Installing GPU Drivers in Linux.

[Spud1200]

Is their any relatively easy way of installing a Nvidia 750 GTX Driver in Linux 17.1 Mate release Rebeca. I've been looking around online and it seems complex. I've just had my 750 GTX Delivered today and want to use it for F@H. I'm completely perplexed by the complexity of Installing a GPU driver. I've been reading about the Terminal, installing Via the Command line and I'm really confused.


EDIT: I've got a Number of systems her and I have the one System with the 750 GTX in it. I've just installed some drivers Via. the update manager as that was the only system that was showing updates available Via. the update manager. My PPD has jumped to approx 8000 after installing the updates Via the manger. Now the System is processing a CPU x64 WU on Mint 64BIT Rebecca Release Mate desktop. just by the PPD can anyone tell me if the system has recodnised the GPU I've installed as it was previously taking up too 20 Minutes for a TPF and dropped to 4 Minutes.

I've been looking on Various Web Sites how to install the Graphics Drivers and I've installed and removed, done a reboot and purged I think some old GPU Drivers then done these updates.

Does that sound like the Card has been detected.

I've changed the drivers over from the default ones. To Nvidia Drivers under Driver Manager as Super User and the Control Panel has detected CUDA in the GPU. The System information is saying the GPU Is unsupported. Its Detected CUDA 5.0 . Why would it detect the CUDA cores but say the GPU isn't supported.

[bruce]

As you have discovered, installing GPU drivers in Linux is an advanced topic. Many flavors of Linux come with preinstalled drivers that support a Graphical User Interface but which may not contain the necessary interface for FoldingAtHome or other stream-oriented applications. Moreover, the generalize script supplied by NVidia doesn't work well on all Distros and something that might work on, say Mint, probably won't work on Redhat and might not even work on Ubuntu (which is very closely related to Mint).

Fortunately, those who support the various flavors of Linux understand the need for proprietary GPU drivers and they usually do provide a customized script in their update manager.

Detection of the card is not the same as having the right drivers, but frankly, I'm surprised that FAH worked at all with the old drivers, even if it was only at 20% of the expected speed.

Bottom line: It looks like you've done exactly the right things.

[Spud1200]

I've managed to get the GPU CUDA Cores detected in the FaH Control Panel. I'm wondering why my GPU is showing as unsupported. I've been reading about the GPU list available and I'm thinking it hasn't been downloaded.

I've started and stopped the application several times under Mint. I've re installed the fah client and control application hoping it would mabie update and download the GPU Compatibility list but it hasn't. Its showing as unsupported. I've got the card detected under mint as in the control setting I have the various options of adjusting the GPU settings and I've got the drivers installed. Its detected the CUDA Cores being 512 of them being a GTX 750 Armour MSI Edition. At least I'm hoping its supported b\c if it isn't I'm going to be gutted ma. I know the Ti is but just the 750 GTX I'm not sure tbh.

I've got the system sitting idle until I get a response.

I've been doping some reading and I can't seem to find the GPU.txt file for the white list under the fahclient folder in the main directory and when I try to copy the .txt file from these forums, I've got an message saying I don't have permission.

Code: Select all

*********************** Log Started 2015-05-23T22:46:47Z ***********************
22:46:47:************************* Folding@home Client *************************
22:46:47:    Website: http://folding.stanford.edu/
22:46:47:  Copyright: (c) 2009-2014 Stanford University
22:46:47:     Author: Joseph Coffland <[email protected]>
22:46:47:       Args: --child --lifeline 7863 /etc/fahclient/config.xml --run-as
22:46:47:             fahclient --pid-file=/var/run/fahclient.pid --daemon
22:46:47:     Config: /etc/fahclient/config.xml
22:46:47:******************************** Build ********************************
22:46:47:    Version: 7.4.4
22:46:47:       Date: Mar 4 2014
22:46:47:       Time: 12:02:38
22:46:47:    SVN Rev: 4130
22:46:47:     Branch: fah/trunk/client
22:46:47:   Compiler: GNU 4.4.7
22:46:47:    Options: -std=gnu++98 -O3 -funroll-loops -mfpmath=sse -ffast-math
22:46:47:             -fno-unsafe-math-optimizations -msse2
22:46:47:   Platform: linux2 3.2.0-1-amd64
22:46:47:       Bits: 64
22:46:47:       Mode: Release
22:46:47:******************************* System ********************************
22:46:47:        CPU: Intel(R) Celeron(R) CPU 430 @ 1.80GHz
22:46:47:     CPU ID: GenuineIntel Family 6 Model 22 Stepping 1
22:46:47:       CPUs: 1
22:46:47:     Memory: 1.95GiB
22:46:47:Free Memory: 587.73MiB
22:46:47:    Threads: POSIX_THREADS
22:46:47: OS Version: 3.13
22:46:47:Has Battery: false
22:46:47: On Battery: false
22:46:47: UTC Offset: 1
22:46:47:        PID: 7865
22:46:47:        CWD: /var/lib/fahclient
22:46:47:         OS: Linux 3.13.0-37-generic x86_64
22:46:47:    OS Arch: AMD64
22:46:47:       GPUs: 1
22:46:47:      GPU 0: UNSUPPORTED: GM107 [GeForce GTX 750]
22:46:47:       CUDA: 5.0
22:46:47:CUDA Driver: 6050
22:46:47:***********************************************************************

