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 | architecture | cuda |
Tesla K80 | SM37 or SM_37 | CUDA 5 and later |
Tesla V100 | SM70 or SM_70 | CUDA 9 and later |
cat /proc/driver/nvidia/version
nvcc --version
cat /usr/local/cuda/version.txt
cuDNN | cuda |
7.6.1 | 9.0,9.2,10.0,10.1 |
5.1 | 8.0,7.5 |
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
![](https://i-blog.csdnimg.cn/blog_migrate/ac405cb20e64f749f83bbfbec2af3170.png)
参考文献:
- check-cuda-installed
-
Supported SM and Gencode variations
-
cuDNN Archive
-
tf源码编译
-
Which TensorFlow and CUDA version combinations are compatible?
-
https://pytorch.org/get-started/previous-versions/