021-01-12 14:55:01.228172: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusparse64_10.dll
2021-01-12 14:55:01.228506: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2021-01-12 14:55:01.228872: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1703] Adding visible gpu devices: 0
2021-01-12 14:55:01.675882: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102] Device interconnect StreamExecutor with strength 1 edge matrix:
2021-01-12 14:55:01.675991: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1108]      0
2021-01-12 14:55:01.676978: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1121] 0:   N
2021-01-12 14:55:01.677505: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1247] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 2988 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1050 Ti, pci bus id: 0000:01:00.0, compute capability: 6.1)
2021-01-12 14:55:01.680267: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x1dda93e8fe0 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
2021-01-12 14:55:01.680383: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): GeForce GTX 1050 Ti, Compute Capability 6.1
[I 14:56:02.981 NotebookApp] Saving file at /Untitled4.ipynb
2021-01-12 14:56:56.632067: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
[I 14:58:02.977 NotebookApp] Saving file at /Untitled4.ipynb
[I 15:00:03.824 NotebookApp] Saving file at /Untitled4.ipynb