1

I am trying to set up an NVidia RTX 2070 GPU on my laptop for tensorflow, I am a total newbie when it comes to GPUs. My problem is CUDA installation. More specifically, I get three pieces of information that seem incompatible.

When I type nvidia-smi on my Ubuntu 18.04 terminal I get:

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.64.00    Driver Version: 440.64.00    CUDA Version: 10.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce RTX 2070    On   | 00000000:01:00.0 Off |                  N/A |
| N/A   45C    P8     8W /  N/A |    646MiB /  7982MiB |      5%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0      1218      G   /usr/lib/xorg/Xorg                            72MiB |
|    0      1271      G   /usr/bin/gnome-shell                          52MiB |
|    0      1600      G   /usr/lib/xorg/Xorg                           330MiB |
|    0      1728      G   /usr/bin/gnome-shell                         188MiB |
+-----------------------------------------------------------------------------+

But when I type nvcc --version I get:

Command 'nvcc' not found, but can be installed with:

sudo apt install nvidia-cuda-toolkit

And finally, when I type /usr/local/cuda/bin/nvcc --version I get:

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Sun_Jul_28_19:07:16_PDT_2019
Cuda compilation tools, release 10.1, V10.1.243

In other words, on one side I have CUDA 10.2, on another side I have none, on another I have 10.1.

Sorry for the rookie question but I can-t figure out what's happening here.

Leevo
  • 1,683
  • 2
  • 17
  • 34
  • 4
    nvidia-smi only tells you the maximum CUDA support level of your driver. It says (and cannot say) anything about what version of the CUDA toolkit you may or may not have installed. – talonmies Apr 24 '20 at 09:52
  • 1
    Does this answer your question? [Different CUDA versions shown by nvcc and NVIDIA-smi](https://stackoverflow.com/questions/53422407/different-cuda-versions-shown-by-nvcc-and-nvidia-smi) – Wippo Apr 24 '20 at 09:53
  • I see thanks. I still don't understand one thing: what version do I have installed? I still don't understand why `nvcc --version` returns nothing, but `/usr/local/cuda/bin/nvcc --version` returns 10.1. What do I have here? – Leevo Apr 24 '20 at 09:55
  • 1
    nvcc not being found can be as simple as it not being in your shell search path – talonmies Apr 24 '20 at 09:56
  • 2
    Perhaps you forgot to add the cuda path to your `PATH` and `LD_LIBRARY_PATH` enviroment variables? – Teo Cherici Apr 24 '20 at 10:03
  • https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#post-installation-actions – Robert Crovella Apr 24 '20 at 21:07

0 Answers0