GUIDE: Linux Mint 18.3 Cinnamon with Nvidia GPU(s)

[brownstone]

It looks like the Nvidia drivers are indeed installed. FAH is setup to start automatically at boot. Here's the log:

Spoiler: FAH log

*********************** Log Started 2017-12-10T04:25:18Z ***********************
04:25:18:************************* Folding@home Client *************************
04:25:18: Website: http://folding.stanford.edu/
04:25:18: Copyright: (c) 2009-2014 Stanford University
04:25:18: Author: Joseph Coffland <joseph@cauldrondevelopment.com>
04:25:18: Args: --child --lifeline 1703 /etc/fahclient/config.xml --run-as
04:25:18: fahclient --pid-file=/var/run/fahclient.pid --daemon
04:25:18: Config: /etc/fahclient/config.xml
04:25:18:******************************** Build ********************************
04:25:18: Version: 7.4.4
04:25:18: Date: Mar 4 2014
04:25:18: Time: 12:02:38
04:25:18: SVN Rev: 4130
04:25:18: Branch: fah/trunk/client
04:25:18: Compiler: GNU 4.4.7
04:25:18: Options: -std=gnu++98 -O3 -funroll-loops -mfpmath=sse -ffast-math
04:25:18: -fno-unsafe-math-optimizations -msse2
04:25:18: Platform: linux2 3.2.0-1-amd64
04:25:18: Bits: 64
04:25:18: Mode: Release
04:25:18:******************************* System ********************************
04:25:18: CPU: AMD Ryzen 3 1200 Quad-Core Processor
04:25:18: CPU ID: AuthenticAMD Family 23 Model 1 Stepping 1
04:25:18: CPUs: 4
04:25:18: Memory: 3.87GiB
04:25:18:Free Memory: 3.27GiB
04:25:18: Threads: POSIX_THREADS
04:25:18: OS Version: 4.8
04:25:18:Has Battery: false
04:25:18: On Battery: false
04:25:18: UTC Offset: -7
04:25:18: PID: 1705
04:25:18: CWD: /var/lib/fahclient
04:25:18: OS: Linux 4.8.0-53-generic x86_64
04:25:18: OS Arch: AMD64
04:25:18: GPUs: 2
04:25:18: GPU 0: NVIDIA:7 GP102 [GeForce GTX 1080 Ti] 11380
04:25:18: GPU 1: NVIDIA:7 GP102 [GeForce GTX 1080 Ti] 11380
04:25:18: CUDA: Not detected
04:25:18:***********************************************************************
04:25:18:<config>
04:25:18: <!-- Client Control -->
04:25:18: <fold-anon v='true'/>
04:25:18:
04:25:18: <!-- Folding Slot Configuration -->
04:25:18: <gpu v='false'/>
04:25:18:
04:25:18: <!-- Network -->
04:25:18: <proxy v=':8080'/>
04:25:18:
04:25:18: <!-- Folding Slots -->
04:25:18: <slot id='0' type='GPU'/>
04:25:18: <slot id='1' type='GPU'/>
04:25:18:</config>
04:25:18:Switching to user fahclient
04:25:18:Trying to access database...
04:25:18:Successfully acquired database lock
04:25:18:Enabled folding slot 00: READY gpu:0:GP102 [GeForce GTX 1080 Ti] 11380
04:25:18:Enabled folding slot 01: READY gpu:1:GP102 [GeForce GTX 1080 Ti] 11380
04:25:20:WU00:FS00:Connecting to 171.67.108.45:80
04:25:22:WU01:FS01:Connecting to 171.67.108.45:80
04:25:23:WU00:FS00:Assigned to work server 171.67.108.157
04:25:23:WU00:FS00:Requesting new work unit for slot 00: READY gpu:0:GP102 [GeForce GTX 1080 Ti] 11380 from 171.67.108.157
04:25:23:WU00:FS00:Connecting to 171.67.108.157:8080
04:25:23:WU01:FS01:Assigned to work server 171.67.108.157
04:25:23:WU01:FS01:Requesting new work unit for slot 01: READY gpu:1:GP102 [GeForce GTX 1080 Ti] 11380 from 171.67.108.157
