Hi all, I'm running C7 on i7 8700k/asus z370-a with an GTX1050. I noticed that there are several problems installing Nvidia proprietary driver. After one week of troubleshooting I got my solution. Hope that can help other user. I found this https://access.redhat.com/solutions/3438891 but I've not access to this content. In this content seems that the workaround is using lightdm and mask gdm. The case: I tried using proprietary driver from nvidia site and from elrepo. In all the case I got system that hungs during boot. In a first time I got message about a device probing (and some kernel trace about i915) and running systemd-analyze blame seems that udev-settle hungs for several minutes. Some suggested to disable udev-settle if you does not have LVM but this is not my case. After this error I got error with GDM (infinite hungs) so I tried lightdm that works blinking. So I switched to multi-user level, I got always udev-settle hung but running startx system works but with some strange behaviuour. Now I tried another solution. I installed kernel-lt, kernel-lt-devel and reinstalled NVIDIA driver. All works very well also reenabling graphical runlevel. best regards
On 09/20/2018 07:04 AM, Alessandro Baggi wrote:> Hi all, > I'm running C7 on i7 8700k/asus z370-a with an GTX1050. > I noticed that there are several problems installing Nvidia proprietary > driver. After one week of troubleshooting I got my solution. Hope that > can help other user. > ... > I tried using proprietary driver from nvidia site and from elrepo.NVIDIA actually releases their drivers in official RPM format for RHEL/CentOS. Much easier/faster to use and update than their shell script driver. They are in NVIDIA's cuda repo. You can install the repo files from here: https://developer.nvidia.com/cuda-downloads Choose "rpm (network)" for the last step to get to the yum repo. I use cuda but I think you can just install nvidia-kmod to get the driver only. A minor problem is that one cuda version and repo are only maintained for about half a year, then NVIDIA moves to next version. If you always want the latest NVIDIA driver, you might need to change the cuda repo URL to point to the latest every year. I only upgrade when TensorFlow supports a new version of CUDA. This may not solve your specific problem. I just think that there's no reason to going back to use NVIDIA's clumsy .run driver or any 3rd party driver when the official RPMs are easy to use/update. -- Yan Li
Alessandro Baggi:> > I found this https://access.redhat.com/solutions/3438891 but I've not > access to this content. In this content seems that the workaround is > using lightdm and mask gdm.That page is not exactly helpful ... virtually no details Strangely, it has 'Solution Verified' at the top of the page, but the listed 'Resolution' is:> At this time there is no guaranteed resolution of this issue. Using the > Nvidia Beta graphics module sometimes works, but is not guaranteed > or supportedWhat is more puzzling, is that we have probably over a thousand workstations running CentOS 7.5 with Nvidia cards using the proprietary driver and haven't seen any of the 'issues' listed - which are: * Nvidia proprietary module doesn't work after upgrading to RHEL7.5 * No graphics after updating to RHEL7.5 * Desktop does not start after upgrading to RHEL7.5 with Nvidia graphics driver * Black screen after installing Nvidia proprietary graphics module in RHEL7.5 We're using gdm which launches Mate James Pearson
On 09/20/2018 10:49 AM, Yan Li wrote:> NVIDIA actually releases their drivers in official RPM format for > RHEL/CentOS. Much easier/faster to use and update than their shell > script driver. They are in NVIDIA's cuda repo. You can install the repo > files from here: > > https://developer.nvidia.com/cuda-downloads > > Choose "rpm (network)" for the last step to get to the yum repo. I use > cuda but I think you can just install nvidia-kmod to get the driver only.Just checked. The latest repo changed the package name from nvidia-kmod to dkms-nvidia.> A minor problem is that one cuda version and repo are only maintained > for about half a year, then NVIDIA moves to next version. If you always > want the latest NVIDIA driver, you might need to change the cuda repo > URL to point to the latest every year. I only upgrade when TensorFlow > supports a new version of CUDA.I was wrong on this. NVIDIA now actively updates the repo to always point to the latest driver. No manual update needed. -- Yan Li
On 09/20/2018 10:51 AM, James Pearson wrote:> What is more puzzling, is that we have probably over a thousand workstations running CentOS 7.5 with Nvidia cards using the proprietary driver and haven't seen any of the 'issues' listed - which are: > > * Nvidia proprietary module doesn't work after upgrading to RHEL7.5 > * No graphics after updating to RHEL7.5 > * Desktop does not start after upgrading to RHEL7.5 with Nvidia graphics driver > * Black screen after installing Nvidia proprietary graphics module in RHEL7.5 > > We're using gdm which launches MateI am using the latest CentOS 7.5 gdm/GNOME with 390.30-2.el7 driver from cuda repo. Haven't seen any problem either. Chip is Quadro M2000M. -- Yan Li
Hi James, I tried as suggested installing cuda but the problem persist. In a previous c7.5 installation all works as expected but the driver was less then 390.77. So installing latest cuda nvidia driver for suggested links does not work. What other can I try? Il Gio 20 Set 2018, 19:52 James Pearson <james-p at moving-picture.com> ha scritto:> Alessandro Baggi: > > > > I found this https://access.redhat.com/solutions/3438891 but I've not > > access to this content. In this content seems that the workaround is > > using lightdm and mask gdm. > > That page is not exactly helpful ... virtually no details > > Strangely, it has 'Solution Verified' at the top of the page, but the > listed 'Resolution' is: > > > At this time there is no guaranteed resolution of this issue. Using the > > Nvidia Beta graphics module sometimes works, but is not guaranteed > > or supported > > What is more puzzling, is that we have probably over a thousand > workstations running CentOS 7.5 with Nvidia cards using the proprietary > driver and haven't seen any of the 'issues' listed - which are: > > * Nvidia proprietary module doesn't work after upgrading to RHEL7.5 > * No graphics after updating to RHEL7.5 > * Desktop does not start after upgrading to RHEL7.5 with Nvidia graphics > driver > * Black screen after installing Nvidia proprietary graphics module in > RHEL7.5 > > We're using gdm which launches Mate > > James Pearson > _______________________________________________ > CentOS mailing list > CentOS at centos.org > https://lists.centos.org/mailman/listinfo/centos >