V7 Linux GPU with Wine

[SandStar]

Running with wine works a bit different. Besides being unsupported I've not managed to get it properly working.
It failed to shutdown the core properly and I've only seen it work with core_11 and not 15. I do believe that it can be done though. But since I'm no expert on this I've stopped researching how to fix this. Maybe somebody else can get it to work
Here's how to proceed:

First make sure your pc can run V6 GPU work in wine
V7 will not work if you can't get it to work with V6. For information how to run V6 see this subforum: viewforum.php?f=54

The big difference is that with V7 you don't need the windows client.
I'm running it on my Gentoo system with the 64bit linux client. If that client is told it's "win32" it will act as a windows client and download cores for it.

Code: Select all

  <client-subtype value='STDCLI'/>
  <os-type v='WIN32'/>


It however does require that wine is installed and configured to auto-run windows executables.
The latter is done by "binfmt" and works out of the box on Debian and Ubuntu.

Gentoo (and possible others) require that Executable file formats -> Kernel support for MISC binaries is enabled.
I then added the following line to /etc/conf.d/local.start

Code: Select all

echo ':DOSWin:M::MZ::/usr/bin/wine:' > /proc/sys/fs/binfmt_misc/register

binfmt_misc needs to be mounted for this to work. On my system that was done automatically but here's the line for your fstab if it didn't:

Code: Select all

none /proc/sys/fs/binfmt_misc binfmt_misc defaults 0 0

The result of this is a Linux x86-64 client that acts as a windows client.
I'm only running a GPU slot but a UNIPROCESSOR slot also worked and I think SMP also works fine.

Code: Select all

14:39:40:************************* Folding@home Client *************************
14:39:40:   Website: http://folding.stanford.edu/
14:39:40: Copyright: (c) 2009,2010 Stanford University
14:39:40:    Author: Joseph Coffland <joseph@cauldrondevelopment.com>
14:39:40:      Args:
14:39:40:    Config: /home/erwin/folding-gpu/config.xml
14:39:40:******************************** Build ********************************
14:39:40:   Version: 7.1.4
14:39:40:      Date: Oct 23 2010
14:39:40:      Time: 22:23:20
14:39:40:   SVN Rev: 2383
14:39:40:    Branch: fah/trunk/client
14:39:40:  Compiler: Intel(R) C++ g++ 4.3 mode 1110
14:39:40:   Options: -std=gnu++98 -O3 -funroll-loops -xsse2 -axSSE3,SSSE3,SSE4.1,SSE4.2
14:39:40:            -restrict
14:39:40:  Platform: linux2 2.6.28-11-server
14:39:40:      Bits: 64
14:39:40:      Mode: Release
14:39:40:******************************* System ********************************
14:39:40:        OS: Linux 2.6.34-gentoo-r12 x86_64
14:39:40:    Memory: 3.87 GB
14:39:40:       CPU: Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz
14:39:40:    CPU ID: GenuineIntel Family 6 Model 15 Stepping 11
14:39:40:      CPUs: 1 Physical, 4 Logical, 4 Core(s), 1 Thread(s) per Core
14:39:40:   Threads: POSIX_THREADS
14:39:40:      GPUs: 1
14:39:40:     GPU 0: NVIDIA_G94 GeForce 9600 GT
14:39:40:On Battery: false
14:39:40:UTC offset: 1
14:39:40:***********************************************************************
14:39:40:Enabled computation slot 00: READY gpu:0:NVIDIA_G94
14:39:41:News: Welcome to Folding@Home
14:39:41:Assigned to work server 171.64.65.71
14:39:41:Requesting new work unit for slot 00: READY gpu:0:NVIDIA_G94 from 171.64.65.71
14:39:42:Downloading 80.55KiB............................................................................done
14:39:43:Received Unit: ID:00 State:DOWNLOAD Project:10112 Run:777 Clone:0 Gen:77 Core:0x11 Unit:0x0000005d0a3b1e6b4ccf4feeb2b31b32
14:39:43:Downloading core from http://www.stanford.edu/~pande/Win32/AMD64/NVIDIA/G80/Core_11.fah
<snip>
14:39:48:Starting Unit: 00
14:39:48:Core 0x11 started
14:39:56:Unit 00:
14:39:56:Unit 00:*------------------------------*
14:39:56:Unit 00:Folding@Home GPU Core
14:39:56:Unit 00:Version 1.31 (Tue Sep 15 10:57:42 PDT 2009)
14:39:56:Unit 00:
14:39:56:Unit 00:Compiler  : Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 14.00.50727.762 for 80x86
14:39:56:Unit 00:Build host: amoeba
14:39:56:Unit 00:Board Type: Nvidia
14:39:56:Unit 00:Core      :
14:39:56:Unit 00:Preparing to commence simulation
14:39:56:Unit 00:- Looking at optimizations...
14:39:56:Unit 00:DeleteFrameFiles: successfully deleted file=00/wudata_01.ckp
14:39:56:Unit 00:- Created dyn
14:39:56:Unit 00:- Files status OK
14:39:56:Unit 00:- Expanded 81971 -> 425170 (decompressed 518.6 percent)
14:39:56:Unit 00:Called DecompressByteArray: compressed_data_size=81971 data_size=425170, decompressed_data_size=425170 diff=0
14:39:56:Unit 00:- Digital signature verified
14:39:56:Unit 00:
14:39:56:Unit 00:Project: 10112 (Run 777, Clone 0, Gen 77)
14:39:56:Unit 00:
14:39:56:Unit 00:Assembly optimizations on if available.
14:39:56:Unit 00:Entering M.D.
14:39:56:Unit 00:Tpr hash 00/wudata_01.tpr:  2221668042 1760085118 1497456037 4136928347 3505360100
14:39:56:Unit 00:
14:39:56:Unit 00:Calling fah_main args: 14 usage=100
14:39:56:Unit 00:
14:39:56:Unit 00:Working on 1174 p10112_ubiquitin_300K
14:40:00:Unit 00:Client config unavailable.
14:40:00:Unit 00:Starting GUI Server
14:42:45:Unit 00:Completed 1%
14:45:33:Unit 00:Completed 2%
14:48:17:Unit 00:Completed 3%
14:51:05:Unit 00:Completed 4%