04:25:23:WU01:FS01:Connecting to 171.67.108.157:8080
04:25:23:WU00:FS00ownloading 8.86MiB
04:25:24:WU01:FS01ownloading 5.16MiB
04:25:29:WU00:FS00ownload 21.86%
04:25:32:WU01:FS01ownload 36.36%
04:25:35:WU00:FS00ownload 31.74%
04:25:38:WU01:FS01ownload 87.26%
04:25:40:WU01:FS01ownload complete
04:25:40:WU01:FS01:Received Unit: id:01 stateOWNLOAD error:NO_ERROR project:9414 run:1297 clone:3 gen:195 core:0x21 unit:0x000000ddab436c9d585e0699cac25a6f
04:25:40:WU01:FS01ownloading core from http://fahwebx.stanford.edu/cores/Linux/AMD64/NVIDIA/Fermi/Core_21.fah
04:25:40:WU01:FS01:Connecting to fahwebx.stanford.edu:80
04:25:40:WU01:FS01:FahCore 21: Downloading 3.23MiB
04:25:41:WU00:FS00ownload 76.87%
04:25:46:WU01:FS01:FahCore 21: 50.31%
04:25:47:WU00:FS00ownload 90.27%
04:25:52:WU01:FS01:FahCore 21: 59.99%
04:25:54:WU00:FS00ownload 93.09%
04:25:58:WU01:FS01:FahCore 21: 67.73%
04:25:58:WU00:FS00ownload complete
04:25:58:WU00:FS00:Received Unit: id:00 stateOWNLOAD error:NO_ERROR project:9431 run:297 clone:2 gen:43 core:0x21 unit:0x00000036ab436c9d586fdd3686053ca0
04:25:59:WU01:FS01:FahCore 21: Download complete
04:25:59:WU01:FS01:Valid core signature
04:25:59:WARNING:WU01:FS01:FahCore has not changed since last download, aborting core update
04:25:59:WU01:FS01:Starting
04:25:59:WU01:FS01:Running FahCore: /usr/bin/FAHCoreWrapper /var/lib/fahclient/cores/fahwebx.stanford.edu/cores/Linux/AMD64/NVIDIA/Fermi/Core_21.fah/FahCore_21 -dir 01 -suffix 01 -version 704 -lifeline 1705 -checkpoint 15 -gpu 1 -gpu-vendor nvidia
04:25:59:WU01:FS01:Started FahCore on PID 2196
04:25:59:WU01:FS01:Core PID:2200
04:25:59:WU01:FS01:FahCore 0x21 started
04:26:00:WU00:FS00:Starting
04:26:00:WU00:FS00:Running FahCore: /usr/bin/FAHCoreWrapper /var/lib/fahclient/cores/fahwebx.stanford.edu/cores/Linux/AMD64/NVIDIA/Fermi/Core_21.fah/FahCore_21 -dir 00 -suffix 01 -version 704 -lifeline 1705 -checkpoint 15 -gpu 0 -gpu-vendor nvidia
04:26:00:WU00:FS00:Started FahCore on PID 2201
04:26:00:WU00:FS00:Core PID:2205
04:26:00:WU00:FS00:FahCore 0x21 started
04:26:00:WARNING:WU00:FS00:FahCore returned: UNKNOWN_ENUM (127 = 0x7f)
04:26:00:WARNING:WU01:FS01:FahCore returned: UNKNOWN_ENUM (127 = 0x7f)
04:26:00:WU00:FS00:Starting
04:26:00:WU00:FS00:Running FahCore: /usr/bin/FAHCoreWrapper /var/lib/fahclient/cores/fahwebx.stanford.edu/cores/Linux/AMD64/NVIDIA/Fermi/Core_21.fah/FahCore_21 -dir 00 -suffix 01 -version 704 -lifeline 1705 -checkpoint 15 -gpu 0 -gpu-vendor nvidia
04:26:00:WU00:FS00:Started FahCore on PID 2206
04:26:00:WU00:FS00:Core PID:2210
04:26:00:WU00:FS00:FahCore 0x21 started
04:26:00:WU01:FS01:Starting
04:26:00:WU01:FS01:Running FahCore: /usr/bin/FAHCoreWrapper /var/lib/fahclient/cores/fahwebx.stanford.edu/cores/Linux/AMD64/NVIDIA/Fermi/Core_21.fah/FahCore_21 -dir 01 -suffix 01 -version 704 -lifeline 1705 -checkpoint 15 -gpu 1 -gpu-vendor nvidia
04:26:00:WU01:FS01:Started FahCore on PID 2211
04:26:00:WU01:FS01:Core PID:2215
04:26:00:WU01:FS01:FahCore 0x21 started
04:26:01:WARNING:WU00:FS00:FahCore returned: UNKNOWN_ENUM (127 = 0x7f)
04:26:01:WARNING:WU01:FS01:FahCore returned: UNKNOWN_ENUM (127 = 0x7f)

