nVidia Driver Install - crashing dpkg...

[ytene]

Apologies in advance for the length and complexity of this post. Also, I *hope* I'm following the latest recommended approach with what I'm trying to do...

I've had Mint 18.3 running perfectly on a dedicated SSD on my "gaming" PC ever since it went GA. Until yesterday, when an attempt to run Mint Update resulted in a machine that just went directly to kernel panic each time I tried to boot it [!!]. Today I decided to perform a clean installation - but things have gone from bad to worse and now I am questioning my sanity.

1. Hardware:-
Motherboard: MSI Big Bang Power II [Weird Name; nice board]
CPU: Intel Core i7 4960X Extreme Edition @ 4.3GHz
RAM: 32Gb of 2400MHz DDR3
GPU: nVidia 1080GTX

2. Installation Process:-
2.1. Create installation Media from the 18.3/Cinnamon/64 ISO to a USB 3 memory stick
2.2. Boot from USB stick in non-UEFI mode
2.3. Do not install proprietary drivers during installation
2.4. Manually partition a Samsung 850 Pro 128Gb SSD and install
2.5. Remove USB stick and reboot in software-rendering mode - all works perfectly...

Now, at this point I've tried 2 different approaches...

3. Approach to Install Graphics Driver [1]
3.1. Launch Driver Manager
3.2. Select latest nVidia Driver [384.11]
3.3. Attempt installation

4. Approach to Install Graphics Driver [2]
4.1. Perform Mint Update
4.2. Launch Driver Manager
4.3. Select latest nVidia Driver [348.11]
4.4. Attempt installation

Both of the above mechanisms get me "so far" ... the progress bar on the installer within Mint's Driver Manager gets to approximately 80% complete and then stops. When this happens input devices [mouse, keyboard] stop responding. After a brief period of time, the system reboots and returns in a software-rendering mode - ***but, there are nVidia software components now visible in my Menu, i.e. "Nvidia X Server Settings"

If I then attempt to resolve the problem, I quickly discover that there is some instability in the machines package database and I am recommended to run,

"dpkg --configure -a"

which I have used in the past without problem. This time, however, I get the following:-

Deleting 384.11
Loading new
First installation: checking all kernels
Building for only 4.10.0-38-generic
Building for architecture x86_64
Building initial module for 4.10.0-38-generic

and then, after another brief interval, my input devices lock, the system undergoes a soft reset, boots back in to Mint and once more I'm in software rendering mode with, apparently, no change...

I am wondering if this problem might relate in some way to the Intel Meltdown or Spectre bugs and/or code which might be associated with this. The hardware on which I'm trying to install Mint also runs a couple of different copies of Windows 10 Pro/64 [I have 4-bay, 2.5" drive dock on the front of my case to hold my boot drives - I can swap between different OS builds in the time it takes to complete a power cycle].

Could it be possible that my Windows builds have made a change to firmware on my system that is somehow now not compatible with a vanilla version of Mint 18.3? I've had the ISO image I'm using since it was released, if that's relevant.

If anyone can recommend a reliable mechanism to install a good driver for my 1080GTX, I'd be very grateful. Sorry again for the length of this post - and thanks for reading this far!

[catweazel]

3) Building for only 4.10.0-38-generic

1) Could it be possible that my Windows builds have made a change to firmware on my system that is somehow now not compatible with a vanilla version of Mint 18.3?

2) I've had the ISO image I'm using since it was released, if that's relevant.

If anyone can recommend a reliable mechanism to install a good driver for my 1080GTX, I'd be very grateful. Sorry again for the length of this post - and thanks for reading this far!

Numbered bullets added

1) Yes. Windwoes downloads firmware to the CPU. The firmware is not loaded permanently to the CPU so this depends on starting Windwoes then rebooting into linux. If you cold booted into linux then there is no Windwoes imposed firmware in the CPU.

2) Download a new iso, check the hash value and burn a new flash drive, and reinstall.

3) Update the kernel to version 4.13.0.36 before you install any additional drivers. If you can't get nVidia drivers installed with that, drop back and try 4.4.0.116.

[ytene]

Thank you Sir!

I'll give it a try - and report back...

[ytene]

OK, interim update...

I've downloaded a fresh copy of the 64-bit Cinnamon ISO and confirmed both the hash value and the signature - all look OK.

I've also checked the integrity of my original download ... and that looks good too. I am now about to re-format my USB key, load from the fresh download image and try again, using the specific sequence that you recommend here.

Thanks again for your guidance.

[ytene]

Interim Update #2...

Followed your recommendations. Having performed the initial build and rebooted, I then used Update Manager to select the 4.13.0.36 kernel and that item alone. The application completed without error.

I rebooted the system and confirmed the operation of the kernel through use of "uname -r" at a command prompt.

