tensorflow cuda cudnn

绝地灬酷狼 2021-11-27 05:06 402阅读 0赞

*  tf版本问题

问题： tensorflow==1.12.0 提供非阻塞服务时，报错tensorflow/core/common_runtime/bfc_allocator.cc:380] Check failed: h != kInvalidChunkHandle；另一方面，阻塞服务时运行正常 
    解决方法： 替换为1.13rc0后运行正常

*  [gpu model及其编译结构][gpu model]

<table style="width:500px;"> 
 <tbody> 
  <tr> 
   <td>gpu model</td> 
   <td>architecture</td> 
   <td>cuda</td> 
  </tr> 
  <tr> 
   <td><strong>Tesla K80</strong></td> 
   <td>SM37 or SM_37</td> 
   <td><strong>CUDA 5 and later</strong></td> 
  </tr> 
  <tr> 
   <td><strong>Tesla V100</strong></td> 
   <td>SM70 or SM_70</td> 
   <td><strong>CUDA 9 and later</strong></td> 
  </tr> 
 </tbody> 
</table>

*  [cuda 驱动版本查看][cuda]

cat /proc/driver/nvidia/version

*  [cuda编译器版本查看][cuda]

nvcc --version
    
    cat /usr/local/cuda/version.txt

*  [cuDNN与cuda版本对应关系][cuDNN_cuda]

<table style="width:500px;"> 
 <tbody> 
  <tr> 
   <td>cuDNN</td> 
   <td>cuda</td> 
  </tr> 
  <tr> 
   <td>7.6.1</td> 
   <td>9.0，9.2，10.0，10.1</td> 
  </tr> 
  <tr> 
   <td>5.1</td> 
   <td>8.0，7.5</td> 
  </tr> 
 </tbody> 
</table>

*  cuDNN版本查看

cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2
    cat /usr/include/cudnn.h | grep CUDNN_MAJOR -A 2  //或者试试这个

*  tensorflow[测试的构建配置][Link 1]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2h6aGoyMDA3_size_16_color_FFFFFF_t_70][]

*  [pytorch与cuda版本的对应][pytorch_cuda]

参考文献：

1.  [check-cuda-installed][cuda]
2.  [Supported SM and Gencode variations][gpu model]
3.  [cuDNN Archive][cuDNN_cuda]
4.  [tf源码编译][tf]
5.  [Which TensorFlow and CUDA version combinations are compatible?][Link 1]
6.  [https://pytorch.org/get-started/previous-versions/][pytorch_cuda]

[gpu model]: https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
[cuda]: http://arnon.dk/check-cuda-installed/
[cuDNN_cuda]: https://developer.nvidia.com/rdp/cudnn-archive
[Link 1]: https://stackoverflow.com/questions/50622525/which-tensorflow-and-cuda-version-combinations-are-compatible
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2h6aGoyMDA3_size_16_color_FFFFFF_t_70]: /images/20211126/8884a72be52f4f95ab244d9d5bfe2340.png
[pytorch_cuda]: https://pytorch.org/get-started/previous-versions/
[tf]: https://www.tensorflow.org/install/source?hl=zh-cn#linux