 

[TennesseeTony]

LOL! The forum converted so many parts of the log into 'big grin' smileys!

CUDA not detected. If you want to try something until someone smarter comes along.....

Open console:

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

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

sudo add-apt-repository ppa:graphics-drivers/ppa

sudo apt-get update

edit, missed a step: sudo apt-get -y install python-gnome2 mesa-common-dev freeglut3-dev nvidia-settings gedit

Reboot

 

[TennesseeTony]

Be sure to enter your PassWord, press ENTER, and/or type y for yes when it asks it of you.

 

[StefanR5R]

(from the Folding race thread)

I didn't think to try a disc...cpu stalled with usb a time or ten...

Is this a USB attached HDD or SSD, or is it only a memory stick? If the latter, both the flash memory and the controller will have much larger latencies than the former. In addition, it could wear out and die eventually from the write load which a system + applications partition needs to support.

Usb stick, but only as the boot media, installed to ssd or hard disk.

I take it your CPU Stalled errors were during the boot stage, right? I think I solved that with adding pci=nomsi to GRUB. I say 'think', because I had to do a lot of tweaks, so I can't be certain if that was the solution to the stalled errors, or other errors I had.

As a background: PCI Express devices may send interrupts in two ways: Via IRQ lines like classic PCI devices, or as Message Signaled Interrupt (MSI). The latter gives better performance and eliminates conflicts from shared IRQs. But the downside is that the MSI implementation is buggy on lots of PCI Express devices.

The kernel command line parameter pci=nomsi switches MSI off globally. There are also ways to switch off MSI per device. Linux device drivers also contain blacklists or whitelists to disable/ enable MSI on known bad/ good devices. I don't know how Windows is handling this widespread problem.

So in short, perhaps you had the USB stick with the Linux installer attached to a discrete PCIe USB controller with an MSI bug. (I wouldn't think that an Intel integrated USB controller had such a bug, or at least not have it but be whitelisted for MSI in the kernel.) Unless the stalls came really from a different device.

 

[biodoc]

@brownstone, it appears the nvidia drivers you downloaded don't have cuda/OpenCL support.

On the other hand, since you are starting the FAHClient as a service, perhaps the client is starting before the nvidia drivers load.

With FAHControl open, try the following command to restart FAHClient.

sudo /etc/init.d/FAHClient restart

 

[brownstone]

@Tony, I retried the steps you mentioned from the guide with no love.

@biodoc, restarting the FAHClient after loading to desktop and waiting a few minutes also does not work.

Any ideas how a guy gets the CUDA/OpenCL drivers installed?

I've just been using this command to download and install the drivers;
sudo apt-get install nvidia-xxx (xxx=driver # the last of which I tried was 387 I believe)

 

[biodoc]

If you have the 387 driver try:

sudo apt-get install nvidia-opencl-icd-387

sudo apt-get install libcuda1-387

 

[TennesseeTony]

I see that I missed the last line of code previously (what is python, cuda related or no?) : sudo apt-get -y install python-gnome2 mesa-common-dev freeglut3-dev nvidia-settings gedit

Hopefully Doc's direct-to-cuda method solves your issue. If not, maybe try the GUI, 'Driver Manager' and install drivers from there, if you haven't already tried that. And update the CPU micro-code if not already done.

