Installing GPU Deployment Kit - do I have to revert to an older graphics driver?

I’m on Windows 7, with a GTX 980 Ti. I installed the 353.30 graphics driver immediately after getting the card.

Now, when I try to install CUDA, if I deselect the graphics driver (which is older than my current driver in both the CUDA 7 release and the CUDA 7.5 release), it also deselects the GPU Deployment Kit. If I select the GPU Deployment Kit again, it also selects the graphics driver.

Is it possible to install the GPU Deployment Kit without reverting my driver to an older version, or are the two inextricably linked?

I ran into the same issue this morning. I found some sources (sort of) saying that the 980 Ti works with it’s current driver with CUDA 7.5. I applied to be a registered developer this morning and am downloading 7.5 now… I’ll let you know if 7.5 works soon! If someone knows whether there’s a way to make a 980 Ti work with 7, please let us know! Thank you!

7.5 did not work immediately. The message I am looking at right now reads

This graphics driver could not find compatible graphics 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.

This message is different from the one that comes up with the CUDA 7 Toolkit, and is far more reassuring in it’s content. I’m going to go with what it says and download without the drivers and will come back with my findings later tonight or next week.

It’s not the “compatible graphics hardware” message I’m concerned about - I’d already read elsewhere that this message comes up if you have a newer driver. My concern is that when I untick the older graphics driver in the custom installation, it also unticks the GPU Deployment Kit. If I then select the GPU Deployment Kit, it also re-selects the graphics driver. i.e. when I follow the instructions in the message you’ve posted, I run into a further problem.

I’d be interested in hearing where you got to, anyway!

The GPU Deployment kit is at least loosely tied to a particular driver version. You can get a sense of this by reviewing the GPU Deployment kit homepage.

This connection in the installer is probably done out of an abundance of caution (it cannot be ascertained if the deployment kit bundled with the CUDA toolkit installer is compatible with your existing driver).

Not installing the GPU deployment kit should not have a tremendous impact on your CUDA install. You should still be able to compile and run CUDA programs. I think the main observable difference would be that the nvml SDK would not get installed.

In the future, when a newer CUDA toolkit is released (which would probably then have proper support for your GPU) you should be able to get the GPU Deployment kit at that point.

Thanks, txbob! I gave it a go, and I now have CUDA installed, and I’ve successfully run a few of the samples.

I’m new to this so wasn’t too sure how important the GPU Deployment Kit was; after reading into the NVML SDK I can see it’s not necessarily a must-have, so thanks for pointing me to that.