RTX2080ti + tensorflow

Hello,

I have a RTX2080ti, and now I want to use for some experimentations with tensorflow.

I see in the Tensorflow Installation Guide that I need:

  • Ubuntu 16.04 o greater
  • NVIDIA® GPU drivers —CUDA 9.0 requires 384.x or higher.
  • CUDA® Toolkit —TensorFlow supports CUDA 9.0.
  • CUPTI ships with the CUDA Toolkit.
  • cuDNN SDK (>= 7.2)

And I probe, with Ubuntu 16.04 + Driver 410.57 + Cuda 9.0 + cuDNN SDK 7.2, and not work.
The error is that “libcublas.9.0 libcublas.so.9.0 cannot open shared object file no such file or directory”

Also I probe, with Ubuntu 18.04 + Driver 410.57 + Cuda 10.00 + cuDNN SDK 7.5, and not work.

And I doubt, if is it possible install a tensorflow and use the rtx2080?

For the RTX, cuda 10 is recommended, 9.0 doesn’t really work on those. The error you were getting was pointing to not having set the correct library paths, though.
To install the driver and cuda 10 on Ubuntu, please follow this:

  • Don’t use the .run installers, use --uninstall to uninstall them
  • purge anthing nvidia/cuda
  • add the ubuntu graphics ppa https://launchpad.net/~graphics-drivers/+archive/ubuntu/ppa
  • install the driver from that (sudo apt install nvidia-driver-415)
  • download the cuda .deb
  • add the repo to your system (first three steps from install instructions on download page)
  • don’t install cuda
  • instead, run sudo apt install cuda-toolkit-10-0

Ok then I understand that tensorflow isn’t compatible with rtx2080.

Because rtx2080 need cuda 10 and tensorflow only cuda 9.

https://medium.com/@saitejadommeti/building-tensorflow-gpu-from-source-for-rtx-2080-96fed102fcca

I follow this post, and don`t work. When I put “vncc -V” said that the program wasn’t installed.