I'm pretty sure you've added the GPU list text file, as the Ti's show up in your log. But if not, that was:

cd /var/lib/fahclient

ls

sudo wget http://fah-web.stanford.edu/file-releases/public/GPUs.txt

 

[biodoc]

Tony, python is a programming language as is cuda but they are not related.

I think the only sure way to get nvidia drivers with CUDA and OpenGL suppport is to download them directly from nvidia. Unfortunately they are very difficult for linux beginners to install.

 

[brownstone]

If you have the 387 driver try:

sudo apt-get install nvidia-opencl-icd-387

sudo apt-get install libcuda1-387

Brilliant! That seemed to be just what was needed. Thanks for your help biodoc and Tony!

 

[biodoc]

Excellent! Congrats!

 

[TennesseeTony]

AWESOME!!! Another 2M ppd for the TeAm! Well done, doc, and thanks for your persistence, brownstone!

 

[StefanR5R]

Thanks to all who contributed to this thread. I just finished setting up a new dual-GPU PC and it folds like there is no tomorrow.

 

[StefanR5R]

Just now I noticed that the FAHClient config.xml file still had

<fold-anon v='true'/>​

at the top. Actually I saw this not by inspecting the file itself, but by looking at FAHControl's Configure/ Expert tab. However, I believe the fold-anon tag was overridden by the passkey and user tags further down the config.xml file, because the PPD which I got recently were at the expected level. Nevertheless, I deleted the fold-anon option from the Expert tab now (and hence from config.xml).

 

[Pokey]

StefanR5R
I see the flags in the expert tab but I can find no discussion of them. They obviously have meaning for the Linux FAH Control. Why would you want to delete the fold-anon flag? Just asking. I am somewhat inexperienced when it comes to Linux.

 

[StefanR5R]

It's not really Linux-specific, or Mint-specific, but it was how the config.xml came with by default on my Mint installation. Hence I mentioned it in this thread.

But I probably misunderstood what this config option is about. I thought it instructs the client to submit results anonymously. Unfortunately the client user guide does not document this particular option. However, some older release notes tell what it is, e.g. this post:

v7.2.10:

• Added 'fold-anon' configuration option, fold even if not configured.

IOW the option, if set to true, tells the client to start folding even before user and passkey were entered. I presume the option has no further effect after user and passkey were added to the config.

 

[Pokey]

OK, thanks. I think I'll leave my flags alone, especially since things seem to be running smoothly.

I did look at my lone windows rig and it has none of these flags. Just open-web-control:true and I know what that does.

 

[TennesseeTony]

Oh, good, how do you turn that crap off, make it 'false'?

Also, Mint related but not Nvidia, my Intel HD 530 graphics is 'installed' but BOINC can't use it (does not show GPU options under 'Activity')(no openCL installed?). (Folding doesn't support Intel integrated graphics at all).

Any thoughts on getting BOINC to use the Intel graphics, on Mint/Linux? Only three projects support Intel GPU, that I know of, SETI, Collatz (about 1M points per week under Windows), and Einstein, and it's slower than Molasses in winter, but I like to think of it as an extra CPU core, for only 15-20 watts. And I'd like to make use of it.

 

[Kiska]

Any thoughts on getting BOINC to use the Intel graphics, on Mint/Linux? Only three projects support Intel GPU, that I know of, SETI, Collatz (about 1M points per week under Windows), and Einstein, and it's slower than Molasses in winter, but I like to think of it as an extra CPU core, for only 15-20 watts. And I'd like to make use of it.

You'd probably need to install some opencl headers, look on google for them

sudo apt install ocl-icd-libopencl1
sudo apt install opencl-headers
sudo apt install clinfo

These are the generics, that should expose opencl functionality to boinc

 

[TennesseeTony]

Some snags with Linux Mint....monitoring the systems.

