The 12th Annual Folding@Home Holiday Season Race

[hiigaran]

Already have that. These are all the packages I have installed.

GPU folding on a headless system is definitely possible. My issue isn't on the fahclient end of things with nvidia-settings and Xorg. It's that I can't manually set the fan speeds, which prevents me from safely folding, as the temperatures can get quite high if left to auto speeds.

 

[biodoc]

Maybe you could generate a ~./nvidia-settings-rc with the appropriate commands and then put a line loading the config file in the ~/.xinitrc file?

nvidia-settings --load-config-only &
xterm &

See section 3 in nvidia-settings man page

 

[hiigaran]

I'm unable to perform any nvidia-settings function. The two aforementioned messages keep returning (connection refused and undefined display).

 

[StefanR5R]

Are you running nvidia-settings from within an X session (or outside an X session, but with an X session running in parallel and specified via -c to nvidia-settings)? You need to, as the man page states. Just having Xorg installed isn't sufficient; you need to have the X server running for nvidia-settings.

 

[Ken g6]

Come to think of it, I've never gotten Nvidia settings working over an ssh connection. Do it without X forwarding and it fails like yours does. Do it with X forwarding and it tries to change my local machine instead.

 

[hiigaran]

Apparently you can get it working over SSH with the magic cookie method.

Are you running nvidia-settings from within an X session (or outside an X session, but with an X session running in parallel and specified via -c to nvidia-settings)? You need to, as the man page states. Just having Xorg installed isn't sufficient; you need to have the X server running for nvidia-settings.

ps aux reports nothing X related running. Here's a list of all processes output by that command. I'm not actually sure how to get X started, either manually, or automatically. I had assumed it would start automatically after installing it and rebooting.

 

[Ken g6]

Usually you start X by starting a window manager.

What's the magic cookie method?

 

[hiigaran]

4th post. Came across it while trying to figure out the connection refused issue.

I'm going to guess I don't have a window manager installed. LightDM and GDM are the only two I can think of, neither of which are in my list of installed packages. Going to install LightDM and see what happens.

EDIT: Well, installing LightDM didn't help. Still has the connection refused message. Can only interact via SSH now, because it loads a GUI login that I cannot interact with. No mouse pointer, and can't switch to terminal with keyboard. Looks frozen.

One thing that's different is that nvidia-smi doesn't work now. The command gives me the message:

Unable to determine the device handle for GPU 000:04:00.0: GPU is lost. Reboot the system to recover this GPU.

dmesg also shows the following:

NVRM: Xid (PCI:0000:04:00): 79, GPU has fallen off the bus.

A quick search indicates that 79 could potentially mean hardware issues, such as defective GPU, or insufficient power, but this can't be the case, given that before I started this whole reinstallation thing, it had an uptime of several months running at full load.

Back to the drawing board!

EDIT2: Removed LightDM and installed GDM3. Not much better. Locally, the system is stuck in the late phases of booting, after the large list of [OK] messages. The monitor constantly goes blank, then returns to the same screen, almost like someone just keeps turning the screen off and on.

I can SSH just fine, however. smi works again, and top reports a Xorg process that seems to be using 50% to 75% CPU at all times, but smi reports no Xorg processes on the cards themselves. Going to try another display manager.

EDIT3: Aaaaand KVM/SDDM gives the same issues as LightDM. I've run out of display managers to try.

EDIT4: Apparently I don't need a display manager. I can just use startx. All it did was turn the screen blank for a couple of seconds, then display a message saying x successfully shut down. Logs show that it failed to initialise the GLX module. More research!

 

[crashtech]

So you want to fold on GPUs in a headless system. I'm not sure that's possible.

The one you helped me with has been headless since I got done configuring it. It was just set up as a regular desktop, though.

 

[biodoc]

The one you helped me with has been headless since I got done configuring it. It was just set up as a regular desktop, though.

If you mean headless as no monitor attached, yes 2 of my 3 computers are also "headless". I thought headless meant no x-server and windows manager. My mistake.

 

[crashtech]

If you mean headless as no monitor attached, yes 2 of my 3 computers are also "headless". I thought headless meant no x-server and windows manager. My mistake.

Oh, you are probably right! Not running X while also getting full GPU functionality is certainly above my pay grade.

 

[hiigaran]

This makes no sense. I tried installing the exact same version of Mint and nvidia drivers that worked on the original setup...and it still doesn't work.

I think I'm just going to give up on this.

 

[StefanR5R]

If fan control is the only feature you are missing, then a desperate(?) workaround would be to either unplug the GPU fans from the card and feed them externally instead (you would probably need to solder an adapter plug yourself), or have additional fans blowing onto the card.

This makes no sense. I tried installing the exact same version of Mint and nvidia drivers that worked on the original setup...and it still doesn't work.

Did they update packages in the meantime? (Since December I have seen a minor-version update of the driver reported in the updater, which I haven't installed yet myself. There were updates to cinnamon too; don't recall if any Xorg updates.) Do you start a different desktop? Do you have a different .xinitrc compared with your earlier setup?

 

[biodoc]

I think I'm just going to give up on this.

Don't give up!

A few days ago I decided to upgrade my main rig (3930K w/2x GTX1080) from mint 17.3 to 18.3. I decided to do a clean install. This rig is also a dual boot with windows 10.

This installation is similar to the one at the folding forum here.

