Hello Fabian,
On 26.09.19 09:46, Fabian Arrotin wrote:
On 25/09/2019 10:30, Ralf Aumüller wrote:
...
today I updated a CentOS 7.6 ppc64le machine to CentOS 7.7. After reboot to the new kernel (4.18.0-80.7.2.el7.ppc64le) dkms could not build the nvidia-module.
...>> Any comments?
thanks for Your quick response.
Well, if you use that kernel, that means you're on Power9 variant, and that architecture doesn't exist anymore upstream (so no RHEL 7.7 for Power9). As almost all packages are just ppc64le (which still exist upstream), the decision was to still provide 7.7.1908 for Power9 users, but using the kernel from CentOS 8, rebuilt for CentOS 7. (same is also true for aarch64)
For that kernel to be built, we had to use newer gcc, that you can find/use through devtoolset-8 : http://mirror.centos.org/altarch/7/sclo/ppc64le/rh/devtoolset-8/
Ok. So I try to install devtoolset-8 and build the nvidia-driver with that gcc.
Curious : which kind of machine do you have that has both a Power9 and nvidia ? that seems to *not* be an IBM node, but a kind of openpower workstation ?
It'a a IBM Power System AC922 (8335-GTH) with Nvidia Tesla V100 graphic cards. Supercomputer "Summit" uses this nodes (https://www.olcf.ornl.gov/summit/).
PS2 : worth creating a bug report on https://bugs.centos.org for easier tracking and also indexing, so that other people in your situation would follow the bug report (index by crawlers) and eventually discussion can happen there.
I will do that.
Best regards, Ralf