1.) TeamViewer: Works well enough, same as on Windows, not great, but well enough. BUT, even though turning off auto-monitor detection, and setting resolution where I want it, when using a KVM, one system continues to revert to 640x480 every time I switch the KVM to it (but only if running TeamViewer)

2.) HFM.net...Where can one "typically" find the log file for Folding@Home on Linux? I seem to recall Linux can sorta put your files anywhere....but typically the home folder, or some such, unless you direct otherwise? More directly, perhaps, what are you experienced Linux Folders using to monitor your remote systems, preferably similar to what Mark uses, HFM.net?

 

[Kiska]

2.) HFM.net...Where can one "typically" find the log file for Folding@Home on Linux? I seem to recall Linux can sorta put your files anywhere....but typically the home folder, or some such, unless you direct otherwise? More directly, perhaps, what are you experienced Linux Folders using to monitor your remote systems, preferably similar to what Mark uses, HFM.net?

Usually in /var/logs/..........
Or its in /etc/fahclient/.........
Or here: /var/lib/fahclient/..........

Those are some choices, use ls

 

[StefanR5R]

On the Linux machine I simply went to FAHControl's "Configure" -> "Remote Access" and entered:

Console client remote access

Allow 127.0.0.1 192.168.0.0/24
Deny 0/0​

Passwordless IP Address Restriction

Allow 127.0.0.1 192.168.0.0/24
Deny 0/0​

You may need to adjust the addresses above to those in your LAN, of course.

This resulted in the following lines being added to config.xml (default path on Linux mint: /etc/fahclient/confing.xml):

  <!-- HTTP Server -->
  <allow v='127.0.0.1 192.168.0.0/24'/>

  <!-- Remote Command Server -->
  <command-allow-no-pass v='127.0.0.1 192.168.0.0/24'/>

Edit: The "/24" suffix means that only the the first 24 bits of the address are taken literally, the lowermost 8 bit are a wildcard. I.e. 192.168.0.0/24 really means 192.168.0.0...192.168.0.255. This range needs to cover the address of the node which is going to be used as central control. Note, if you want to perform remote access over public networks, you should only use password-protected access, i.e. drop the Passwordless/ command-allow-no-pass section.

On an OpenSUSE machine running FAHClient I also had to open port 36330 in the local firewall (or in fact, I disabled the local firewall). On my Mint installation, there is a local firewall too but it was off by default, hadn't had to change it. I don't have a local firewall on my Gentoo boxes either. Edit: I don't know if firewall modifications are needed on the node which is acting as the central control.

I haven't used HFM.net with this yet, only FAHControl on a central machine. Edit: HFM.net uses the same protocol though. At least with FAH v7, HFM.net no longer needs filesystem access to the log file on the FAHClient host.

I added all my hosts which may run FAHClient in the left side bar of the central FAHControl instance.

 

[biodoc]

StefanR5R, nice tutorial on how to use FAHControl to monitor and control other hosts running FAH. I use this method too.

If anyone would like to use HFM.net on linux, this protocol should work.

HFM.net is a microsoft NET framework app so you need to install mono.

On Mint or Ubuntu:

sudo apt-get install mono-complete

Create an HFM directory in your home directory.

cd ~
mkdir HFM

Download the latest version of HFM.net (ZIP file)

Find the zip file using the "gui file explorer", double-click on it to open the archive manager and then extract it in the newly create HFM directory.

To run HFM.net:

mono ~/HFM/HFM.exe

 

[StefanR5R]

How to control whether services (such as FAHClient and boinc-client) are started automatically at boot, and in which order:

Mint 18.3 apparently does not have a GUI for this. (E.g. OpenSUSE has one.) Nor did I find official documentation how one is supposed to control services on Mint. But this seems to work:

sudo update-rc.d FAHClient disable|enable [ runlevel ]​

(Don't cut and paste this. State either disable or enable, and optionally state a runlevel (e.g. 5) without brackets. See "man update-rc.d".) AFAIU, FAHClient with GPU slots needs to be started after X11. More research required...

 

[Howdy]

What's the trick to OCing a GPU in Mint? I have tried the cool-bits=5 (from what I have researched) which was supposed to add it to Nvidia X server settings. It's nowhere to be found.....I have worn out the "reboot" button.