21:10:57.123186: W tensorflow/stream_executor/platform/default/dso_:55] Could not load dynamic library 'libcudart.so.10.0' dlerror: libcudart.so.10.0: cannot open shared object file: No such file or directory LD_LIBRARY_PATH: /usr/local/cuda/lib64 Name: NVIDIA GeForce RTX 3090 major: 8 minor: 6 memor圜lockRate(GHz): 1.695 Installing CUDA 10.2 does not work with the following errors: 21:10:57.123081: I tensorflow/core/common_runtime/gpu/gpu_:1618] Found device 0 with properties:
#Nvidia cuda toolkit driver install#
For example, if you want to run TensorFlow 1.15, you must install CUDA 10.0. Note that the minor CUDA version is required. Official page with the compatibility chart Linux CPU (Linux) Version Let's solve these compatibility requirements one by one. tensorflow-datasets=3.2.1 (my project's need).tensorflow-probability=0.7 (my project's need).NVIDIA driver 410.48 (comes with CUDA 10.0 Toolkit) or NVIDIA 470 (I installed via apt and confirmed that it works).Note: install a different minor version CUDA 10.2 does not work. Python 3.7 (the latest version supported by TF 1.15).You can use TensorFlow version 1, by installing exactly the following versions of the required components: Quick resolution for Tensorflow version 1 user TensorFlow → Cudnn/Cuda → NVIDIA driver/GCC.The general flow of the compatibility resolving process is This post will show the compatibility table with references to official pages. Running (training) legacy machine learning models, especially models written for TensorFlow v1, is not a trivial task mostly due to the version incompatibility issue.
0 kommentar(er)
