Well, SOMETHING has changed with my rig.
I can no longer add more than eight GPU's.
If I do, the OS wont boot.
When I try and add #9 I get this across the top of the screen...

/dev/sda2: clean, 262462/3571712 files, 1897677/142725512 blocks and it stops right there...

THANK YOU TBAR!!!!

I LOVE YOU MAN!

Are we there yet? Tick tock... C'Mon download, you can do it!

So tell me TBar, what is it that you do exactly? For a living I mean? You cant possibly do apps for Seti and get paid fer it can you? Do you "work" for Seti?? If you're doing this sort of thing, what other kind of apps do you tinker with besides Seti? Are they all number crunching things?

Yawn...

Naps are good when youre old.

Downloading the new version now.
Dont know the diff between the two apps so I'll stick with the BOINC all-in-one.
At least I had that working. I see it has grown a few Kb...

I wonder what happened to what I had?

Computer forensics... Heck of a job to have...

Stay with driver 384 if it works for you, I'm going to upload a new App to Crunchers Anonymous that will work with driver 384, so you can try that in a little while.

I did Say a Little While. Now is a Little While later.
Both Apps have been changed. You can use the All-In-One or the Linux_MultiGPU-v0.97_Special link, they both have been updated. You want to use the App setiathome_x41p_V0.97_x86_64-pc-linux-gnu_cuda90 with the NEW app_info.xml file.
The New CUDA 9.0 App Will Work with driver 384. It's the same App I'm using Now on my mining machine.

BLEH...Nap time

Well, nevermind. All the CUDA units are errors...

BUT WAIT!
The log says its a CUDA 9.0 driver so what the heck?
Odd, it says its both. CUDA 9.0 & OpenCL 1.2 CUDA... ???

Maybe I just dont have any CIDA work units yet?

Poop. My monster is dead...
That 384 driver is NOT a CUDA driver. Its running opencl.

Reinstall #9...

Cant do any harm at this point so I tried to reinstall it and its telling me its already in there so there MUST be something wrong with something thats been updated in Linux. None of this happened until I allowed it to update and whatever that update is seems to be staying with it somehow even though I keep the unit disconnected from the network.

Nope. Crapped out again. Connected to the internet or not, updated or not, something has changed. I am no longer able to get this running again.
Back at the same corrupt screen with the mucked up logon loop.
Gonna try removing the driver with using console.

Afraid to try the 384 driver with this BOINC. If it gets mucked up, reinstall is all I know how to do. UGH!

Wait, you said early on NOT to use the 384 driver with this VER of BOINC casue it wont work/will crash.
Is that not necessarily true? Might it work? I thought we discussed the 384 driver NOT being a CUDA 90 driver...

BTW, I definitely did register the driver module when you installed the driver...

Yep. Did all of that for sure. Always install the dependencies first.
Right now I'm staring at a new working install with the 384.10 driver from NVIDIA installed as told to do....
Right now the system is stable and runs fine with one GPU, no BOINC yet.

I'll try once more to install 396.54

Well, it certainly doesn't sound encouraging.
I have 11 cards working on my miner, and it's showing 8.1 GBs of memory being used. From that, I'd say you shouldn't try running more than 8 or 9 GPUs on a machine with just 8 GB.
Look at this post again;

I have tried several times to install this on the SUPER32 system with no success.
Same problem several tries. Log on loop. I was sure my execution of your instructions was flawless last two attempts.
Must be a driver system comparability issue? Dunno.
Cant hurt to try the 390 driver. ?? This is an old dual Xeon server board from ASRock.
If 390 fails the same way, I'll try the app anyhow.
If it runs without errors, I'll let it run for a bit to see if Linux crunches SOG units faster than windows.
I'm wasting 32 threads by not running this.

That is rather strange, I don't ever remember hearing those particular results before. Usually they eventually get it to work. There are a number of different things you could try, one would be the nVidia 390 driver. First a word on drivers. The Repository drivers are different than the drivers downloaded from nVidia and it is possible to mistakenly install both at the same time for the same device. Having both installed will trash your system, you must completely remove one before installing the other. However, it is possible to just install another driver from the same source, as the installed driver will be replaced by the new one. So, you can just run the 390 driver installer from nVidia without removing the 396 driver from nVidia. In that case the installer will remove the installed driver automatically before installing the new one. With the Repository driver you can install it from the console even if you can't get to the desktop. So, if you are in the driver loop, and in the console, you would remove the nVidia driver by either running nvidia-uninstall or ./drivername --uninstall. Then install the Repository driver while still in the console. Since we know your system has the 384 repository driver, after removing the nVidia driver run, sudo apt-get install nvidia-384. That will install the repository driver and solve the driver loop after rebooting...

I have tried it recently. To remove the Driver from nVidia while in the console first Stop lightdm, then run, nvidia-uninstall. After removing the driver from nVidia then run, sudo apt-get install nvidia-384 to install the Repository driver. That should solve the login loop.
Stay with driver 384 if it works for you, I'm going to upload a new App to Crunchers Anonymous that will work with driver 384, so you can try that in a little while.
BTW, if an update breaks the driver it means you didn't register the driver module when you installed the driver. Before the driver from nVidia starts to install it will ask you if you want to register the module to work with updated kernels, as before, I suggest you chose YES and register the module, else a kernel update will break the driver. You must have DKMS installed to register the driver module, that was covered earlier as well.

Well, thats it. I give up. I cannot restore the success I had achieved. The miner is down for the count and I have no idea why.

Several fresh installs this morning.

Allowed it to update software. Even tried it without allowing that to happen.

Only one GPU installed.

Installed the NVIDIA alternate driver.

Restarted.

Followed the instructions from TBar.

Doesnt work.

It comes back to a flickering messed up log in page that cant be gotten around.

I'm screwed.

Huh, now I cant get it to work no matter what I do...