I downloaded the 18.3 mint cinnamon image and started the installation. Once installed, I opened the update manager and did several rounds of system updates until the manager informed me that everything was up to date. I rebooted and then installed:

sudo apt install build-essential (this includes all the basic packages including kernel headers, gcc, DKMS, etc for compiling software and building new kernel modules. These packages are required for installation of the Nvidia drivers downloaded directly from Nvidia.

sudo apt install ia32-libs (32-bit app support for those rare times you need to run a legacy app and 32-bit openGL support)

Reboot.

Download the latest Nvidia driver (NVIDIA-Linux-x86_64-390.25.run)

chmod +x NVIDIA-Linux-x86_64-390.25.run (makes it executable)

blacklist the nouveau and other drivers:

sudo nano /etc/modprobe.d/blacklist.conf then append the following to the end of the file and save.

blacklist vga16fb
blacklist nouveau
blacklist rivafb
blacklist nvidiafb
blacklist rivatv

then to apply these changes:

sudo update-initramfs -u

sudo nano /etc/default/grub and changed "quiet splash" to "text" in the GRUB_CMDLINE_LINUX_DEFAULT line
(Linux will now start without a GUI)

sudo update-grub

reboot to terminal

log in and install the nvidia driver.

sudo ./NVIDIA-Linux-x86_64-390.25.run

sudo nano /etc/default/grub and changed "text" back to "quiet splash" in the GRUB_CMDLINE_LINUX_DEFAULT line

sudo update-grub

reboot

run nvidia-settings to confirm driver installation.

install python support for running FAHControl as per hiigaran guide.

wget http://launchpadlibrarian.net/109052632/python-support_1.0.15_all.deb

sudo dpkg -i python-support_1.0.15_all.deb

Then install FAHClient and FAHControl as per hiigaran guide.

 

[hiigaran]

I haven't installed ia32-libs, and I do recall the /.run file from Nvidia mentioning that it could not install 32bit compatibility, despite saying installation successful. Maybe that's the missing piece of the puzzle?

Right, I'll try it again...

EDIT: Can't do anything with it at the moment. Doesn't seem to be connecting for updates. Couldn't even connect to get the time during install. Not sure what's causing it.

EDIT2: And the original system does work, so I know there's no hardware issue.

 

[biodoc]

Maybe it is time to switch to Debian. I'm sure it will be frustrating at first but I'm tired of the "staged releases" of mint, ubuntu, etc. Maybe I'll try the Mint version (LDME 2 with cinnamon).

 

[hiigaran]

Tried Debian before. Several packages couldn't be installed. Message said certain packages were 'uninstallable'. Tried to manually install .deb packages in place of those, but no luck.

Anyway, I've hijacked this thread long enough. Going to continue in this thread.

 

[LANMAN]

Ugh.. getting another chassis delivered to power the last GPU I have. (Titan Z)
Hoping to get 1M more a day, but we'll see.
Its not scheduled to ship until 21 Feb..

I noticed that ppl in the Nvidia threads are saying miners are paying $1200 for 1080TI's?
That's absolutely insane...
Still haven't found anyone on the net running F@H on a Titan V either; case in point.. that's a lot of $ for one GPU.
I know is designed for more AI learning, so I wouldn't be surprised the cost vs. results would not be worth it.

 

[StefanR5R]

Check out the single-precision FAHBench results:
https://www.anandtech.com/show/12170/nvidia-titan-v-preview-titanomachy/5
What's missing there is how much power each card draws during this benchmark...

Here are some interesting findings:
https://hardforum.com/threads/nvidia-titan-v-observations.1951950/

 

[LANMAN]

Wow.. half the power consumption? That would almost pay for itself over time.
Interesting... thanks for sharing!

 

[StefanR5R]

Half the power consumption at almost same performance sounds a bit drastic. On the other hand,

shader count: 5120 vs. 3584
core clock: 1.34 GHz vs. 1.87 GHz (from the hardforum post)
core voltage: ? vs. ? (would be interesting to know)
manufacturing node: TSMC 12 nm vs. TSMC 16 nm
furthermore, memory: HBM2 vs. GDDR5​

And while GV100 is a lot larger than GP102, a classic single precision workload like Folding@Home certainly leaves many of GV100's transistors unused.

 

[ao_ika_red]

The beacons are lit! TeAm calls for aid!
http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...back-toms-vs-anand-part-2-march-12th.2539469/

 

[hiigaran]

Well, this is a bummer. PSU blew up on one of my folding rigs. At least it happened after the competition I guess!

Now I need to figure out how to use the 10 year warranty on this thing, since it was purchased in the US, and I'm a little over 12 000 Km away...

 

[Pokey]

Well, this is a bummer. PSU blew up on one of my folding rigs. At least it happened after the competition I guess!

Now I need to figure out how to use the 10 year warranty on this thing, since it was purchased in the US, and I'm a little over 12 000 Km away...

ugh. Where is a good teleporter when you need it??

 

[hiigaran]

Nah, teleporters would put me out of the job!

Currently going through the RMA process. Waiting for approval. Not sure if it takes long, or if I'm stuck on something though. I got a message saying my uploaded invoice of the PSU was rejected for a list of possible reasons and automatically removed from the system, but at the same time, it still shows that the invoice is still on their servers.

The only one of those possible reasons for rejection is that the name and address on the invoice don't match my own. Because I'm always flying around, I needed a reliable place for the hardware to be delivered to when I purchased it, so I had a friend buy the parts and have them delivered to his place for me to pick up from and pay in cash to. Could there be some sort of technicality about it being second hand as a result?