[bruce]

The GTX 750 is a supported GPU, but there's a file on your system called GPUs.txt which needs to be updated.

The FAHClient normally updates that file automatically but in a number of cases, it fails to do so if FAHClient gets started before your internet connection has been activated. I believe it will take care of itself if you do a sudo /etc/init.d/FAHClient restart. It that doesn't do it, we can help you dig deeper.

Also, it appears that you NVidia drivers may not be the latest ones.

[Spud1200]

The GTX 750 is a supported GPU, but there's a file on your system called GPUs.txt which needs to be updated.

The FAHClient normally updates that file automatically but in a number of cases, it fails to do so if FAHClient gets started before your internet connection has been activated. I believe it will take care of itself if you do a sudo /etc/init.d/FAHClient restart. It that doesn't do it, we can help you dig deeper.

Also, it appears that you NVidia drivers may not be the latest ones.

I have just closed down the fah application and restarted it using the command above with out any luck. Now however, I have just been playing around with some of the GPU settings and its detected the card then presented a error message with the slots failed after downloading a WU of approx 4 Days.I have selected the GPU under slots in the configure menu then set it as 0 as opposed to -1 and then its been detected then failed. When I selecting a -1 option for the system to decide on the GPU its showing as "No available GPUs."

Its going threw the process of Download, Ready and then Failed. I'm making progress but its slow.

[7im]

In the config.xml, check for a setting gpu=false and change it to true. Save and restart.

[Spud1200]

Hear is the Log File. The Slot has failed again.