Next, I attempted to deploy the nVidia driver, using the driver manager, before attempting to install any other software [including any other drivers]. I observed exactly the same results as identified in my original post.

I will now try with 4.4.0.116.

Here's a curiosity, however... Last weekend, when I booted this machine for the first time in a couple of months, the system came up perfectly. The nVidia driver was working. I applied updates to the system at that time and it was that round of updates that caused the crash that led me here. I mention this only because I was clearly able to perform a build from this image and got a working nVidia driver. I'm still trying to figure out what I've changed or done differently, but nothing springs to mind yet.

Thank you

[ytene]

OK, final update for this attempt - and I regret that I have to advise of failure thus far.

Although I was able to find a copy of the 4.4.0-116 kernel at "ubuntu.pkgs.org", when I attempted to apply it directly I received an error message warning me that the package I was attempting to deploy was available directly via one of the upstream binary repositories... [Duh]. [ I have got *so* out of the habit of using Synaptic since moving from ubuntu to Mint... ]

Anyway, I used Synaptic to deploy the relevant meta-package, which brought in the following specific packages:-

linux-generic [the meta-package]
linux-headers-4.4.0-116
linux-headers-4.4.0-116-generic
linux-headers-generic
linux-image-4.4.0-116-generic
linux-image-extra-4.4.0-116-generic
linux-image-generic

As before, this installation went very smoothly. I had to hack grub in order to get it to display a boot menu [quick edit to /etc/default/grub followed by an grub-update] and upon reboot I booted into the 4.4.0-116 kernel and verified operation with "uname -r"

Unfortunately, I then attempted to install the nVidia drivers by means of the Driver Manager, only to see *exactly* the same results once more.

I am completely at a loss to explain this... How can I have possibly got this to work in the past? And I *know* I wasn't hallucinating - I booted the darn thing last Saturday!

I am sorry that this proved fruitless... Happy to take further guidance as to what you'd recommend next. [I've a little time on my hands today - just for the sheer fun of it [!!!] I'm going to have a go with 18.2 or maybe even an earlier release - just to see what I get. I appreciate that this won't be an ideal answer even if it works, but at this point I'd declare it a win if I get my baby back....

[ytene]

OK, 18.2 was another bust, but... I wonder if I’m on to something?

Up until now I’ve kind of been assuming that the latest nVidia driver is going to work perfectly; after all I have an entirely standard 1080GTX [albeit water cooled] and thus I have been happy to keep on loading that piece of software.

I noticed that when I tried to deploy the nVidia driver to the 18.2 build that I just created, it attempted to install the same 384.111 version of the nVidia drivers.

What if that is the problem?

When I go in to Mint’s ‘Driver Manager’ to look for proprietary drivers, it will only ever show me the latest available nVidia driver and doesn’t give me the option to try an earlier version. I’ve never tried to install an nVidia driver to Mint by any other means, but I’m going to build another fresh Mint 18.3 image and see if I can find a slightly older version of the nVidia driver to try.

At this stage in the proceedings, I can’t think of much else to try - and this is all just basic “trial and error” from here onwards...

[ClixTrix]

If I may ask, what is the make, model, and total wattage of your PSU?

[ytene]

Hello...

Thank you for chipping in - any contributions gratefully received at this point.

My PSU is a Zalman ZM-1250 Platinum PSU... It has one of the cleanest power deliveries of any PSU I've ever tested... and it has the added bonus of running in "silent mode" [i.e. load-based fan activation] when the total system draw is under 500 watts... I've not actually heard the fans yet...

In addition to the single 1080GTX and the single Intel CPU [the latter being slightly over-clocked to 4.3GHz] I have 32Gb RAM, currently 1 x Samsing 850 Pro SSD, 2 optical drives and a sound card fitted. I don't run any HDDs of any kind [2 NAS boxes for that], so hopefully the load on the PSU is actually quite reasonable...

[ytene]

I came up with one more thing to try [also a failure]...

I downloaded several of the most recent nVidia install files from their web site and had a go at installing the latest of this [version 390.42], just to see what would happen. It's not straightforward as you have to do this without the X server running, so I managed to tweak my grub boot loader to get me a command-line boot option and give it a try.

