Folding Forum

Hey everyone,

My brand-new Asus GTX 970 just arrived from Amazon, and I'm really excited for its potential. Unfortunately I haven't been able to fold with it yet. I'm running Linux Mint 17 LTS (Ubuntu 14.04 package base of course) and I installed the Nvidia 343.22 drivers from xorg-edgers following http://www.unixbuzz.com/2014/09/install ... 14-04-ppa/ as a guide. I also installed nvidia-settings, nvidia-modprobe, and nvidia-libopencl1-343 for good measure, then removed that PPA.

The machine has the 970 and a 480, so both support OpenCL.

Nvidia-settings shows 343.22 is installed, and it appears to be working just fine video-wise. The Drivers tool has Nouveau selected, but shows the 480 has using "an alternative driver." Steam games work great, but OpenCL doesn't. When I run "./ocore --devices" it throws "No OpenCL Compatible Devices Found" and work units download by a FAHClient GPU slot error out almost instantly with "Bad platformID size" and "BAD_WORK_UNIT (114 = 0x72)"

Any suggestions? Should I uninstall 343, use the Drivers tool to switch to nvidia-331, then install 343 again? Any other ideas?

My suspect would be nouveau ... I needed to blacklist it via grub to get NV installed (but original version, no ppa)
Right now still on 14.04 working. 343.22

Ocores I tried but mixed up CUDA 6 and 6.5; need to wait until weekend

ChristianVirtual wrote:My suspect would be nouveau ... I needed to blacklist it via grub to get NV installed (but original version, no ppa)
Right now still on 14.04 working. 343.22

Ocores I tried but mixed up CUDA 6 and 6.5; need to wait until weekend

You shouldn't have to blacklist Nouveau. This package should work. Installing the drivers is simple with "sudo apt-get install nvidia-current" or nvidia-331. Something's up with 343.22 or I installed it wrong.

Ok, so I tried the 340.58 driver and my 480 is recognized.

So neither my 480 nor the 970 were showing support for OpenCL with 343.22. Out of curiosity, I tried the following:
1. Purged all 343.22 driver packages.
2. Restarted, booted via Nouveau since that's the default.
3. Install 340.58 driver via the Driver tool.
4. Rebooted, booted via 340 driver.
Now the 480 is recognized as having valid OpenCL support, but the 970 isn't. I would conclude that the 343.22 driver does not contain the OpenCL component, but others can fold with it so I don't know what's happening here.

$sudo apt-get install nvidia-current-updates nvidia-opencl-dev

7im wrote:$sudo apt-get install nvidia-current-updates nvidia-opencl-dev

Good idea. nvidia-current-updates is quite old however, (it references 304) so I installed nvidia-343 and nvidia-opencl-dev instead. Unfortunately, that did not work. The 970 is recognized, but now neither card shows OpenCL support according to ./ocore --devices

I'm going to stick with 340 for the time being since at least it has OpenCL support for one card.

https://devtalk.nvidia.com/default/topi ... -releases/

This is really weird Jesse. All I know is that on Ubuntu 14.04 I ran into the same problem after a manual driver install, but then I installed nvidia-opencl-dev and nvidia-modprobe and I think that was enough for me to get it working. Another person on IRC couldn't get it working with Mint either. Could there be something Mint specific throwing it for a loop? I'm scared to touch mine now as people are having so much trouble (and my 980/750Ti are working great). Mine is a old install of Ubuntu so it has many old packages (compilers/etc.) and the newest CUDA toolkit designed specifically for the 970/980 (as I needed to run some CUDA on it). No idea why any of those should help, but I guess you could give them a try (at least it should get CUDA working).

Let's try to come up with some instructions that just work ... at least most of the time.

Prerequisites ( if X is not currently working the following instructions can be entered via root prompt - see * below ) :

Open a terminal (ctrl-alt-t) The latest linux 64-bit driver ( currently 346.35 ) is available from nvidia and is downloaded above using wget ... if you want to use a different version adjust that line, the line following it and the one a few lines below here in the code box.

The following instructions will no longer be on the screen once you switch from X so if you don't have access to another computer it could be useful to print them or save them to a text file so that you can view it later by using the cat command.

ctrl-alt-f1 to proceed to login screen and login ( see * below if ctrl-alt-f1 doesn't give you a login screen ) Answer Accept or Continue or Yes to all questions ... that does mean moving the highlight where appropriate. Simply pressing Enter for each question will not create a working installation.

sudo reboot

*
If ctrl-alt-f1 doesn't work you need to get to the root prompt via the grub menu.
Reboot and select Advanced, then recovery mode (2nd entry ...you might need to hold down the (right) shift key while booting to show the grub menu if you only have one o/s on the system).
In the recovery menu select the grub entry which will change your o/s from read-only to read-write.
Then select the Network entry.
After that's done select the root entry.
Continue with the rest of the instructions ... since you're already root there's no need to use sudo.