Code: Select all

17:09:58:
17:09:58:  <!-- User Information -->
17:09:58:  <passkey v='********************************'/>
17:09:58:  <team v='12864'/>
17:09:58:  <user v='Spud1200'/>
17:09:58:
17:09:58:  <!-- Folding Slots -->
17:09:58:  <slot id='0' type='GPU'>
17:09:58:    <gpu-index v='0'/>
17:09:58:  </slot>
17:09:58:</config>
17:10:01:WU01:FS00:Download 100.00%
17:10:01:WU01:FS00:Download complete
17:10:01:WU01:FS00:Received Unit: id:01 state:DOWNLOAD error:NO_ERROR project:9007 run:1388 clone:2 gen:2 core:0xa4 unit:0x00000008ab40417c554e8b6be6283d87
17:10:01:WU01:FS00:Starting
17:10:01:WU01:FS00:Running FahCore: /usr/bin/FAHCoreWrapper /var/lib/fahclient/cores/web.stanford.edu/~pande/Linux/AMD64/Core_a4.fah/FahCore_a4 -dir 01 -suffix 01 -version 704 -lifeline 3422 -checkpoint 15 -gpu 0
17:10:01:WU01:FS00:Started FahCore on PID 3670
17:10:01:WU01:FS00:Core PID:3674
17:10:01:WU01:FS00:FahCore 0xa4 started
17:10:02:WARNING:WU01:FS00:FahCore returned: FILE_IO_ERROR (117 = 0x75)
17:10:02:WARNING:WU01:FS00:Fatal error, dumping
17:10:02:WU01:FS00:Sending unit results: id:01 state:SEND error:DUMPED project:9007 run:1388 clone:2 gen:2 core:0xa4 unit:0x00000008ab40417c554e8b6be6283d87
17:10:02:WU01:FS00:Connecting to 171.64.65.124:8080
17:10:02:WU00:FS00:Connecting to 171.67.108.200:80
17:10:02:WU01:FS00:Server responded WORK_ACK (400)
17:10:02:WU01:FS00:Cleaning up
17:10:03:WU00:FS00:Assigned to work server 171.64.65.124
17:10:03:WU00:FS00:Requesting new work unit for slot 00: READY gpu:0:GM107 [GeForce GTX 750] from 171.64.65.124
17:10:03:WU00:FS00:Connecting to 171.64.65.124:8080
17:10:04:WU00:FS00:Downloading 711.19KiB
17:10:06:WU00:FS00:Download complete
17:10:06:WU00:FS00:Received Unit: id:00 state:DOWNLOAD error:NO_ERROR project:9014 run:410 clone:12 gen:0 core:0xa4 unit:0x00000008ab40417c554e9abe4c03db89
17:10:06:WU00:FS00:Starting
17:10:06:WU00:FS00:Running FahCore: /usr/bin/FAHCoreWrapper /var/lib/fahclient/cores/web.stanford.edu/~pande/Linux/AMD64/Core_a4.fah/FahCore_a4 -dir 00 -suffix 01 -version 704 -lifeline 3422 -checkpoint 15 -gpu 0
17:10:06:WU00:FS00:Started FahCore on PID 3675
17:10:06:WU00:FS00:Core PID:3679
17:10:06:WU00:FS00:FahCore 0xa4 started
17:10:07:WARNING:WU00:FS00:FahCore returned: FILE_IO_ERROR (117 = 0x75)
17:10:07:WARNING:WU00:FS00:Fatal error, dumping
17:10:07:WU00:FS00:Sending unit results: id:00 state:SEND error:DUMPED project:9014 run:410 clone:12 gen:0 core:0xa4 unit:0x00000008ab40417c554e9abe4c03db89
17:10:07:WU00:FS00:Connecting to 171.64.65.124:8080
17:10:07:WU01:FS00:Connecting to 171.67.108.200:80
17:10:07:WU00:FS00:Server responded WORK_ACK (400)
17:10:07:WU00:FS00:Cleaning up
17:10:08:WU01:FS00:Assigned to work server 171.64.65.124
17:10:08:WU01:FS00:Requesting new work unit for slot 00: READY gpu:0:GM107 [GeForce GTX 750] from 171.64.65.124
17:10:08:WU01:FS00:Connecting to 171.64.65.124:8080
17:10:09:WU01:FS00:Downloading 711.13KiB
17:10:12:WU01:FS00:Download complete
17:10:12:WU01:FS00:Received Unit: id:01 state:DOWNLOAD error:NO_ERROR project:9014 run:411 clone:9 gen:0 core:0xa4 unit:0x00000008ab40417c554e9abe5b15070e
17:10:12:WU01:FS00:Starting
17:10:12:WU01:FS00:Running FahCore: /usr/bin/FAHCoreWrapper /var/lib/fahclient/cores/web.stanford.edu/~pande/Linux/AMD64/Core_a4.fah/FahCore_a4 -dir 01 -suffix 01 -version 704 -lifeline 3422 -checkpoint 15 -gpu 0
17:10:12:WU01:FS00:Started FahCore on PID 3680
17:10:12:WU01:FS00:Core PID:3684
17:10:12:WU01:FS00:FahCore 0xa4 started
17:10:13:WARNING:WU01:FS00:FahCore returned: FILE_IO_ERROR (117 = 0x75)
17:10:13:WARNING:WU01:FS00:Fatal error, dumping
17:10:13:WU01:FS00:Sending unit results: id:01 state:SEND error:DUMPED project:9014 run:411 clone:9 gen:0 core:0xa4 unit:0x00000008ab40417c554e9abe5b15070e
17:10:13:WU01:FS00:Connecting to 171.64.65.124:8080
17:10:14:WU01:FS00:Server responded WORK_ACK (400)
17:10:14:WU01:FS00:Cleaning up

Is their anything I can do. Or is it hopeless..

[bollix47]

For some reason your client is downloading CPU projects for your GPU which will not work. The info in your log prior to your user info might be of help (it should list GPU=true) but if it was me I'd completey uninstall FAHClient including data, re-install it and make sure you have a GPUs.txt file in /var/lib/fahclient after the client starts up.

If it's not there then see viewtopic.php?f=96&t=27734 for some help getting the file in the right location.

[Spud1200]

I've uninstalled the Client and reinstalled it several times. I have no option to cut and past the GPU.txt file as I received a message that I don't have the admin rites & I'm not the owner to modify the file \ folder. I have been using GPU Drivers found under Driver Manager in the Control Panel in Linux Mint and it does say their are no propriety divers in use.

BOINC detected the card, but the previous drivers I had on their they were not recognized by BOINC but it did notice the card.

Would it have anything to do with the fact its a single core system and hence why its downloading CPU work and then having to manually set up the GPU after I boot the client. I don't want to ditch any WU if I can help it but I have done so with the CPU needing to support the GPU and hence have removed the slot to accommodate the GPU.

The GPU was being recodnised on Slot 0 not -1 and then returning a failure which you see by the log. ..

