-4

The version I originally installed is 11.6. To match the tensorflow2.11.0, I had to install the v11.2 of CUDA, during which I first uinstall the newer version of CUDA(every thing about it) and then install the earlier version that is 11.2. However, after the atuomatic installation and correctly (I think so) configured system environment variables, the nvcc -V command still dispaly that the version of my CUDA is 11.6. I thought it may due to the downgrade, so I install the v11.7 version of CUDA, thinking it maybe right this time, well, it's not. It still reminded that the current version is 11.6. Hope someone could help me out!

enter image description here And the following img shows how I configure the system environment. enter image description here enter image description here The default configuration didn't work correctly either. And my goal is to keep both version of CUDA so that I can switch between them when I need to do so. Much appreciation!

Steven Cao
  • 1
  • 2
  • nvcc does not show the installed CUDA version, so your whole premise is incorrect. Also note that TensorFlow 2.11 does not have GPU support on Windows (only through WSL). – Dr. Snoopy Mar 05 '23 at 16:38
  • So what is the latest version of tensorflow that support GPU I was wondering? And why it still tolds me about v11.6, it bothers me a lot. In fact the GPU driver supports up to 11.8 in my computer. Every thing just doesn't make sense... – Steven Cao Mar 05 '23 at 16:44
  • On Windows native? That would be 2.10, check the tensorflow website: https://www.tensorflow.org/install/pip#windows-native – Dr. Snoopy Mar 05 '23 at 16:45
  • Thakns, I will check it now, but I'm still not sure that what version of CUDA exactly is running in my laptop. What command can I use or how can I check it? – Steven Cao Mar 05 '23 at 16:51
  • There is a `version.json` in the CUDA directory. Allthough the major and minor version numbers of the toolkit and the compiler normally do match. – paleonix Mar 05 '23 at 18:10

1 Answers1

0

To answer my own question, things turned out that you have to add C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.x(depend on your own version) to the
path. Yes, that's it, case closed. Don't learn from me leaving the CUDA path outside...And nvcc -V does correctly show the CUDA version that you are currently using.

Steven Cao
  • 1
  • 2