[Ubuntu] Setting up the GPU client

Disclaimer: This is the longest, and frankly, most annoying guide I've had to make. I take no responsibility for any monitors that are destroyed in anger.

This guide will cover how to install the GPU client (for nVidia cards only) on Ubuntu v10.04. There is no way to install a client for ATi on Ubuntu at the moment.

It will also show you how to install HFM, and use it to monitor your client.

GPU Guide:

What you need:

You must make sure Ubuntu is up to date. Go to System> Administration> Update Manager, and install all updates:



Installation:

1) Open Terminal. It can be found in Applications> Accessories> Terminal (you can drag it to the panel to make it easier).

Enter this command:

2) Now, download the latest nVidia drivers from here. Don't use the "Hardware Drivers" supplied by Ubuntu. Move that file to the Desktop.
(Direct Link to 195.36)

Run this command:

Now create a file on the Desktop (Rightclick> Create Document> Empty File). Name it "nvidia-graphics-drivers.conf". Open it, and type (or paste) in these lines:


Then open Terminal and enter these commands:

Create another file on the Desktop, this time naming it "xorg.conf". Open it and type this in:

Open Terminal again and use these commands:
2) Restart the computer. When rebooting, you will be told you can't start X, so choose the Low Graphics option instead.

*Get some paper and a pen for the following commands; you won't be able to check them for this section.*

Open Terminal and type this:
Your screen will go blank and you will be brought to a text based login, so have your written commands ready. Enter your username and password.

Now, install the drivers by using this command:
Choose "Accept" to begin installation. When finished, it will ask if you wish to use the OpenGL libraries; say no. Finally, Select "Yes" to reconfigure X.

Now, enter this command to log in again:

You can throw that piece of paper away now.

3) Now we need to download the CUDA 2.3 Toolkit from here (it's a direct link). Move it to the Desktop. Open Terminal and run the installer:

Just press "Enter" when asked where to install it.



Now we need Ubuntu to be able to find the CUDA binaries/libraries. Enter the following commands:

Restart your computer.

4) Now we will configure Wine. First, we will need to install it. Open Terminal, and enter:

To configure it, go to Applications> Wine> Configure Wine.

Make sure the Windows version is "Windows XP":



5) We now need to install the Wine CUDA .dll files. Dowload this file, and put it on the Desktop:
http://www.gpu3.hostei.com/wrapper2n.../cudart.dll.so


Open Applications> Wine> Browse C:/ Drive> windows> system32. Copy "cudart.dll.so" to system32, then rightclick and rename it to "cudart.dll"

Next, rightclick on the .dll file, select "Properties", and go to the "Permissions" tab. Change to:

Open Terminal again, and enter:

Thank you to Tasmac for this part.

6) Now we begin installing the GPU client itself. Head to the Stanford High Performance downloads page. Get the third client in the list, under "Windows XP Console client". Make sure to move it to the Desktop.



Now, you will extract the files into the appropriate directory. Run this command:

Multi GPU users, change nvidia0 to nvidia1 for the 2nd GPU, and so on for all GPUs in that system, then run the command again.



7) Now, while the client is set up, we want to make a launcher (so you don't need to constantly enter commands to start the client.

So, open Terminal and enter this:

When the file opens, copy and paste this in:

If using Multiple GPUs, add -forcegpu nvidia_g80 -gpu 0 instead.
Then, in cd ~/nvidia1/, add the same but using -gpu 1 instead, and so forth.

Press CTRL+X, then 'Y' to save, and ENTER.

Then use this final command:

Now, to add the Launcher itself, rightclick the top panel (the 'taskbar' along the top)> "Add to Panel"> "Custom Application Launcher".

• Change "Type" to "Application in Terminal".
• For "Name", enter "GPU Client".
• For "Command", enter:

Now, download this icon:


Click the spring-like icon, and replace it with the icon above:


Now, click the icon, and move on to the final steps.

8) Once you start the client, you will be asked for your info. Press "Enter" unless the answers are in bold:

• User name [Anonymous]? [Your Folding Name]
• Team Number [0]? 37726
• Passkey []?
• Ask before fetching/sending work <no/yes> [no]?
• Use proxy <yes/no> [no]?
• Acceptable size of work assignment and work result packets <bigger units may have large memory demands> -- 'small' is <5MB, 'normal' is <10MB, and 'big' is >10MB <small/normal/big> [normal]? big
• Change advanced options <yes/no> [no]? yes
• Core Priority <idle/low> [idle]? low
• Disable highly optimized assembly code <no/yes> [no]?
• Interval, in minutes, between checkpoints <3-30> [15]? 5
• Memory, in MB, to indicate <xxxx available> [xxxx]? 512
• Set -advmethods flag always, requesting new advanced scientific cores and/or work units if available <no/yes> [no]? yes
• Ignore any deadline information <mainly useful if system clock frequently has errors> <no/yes> [no]?
• Machine ID <1-16> [2]? Press "Enter" if this is the first GPU, 3 for the 2nd GPU, and so on.
• Launch automatically, install as a service in this directory <yes/no> [no]?
• *The following options require you to restart the client before they take effect
• Disable CPU affinity lock <no/yes> [no]?
• Additional client parameters []?
• IP address to bind core to <for viewer> []?

And a picture for reference:



Now move on to the next post to set up HFM to monitor the client.
Edited by zodac - 12/13/10 at 2:34pm

HFM Guide:

What you Need:

HFM.NET. The .zip file, under "Featured downloads".
(Direct Link)

Installation:

1) Since there isn't support for the framework HFM is based off in Linux, you will need to configure the OS first. Open Terminal (again), and enter the following command:

A text editor will open. At the bottom of this file, add these lines:


Save the file and close it.

2) Now, more commands to add:

Extract the .zip file you downloaded to a folder called HFM on the Desktop.

Now, go back to terminal and enter these lines:

HFM will open. For now, close it. We'll now make a launcher for it.

3) Create a file on the Desktop. Call it HFMLaunch:



Open the file, and type this in and save the file:
4) Open Terminal and use these commands:

5) Now, we make a launcher. Rightclick the top Panel> Add to Panel> Custom Application Launcher.

• Leave "Type" as "Application".
• For "Name", enter "HFM".
• For "Command", enter:
Again, download this icon:


Click the spring thing again, and replace the icon.



It should now look like this:


6) Now we can set HFM to monitor the GPU client. Go to Clients> Add Client.

• For "Instance Name", enter what you'd like to call the client.
• For "Client Processor MHz", enter the GPU shader clocks.
• For "Log Folder", enter:


It should look like this:



7) Finally, we need to save the configuration file. Go to File> Save Configuration As> Save the file as Config:



Then go to Edit> Preferences> Startup _External. Check the "Load Configuration File" box, and select the Config.hfm file you just saved.

You're finished!
Edited by zodac - 11/6/10 at 5:51pm

Not this version; needs to be updated with the GPU3 setup file, and the right CUDA toolkit...

Thank God; I thought someone was going to ask me to do it.