[bollix47]

There is no mention of cut/paste in the link I gave you ... all you need to do is open the following link in your browser and then click on the File menu or right click anywhere on the page and select Save as ... it will fill in the blanks for you and all you have to do is click on Save:

https://fah-web.stanford.edu/file-relea ... c/GPUs.txt

After you've saved it the file should be in your Downloads folder and the instructions for moving that to the proper location are also in the link in my last post:

viewtopic.php?p=276614#p276614 :

Open a terminal (Ctrl-Alt-t) (enter your password when asked)

Code: Select all

cd Downloads   
sudo cp GPUs.txt /var/lib/fahclient
sudo /etc/init.d/FAHClient restart

[7im]

Uninstall is not the same as purge. Not all settings are removed with a simple uninstall. For example, if the client was installed (the very first time) with no supported GPUs, then the default in the config.xml will always say GPU = False. And I have a feeling that is part of the reason some Linux installs are having a problem with the GPUs.txt download/update routine, and hence the problem installing a GPU slot. (same behavior having to purge and reinstall to fix the start as a service check box)

We eventually need to get the drivers up to 346.xx or above to get CUDA to 7000 or higher for core_18. Maybe Jesse's driver install post would help.

[7im]

@ Spud1200 - have you tried changing the GPU setting from -1 to 1?

[Spud1200]

To be honest I think its a Driver problem as BOINC isn't recognizing the GPU either. I have under Driver Manager in the Linux Mint Control panel selected the latest drivers that are available from the list but the device at the top of the pop up box isn't being listed. Its recognizing OpenCL but not the CUDA Cores and hence BOINC is saying that it doesn't recodnize the Driver version from what I presume the Generic List of available drivers. I think 7im you posted a Topic of installing drivers under Mint in which you have to leave the GUI and use the Terminal to get the drivers successfully installed and that might be the only way around it.

I downloaded the latest drivers from the NVidia Web Site and Mint 86_64 wasn't recognizing the .Run File I had downloaded from their web site despite selecting Linux 64 Bit for some reason. I have been threw everything and I'm convinced after posting on the BOINC message boards since its (BOINC) not seeing the GPU as well its a driver Problem. The Driver Manager is also saying no propriety drivers are in use but I am selecting NVidia Driver 352.xx (I think, I would have to check) from the list available.

EDIT: Yeah I've also selected the 0 -1 and 1 option from the GPU settings available. I have on "1" settings been able to get the slot recognized but it fails after downloading CPU Work as posted in my Log.

[bruce]

The other possibility is that this is a simple issue of permissions. The installation normally creates a new user called fahclient. This user owns the files created by the FAHClient executable, including GPUs.txt and the /var/lib/fahclient directory. That's why bollix47 told you to use sudo cp rather than just cp. All of that is taken care of if you use /etc/init.d/FAHClient to start the executable but if you're starting it by a different method or if the files were created with different permissions, that could be why GPUs.txt isn't getting updated automatically.

[Spud1200]

The other possibility is that this is a simple issue of permissions. The installation normally creates a new user called fahclient. This user owns the files created by the FAHClient executable, including GPUs.txt and the /var/lib/fahclient directory. That's why bollix47 told you to use sudo cp rather than just cp. All of that is taken care of if you use /etc/init.d/FAHClient to start the executable but if you're starting it by a different method or if the files were created with different permissions, that could be why GPUs.txt isn't getting updated automatically.

Under driver Manager in the control panel as super user I have, "NVIDIA Corporation: Unknown." Thus below, This Device is listing an alternative Driver. In this panel of option (Driver Manager) I have selected as Version. 340.76 NVidia Binary Driver for the GTX 750. I also have No proprietary drivers are in use.

I have the option from another three Driver Versions being: NVidia 346.72 and or 352.09 or again third and final option 349.16 as drivers direct I presume from NVidia.

What is the latest Driver that will be supported by the fah client do you know for this card the GTX 750.

I have been looking at the permissions of the file \ folders and I looked at the property's box a bit like looking at the file \ folder information under windows and it said I wasn't the owner so I couldn't modify the contents by placing the GPU.txt file in the main directory. Also their was no GPU.txt file in the directory what so ever to start with when I have been looking at the fah client installation under var/lib/fahclient . Do you know why it would download CPU WU's for a GPU.

When I'm starting the client I'm having to ditch the CPU Slot as I need it to support the GPU as its only a single core System and I absolutely hate ditching WU. I just hate doing it. Now thing is thats when starting the Client fah threw the GUI being the Menu and all programs. I have not done a purge of the fah client at all so I don't know if their are any files in other parts of the system like windows such as a main registry. The issue I have what I'm saying is being a single core system, with a GPU Slot available and a GPU I'm having to ditch the CPU Slot for the GPU and I'm curious if that will cause problems in it self. I have selected under config.XML the value of true for the GPU and selected slot 1 but still with a failure so thats why I've posted the Drivers available under driver manager and what I have currently running.