2017 Titan Xp NOT SEEN BY CUDA version 8 ????

I have wasted almost a solid 24 hours of my life trying to get tensorflow gpu on windows 10

I even started with a new install of windows 10 to ensure no issues.

Right now I am stuck with CUDA 8 installation:

  1. at system check I get the DO YOU WANT TO CONTINUE???
    says graphics driver could not find compatible graphics hardware yet I use the 2017 Titan Xp all day so its there and so is its driver???

even if i continue then I get a warning that visual studio is not found…
I downloaded both files on the download page it sends me too:
https://www.visualstudio.com/ the IDE and the CODE

did a reboot since I know its amending PATH

Ran everything as administrator …

this is insane…please help…Im looking to hire someone to get tensorflow gpu up and running

I have similar issue. I’m using GTX 1080 Ti on windows 10 and can’t install CUDA Toolkit 8.0

Please follow the Windows installation instructions here:

If a particular component is failing to install, you may unselect it either by customizing the install GUI or by doing a silent install as specified in the guide above.

I have same problem with gparkis. CUDA8 could not find compatible graphics hardware. I think it’s some driver problem, the driver’s version doesn’t fit with CUDA8, Can nVidia engineer fix this as soon?

It’s note as: This graphics driver could not find compatible hardware. You may continue installation, but you may not be able to run CUDA applications with this driver. This may occur with graphics hardware that is newer than this toolkit. In that case, it is suggested that you keep your existing driver and install the remaining portions of the CUDA Toolkit.

I’m tying to customizing install but it does not work. No GUI option.

How can I do?

I was successful in installing it.

I had to manually move the cudnn files into the cuda folders.

And make sure the path was set.

When I did this, I still got the error that no hardware was seen yet when I would do tensorflow and gpu check it would work. And when I was running models it was using the GPU, so it can still work.

When I get home I will create some screen shots showing the folders and the path.


I had a similar issue with my GTX1080Ti as well. As far as I could figure out the driver contained in the latest CUDA-Release is not compatible to GTX1080Ti and later, at least for Linux.

So probably you you have to install the latest driver manually and install CUDA without driver afterwards. But for me that led to other problems https://devtalk.nvidia.com/default/topic/1010664/ubuntu-17-04-cuda-8-0-linker-problems-with-cuda-examples-dynamic-parallelism-/?offset=3.

However, tensorflow-gpu is up and running (both, binaries and built from source).

As for linux I can tell you it was a dam nightmare. I never was able to get gpu tensorflow to work.

Windows 10 was fine.

Which folders I should move the cudnn file into?

I can’t find the path.