Why did CUDA 8 install remove the nvidia-375 driver and replaced it with nvidia-367?

I successfully installed the GEFORCE nvidia-375.26 driver on my laptop with GTX1060 GPU and Ubuntu 16.04 OS. When I subsequently installed CUDA 8 (cuda-repo-ubuntu1604-8-0-local_8.0.44-1_amd64.deb), it removed the nvidia-375 driver and replaced it with nvidia-367, as per below:

The following packages will be REMOVED:
libcuda1-375 nvidia-375 nvidia-opencl-icd-375
The following NEW packages will be installed:
ca-certificates-java cuda cuda-8-0 cuda-command-line-tools-8-0 cuda-core-8-0
cuda-cublas-8-0 cuda-cublas-dev-8-0 cuda-cudart-8-0 cuda-cudart-dev-8-0
cuda-cufft-8-0 cuda-cufft-dev-8-0 cuda-curand-8-0 cuda-curand-dev-8-0
cuda-cusolver-8-0 cuda-cusolver-dev-8-0 cuda-cusparse-8-0
cuda-cusparse-dev-8-0 cuda-demo-suite-8-0 cuda-documentation-8-0
cuda-driver-dev-8-0 cuda-drivers cuda-license-8-0 cuda-misc-headers-8-0
cuda-npp-8-0 cuda-npp-dev-8-0 cuda-nvgraph-8-0 cuda-nvgraph-dev-8-0
cuda-nvml-dev-8-0 cuda-nvrtc-8-0 cuda-nvrtc-dev-8-0 cuda-runtime-8-0
cuda-samples-8-0 cuda-toolkit-8-0 cuda-visual-tools-8-0 default-jre
default-jre-headless fonts-dejavu-extra java-common libbonobo2-0
libbonobo2-common libcuda1-367 libgnome-2-0 libgnome2-common libgnomevfs2-0
libgnomevfs2-common liborbit-2-0 nvidia-367 nvidia-367-dev nvidia-modprobe
nvidia-opencl-icd-367 openjdk-8-jre openjdk-8-jre-headless
0 upgraded, 52 newly installed, 3 to remove and 18 not upgraded.

My questions are:

  1. Why would it do that as nvidia-375 is supposed to be the latest release?
  2. Can I just re-install nvidia-375 without breaking CUDA 8?



Hi Jon,

I’m just guessing, but I think there is a bug in some of the recent NVidia Linux (and Windows, too I think) drivers that breaks CUDA; more specifically FAH GPU Core21 will randomly report BAD_WORK_UNIT and stop the process, and this is seemingly caused by most recent NVidia drivers (older works fine).

There are several threads floating around (on different forums) about this error. See:

But at least according to the latter thread, both 35X nd 36X [EDIT]and most 37X should be affected (the first one that is not affected, is 376.xxx, but that is not available for Linux yet) [/EDIT].


This might mean that there is some other reason distribution chooses to do what it does (usually, the changelog of the package(s) in question is a good source for the reasoning).

Actually, the bug might actually be in the FAH_Core, according to the second thread I linked. Haven’t had time to investigate, as it is quite tedious to Google and read the forums, and I don’t care that much - I will just wait for new versions that will fix this (and don’t have much to input to help fix the issue via forums or other means).

I have experienced the same problem with cuda-repo-ubuntu1604-8-0-local-cublas-performance-update_8.0.61-1_amd64.deb, only I had the nvidia-384 driver installed on my laptop, and the package uninstalled it and replaced it with nvidia-375.

It would be nice if the NVidia driver installation was completely decoupled from the CUDA install, so that the user could choose which driver to install.

Alternatively, if the latest versions of the NVidia driver don’t play nicely with CUDA, then the installer could warn/notify the user, allowing the user to proceed at their own risk.

Hello! I yesterday buy video adapter Nvidia GT 1030, in previous time i use cuda with GTX650-E, and it working fine, with 375 driver (when i nstall it was same, and 375 was last).

I delete all software and install current 384.66 driver. It work fine. In next step, i nstall cuda, it downgrade driver to 375 version.

But 375 version, is not compatible with GT 1030, and i can see only black screen after start X and many errors in log… :-( And what i must do???

I solve my problem. just install drivers, 384.66 vesion, and “cuda-toolkit-8-0”, without runtime libraryes. also cublas libraryes, and NOT install drivers. It working.