Attempting to register factory for plugin cuRAND when one has already been registered

As I import tensorflow I get:

import tensorflow

2019-06-01 17:36:53.943387: E tensorflow/stream_executor/cuda/] Unable to register cuRAND factory: Attempting to register factory for plugin cuRAND when one has already been registered

Is this a warning that I can ignore, or do I need to solve something? As far as I can tell, CUDA and the toolkit are correctly installed and TF can use the GPU.


A few questions:

  • What version of TensorFlow is this?
  • Is it a pre-built version or did you build it yourself?
  • Is there anything else previously imported before import tensorflow? A complete reproducer would be helpful.



  • I’m running Tensorflow (gpu) 1.13.1 (as per “pip show tensorflow”)
  • It is a pre-built version
  • No other imports necessary for the message to appear. For example, if the very first thing I type after loading python is “import tensorflow” then the message appears.

Some more data:

I’m running Ubuntu 18.04.

nvcc -V
nvcc: NVIDIA ® Cuda compiler driver
Copyright © 2005-2019 NVIDIA Corporation
Built on Wed_Apr_24_19:10:27_PDT_2019
Cuda compilation tools, release 10.1, V10.1.168

lspci | grep -i nvidia
01:00.0 3D controller: NVIDIA Corporation GP107GLM [Quadro P600 Mobile] (rev a1)

Sat Jun 1 19:23:23 2019
| NVIDIA-SMI 418.67 Driver Version: 418.67 CUDA Version: 10.1 |
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| 0 Quadro P600 Off | 00000000:01:00.0 Off | N/A |
| N/A 41C P0 N/A / N/A | 292MiB / 4040MiB | 0% Default |

| Processes: GPU Memory |
| GPU PID Type Process name Usage |
| 0 2131 G /usr/lib/xorg/Xorg 175MiB |
| 0 2305 G /usr/bin/gnome-shell 67MiB |
| 0 2664 G …quest-channel-token=1897608136522406515 46MiB |

Many thanks!

TF 1.13.1 from Google is built with and requires CUDA 10.0 (see ), not 10.1 – do you maybe have both Toolkits installed? If not, please also install CUDA 10.0. (You can keep the newer driver, though.)

Or you can use our NGC TensorFlow container from, which takes care of all of this for you.

Hope this helps!

Got it.