what’s the difference between Cuda and Cudatoolkit it should be the same version to be compatible with Deep learning APIs like tensorflow and pytorch ? i have installed cuda(driver) 11.1 for my RTX3060 on windows but when i am installing the tensorflow-gpu (with anaconda) even the older version of tensorflow its forcing to install the most advaced cudatoolkit 11.3 ? is it ok and compatible ? and if not how to intsall a compatible version of cuda driver ,cudatoolkit and Deep learning APIs, when i am trying to install tensorflow with a specific cudatoolkit version its throw this error
“The following packages are not available from current channels”
CUDA and the cudatoolkit refer to the same thing. CUDA is a library used by many programs like Tensorflow and OpenCV. cudatoolkit is a set software on top of CUDA to make GPU programming easy with CUDA.
You may have installed CUDA in a different path, not at the same folder where you have installed the conda. If you are installing Tensorflow-gpu and it’s installing cudatoolkit 11.3, then of-course it is compatible with that TensorFlow version.
Please check this Tested Build Configurations list to know about the Tensorflow compatible version of CUDA and cuDNN. You can install TF compatible specific cudatoolkit by using below code:
conda install -c conda-forge cudatoolkit=11.2 cudnn=8.1.0
2 Likes
thank you for explanation
1 Like