Everything started, but then I got an error which on-screen is reported as a "hardware error", but which the log file [ /var/log/nvidia-installer.log] reports thus:-


nvidia-installer log file '/var/log/nvidia-installer.log'
creation time: Tue Mar 13 16:55:19 2018
installer version: 390.42

PATH: /usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games

nvidia-installer command line:
./nvidia-installer

Unable to load: nvidia-installer ncurses v6 user interface

Using: nvidia-installer ncurses user interface
-> Detected 12 CPUs online; setting concurrency level to 12.
-> Installing NVIDIA driver version 390.42.
-> Running distribution scripts
executing: '/usr/lib/nvidia/pre-install'...
-> done.
-> The distribution-provided pre-install script failed! Are you sure you want to continue? (Answer: Continue installation)
ERROR: The Nouveau kernel driver is currently in use by your system. This driver is incompatible with the NVIDIA driver, and must be disabled before proceeding. Please consult the NVIDIA driver README and your Linux distribution's documentation for details on how to correctly disable the Nouveau kernel driver.
WARNING: One or more modprobe configuration files to disable Nouveau are already present at: /etc/modprobe.d/nvidia-installer-disable-nouveau.conf. Please be sure you have rebooted your system since these files were written. If you have rebooted, then Nouveau may be enabled for other reasons, such as being included in the system initial ramdisk or in your X configuration file. Please consult the NVIDIA driver README and your Linux distribution's documentation for details on how to correctly disable the Nouveau kernel driver.
ERROR: Installation has failed. Please see the file '/var/log/nvidia-installer.log' for details. You may find suggestions on fixing installation problems in the README available on the Linux driver download page at www.nvidia.com.

So this looks as though even though I've managed to boot into a text-only configuration, the nVidia driver installer is detecting that Mint pre-installs the Nouvea driver and is thus refusing to go any further.

I'm completely out of my depth now - no idea what to do next...

Any suggestions would be gratefully received.

Thank you

[ytene]

OK, one final, final update before I concede defeat...

Because I had nothing left to lose, I used another USB key to make myself a bootable image for Mint 17.3, Rosa, and dropped that on the same hardware [right down to the same SSD]... Another clean installation.

Observation #1: When I accessed "Driver Manager" to see what driver options I'd get, I was left looking at the exact same shortlist-of-one that I saw with both 18.3 and 18.2, namely the nVidia 384.11 driver. Which I already know doesn't work with the 18.x family. In other words, there is some form of [uncontrolled?] overlap between nVidia drivers and Mint releases; rather than having the abililty to say, "the latest version of nVidia driver supported by 17.3 was [ "version x" ], you are automatically left with no option than to take "the latest".

This could easily be an ubuntu design choice given they are the upstream source of packages. However, it does seem a little dicey since it leaves us with no way of performing elegant regression triage if - as appears to be the case for me - there is some form of conflict with the latest driver / latest OS combination...

So... I then went to Synaptic to see what options I could find there. Curiously, all the "versioned" nVidia packages included, by default, "transitional packages" which included everything up to and including the latest 384.11 release. In other words, any attempt to deploy a specifically versioned older driver looked as though it would lay 384.11 over the top... I did, however, find a package "nvidia-current", which did not bundle the 384.11 driver and was instead locked at v304.

That installed smoothly, but when I then attempted to run Cinnamon on top, the desktop just kept crashing on me. Information on line suggests that the v304 drivers go all the way back to approx June 2012 (I did not look too deeply), in which case they massively pre-date even Rosa's Cinnamon release. So it's entirely possible that Cinnamon was crashing because it could not find support for a needed feature in the underlying driver.

Now I'm *really* out of options.

[ytene]

Final update...

Eventually, I was able to get this to work. The problem turned out to be a compatibility issue between the latest kernels, which for Mint are Meltdown and Spectre-safe, compared with the packaged nVidia driver. It turns out that the version of the nVidia driver which is automatically presented to Mint users by the Driver Manager utility is not compatible with the "safe" kernels.

As a result, I had two choices, to regress to an earlier, unsafe kernel, or to forego the nVidia acceleration on my desktop.

After some consideration I decided to regress the kernel [to I think 4.3.36 or 4.3.37 - basically the most recent in the 4.3 branch] and from this platform I was able to successfully install the nVidia driver.

I'm rather hoping that this will sort itself out in the fullness of time. It is likely just going to take enough for nVidia and the kernel folk to realise they have a conflict and address it. But it seems reasonable that this might be a widespread problem and not merely a Mint problem, so hopefully it will happen sooner rather than later.

Sincere thanks to everyone who provided thoughts and assistance through this triage. If anyone else sees a similar problem and I can assist with more information about my final configuration, then please feel free to reach out and I'll do my best to help.

Thank you.

[bitelaum]

I'd like to report that I went into similar trouble with Linux Mint Cinnamon Edition 17.3 (Rosa) and was only able to install the NVIDIA drivers with the Driver Manager after going back to kernel 4.4.0-109 (tried with 121 and 119 before and it didn't work).

Now I'm running kernel 4.4.0-109 with the latest driver in driver manager (recommended option) 384.111.

My system is an Intel i7-3930K with 64 Gb DDR3-1600, running with a Geforce 660 (this is my workstation).

I hope this problem will be fixed soon, As I'd like to use a safest version of the Linux kernel.
I'm excited to check Linux Mint Cinnamon 19 (Tara) when it comes out too!