The repository signing key used for the CUDA RPM and Deb packages has been updated to reflect modern standards. As a result, package managers may have problems with installing or upgrading CUDA 8 packages. If you are a Linux user upgrading from previous versions of the CUDA Toolkit using network installers, follow either of the options before proceeding. Users installing CUDA for the first time are unaffected.
Systems with repositories updated with CUDA 8.0 need to download and re-install the 'cuda-repo-*-8.0.44-1.*' package for your distribution of Linux. Re-installation of this package will import the new signing key with ID 7fa2af80. Select the target platform using the green buttons below to get the base installer instructions (using the network repo) for that platform.
For example, for RHEL7, the installation using the network repo:
a) # sudo rpm -i cuda-repo-rhel7-8.0.44-1.x86_64.rpm b) # sudo yum clean all c) # sudo yum install cuda
Manually import this key by downloading the file 7fa2af80.pub from the appropriate repository (for example, Ubuntu 14.04 x86_64 would be http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64/7fa2af80.pub) and then running, as root:
a) For Ubuntu distributions: `cat 7fa2af80.pub | apt-key add -`
b) For Fedora, RHEL and SUSE distributions: `rpm --import 7fa2af80.pub`
At a future date, CUDA packages will begin removing the old public key with ID 5C37D3BE. If you are only using the newly signed repositories, it is safe to remove the old key manually by running, as root:
a) For Ubuntu distributions: `apt-key del 5C37D3BE`
b) For Fedora, RHEL and SUSE distributions: `rpm -e gpg-pubkey-5c37d3be`
Tesla P100 requires a driver version that needs to downloaded from NVIDIA Driver Downloads. Follow these instructions depending on the installer type.
Xcode 8 is not currently supported with the CUDA Toolkit 8 compiler for Mac. We are working on providing an update shortly. Until then, the developers should use Xcode 7.x with the CUDA Toolkit 8 for compiling their code on Mac macOS 10.12. Alternatively, they can run applications compiled with older versions of CUDA Toolkit and Mac/Xcode on Mac macOS 10.12.