As Jesse mentioned earlier, after the driver is assumed to be loaded properly you can avoid a bunch of instant folding failures by first testing for opencl. Download ocore_601_OpenCL_v20 from http://stanford.edu/~yutongz/ocores/ and run: If the driver has been successfully installed you should see something similar to: If all appears to look good but FAHClient still does not recognize your card in the System Info area try downloading the GPUs.txt file manually and moving it to the folding Data Directory which on a standard Linux install is /var/lib/fahclient. The location could be elsewhere depending on how the client was installed and how it was started but it should be wherever the cores, logs and work sub-directories are:
Edited Jan 04/15 - removed xorg-edgers PPA method and performed a general cleanup based on user feedback

You do realise we are talking about Linux and Nvidia here, which can be quite a frustrating animal, you can do everything right and use as many methods as you want when installing nvidia drivers on a Linux OS and still fail. People should not get discouraged when trying to install nvidia drivers on Linux it can take many attempts to get a Linux nvidia driver to work properly. I think a very important thing to add to your instructions is that if you see a report of someone else getting a driver to work then odds are they will work on your system also it just may take many attempts to get them installed properly.

When attempting to install Nvidia drivers on linux there are a couple of commands that are important to remember for completely removing a failed install on Ubuntu run.

$ sudo mount -o remount,rw /
$ sudo apt-get purge nvidia*
$ sudo reboot

I have found that you may have to do this several times when trying to install nvidia drivers before you get a successful proper working install that will fold and it can be very frustrating. Anyway to go along with bollix47 instructions there is a pretty good instruction on updating nvidia drivers on 14.04 here. http://www.binarytides.com/install-nvid ... ntu-14-04/ and good luck

I couldn't agree more Grandpa. I've had both methods fail at one time or another. If I don't forget to purge the old nvidia software and I follow the steps properly it usually just works. That's why I'm trying to get a set of steps that should work most of the time for most people if done exactly as listed. I doubt anyone has come up with something that works for everyone all of the time but maybe we can get close.

bollix47 wrote:1a. add the ppa and install

Code: Select all

sudo add-apt-repository ppa:xorg-edgers/ppa 
sudo apt-get update
sudo apt-get install nvidia-graphics-drivers-343

I believe it's important to install the Head Package so as not to miss anything necessary.

There is no nvidia-graphics-driver-343, nvidia-343 is the name of the right package AFAIK. See for yourself: "apt-cache search graphics- | grep nvidia" returns nothing.

bollix47 wrote:

Code: Select all

ctrl-alt-f1 to proceed to login screen and login * 
sudo service lightdm stop 
(change lightdm if using a different desktop manager)
sudo apt-get purge nvidia*

On Mint, this is "service mdm stop". If you run the driver installer after that, you will get an error about a X11 or Nvidia component being in use by a process. It will say what process it is. Simply kill that process number with "sudo kill -9 <num>"

It is there though even if it does not show up in a search.

Your picture does not show nvidia-graphics-driver-343. "sudo apt-get install nvidia-graphics-driver-343" returns "E: Unable to locate package nvidia-graphics-driver-343"

nvidia-343 does exist and is the current driver, although like I said when I install it, I get 970 and 480 support, but no OpenCL. I suspect it isn't the headers as I was missing those packages with the nvidia-340 package and I have OpenCL support for the 480 with that. I don't know enough about Linux to know how the ocore queries for OpenCL support, or where the OpenCL component is.

When I ran it a couple of days ago it said in Nvidia Xserver settings and the additional drivers said 343 was installed and was being used I also ran $ sudo apt-get install nvidia-current-updates nvidia-opencl-dev sepratley after installing 343 and rebooting. It did take several attempts at installing and removing before it worked properly. F@H ran just fine with 343 but was no improvement over 340 for the GTX 680 so I went back to 304 which is the best driver for the 680.

I did get an error on several attempts to install. Just curious have you tried running sudo apt-get install nvidia-331 instead of sudo apt-get install nvidia-graphics-driver-343, I can install it again if you need me to verify again that it works.

Just a quick update:

Using the PPA I see the same problem that Jesse sees ... i.e. no opencl compatible devices found.

Downloading and installing the 343.22 drivers from nvidia does render a working system including finding and identifying my Maxwell GPU as being opencl compatible.

Not sure if xorg-edgers have done something related to this or even if it's related in any way.

For now my Maxwell is back to folding ... testing will continue after more investigation.