It however failed to shutdown the core properly and I've only seen it work with core_11 and not 15.
Since I'm no expert on this I've stopped researching how to fix this. Maybe somebody else can get it to work

[johnT89]

I added those options in the gpu's slot's 'extra parameters'. Playing around to test it, Ubuntu 64bit and 8600gt.

EDIT:
- Can't find /etc/conf.d/local.start

[SandStar]

I added those options in the gpu's slot's 'extra parameters'. Playing around to test it, Ubuntu 64bit and 8600gt.

EDIT:
- Can't find /etc/conf.d/local.start

I'm not getting your point. What's with local.start ?

[johnT89]

This file do not exist on my Ubuntu 10.10 machine. I have to create it by myself I guess.
I will report back here in a couple of days, when I will have a Fermi card to play with.

[MtM]

Sandstar the quote in the opening post, does it have a source thread of did you put it in quotes hoping it would be more clear they were instructions?

Also, if I would follow that quote, I'm kind of missing needing to install the nvidia sdk, or don't you need that to run gpu code on linux?

It's abit vague

[SandStar]

This file do not exist on my Ubuntu 10.10 machine. I have to create it by myself I guess.
I will report back here in a couple of days, when I will have a Fermi card to play with.

Like I said:

The latter is done by "binfmt" and works out of the box on Debian and Ubuntu.

You should not have to do that on Debian and Ubuntu

[SandStar]

Sandstar the quote in the opening post, does it have a source thread of did you put it in quotes hoping it would be more clear they were instructions?

Also, if I would follow that quote, I'm kind of missing needing to install the nvidia sdk, or don't you need that to run gpu code on linux?

It's abit vague

The source thread is exactly the same. It's just another thread started by me talking about my findings.
As far as I know, I'm the only one that ever got it to run a bit.

I'll try to update my first post tonight.
Make sure your system is capable of doing V6 GPU work. Once that works try V7

[MtM]

I'm not going to right now, going to work on the class creation and json parsing of fahclient, when I can get both my ide and the gpu client to run at the same time without having to wait 30 seconds before my mouse reacts to anything in the ide

If you're going to make this a thread about experimenting with V7 and wine, the last sentence in particular should be in the first post ( I'm easily confused, quicker then others, if info is incomplete so take what I say with that in mind ). I'm only just checking most recent threads ect and I'm a happy camper since I like the way the forum looks right now, so don't take my comment as being critical, I am most definitely not

Still wouldn't hurt to have a link to the old thread

[SandStar]

Actually I'm very happy with your comments. thanks.

A link to the old thread would not help. It was exactly the same with me as the sole contributor.
I just did a little bit of digging and then put it together in 1 post hoping that someone with more knowledge would be able to take it from there.

[Sidicas]

It however failed to shutdown the core properly and I've only seen it work with core_11 and not 15.

Could you be more specific? Is it failing at 100%? Do you have a log for that?
Are you getting any stubs or other Wine debug information to the console? Have you tried running the Windows client on WINE?

I haven't had time yet to try it.

[SandStar]

It however failed to shutdown the core properly and I've only seen it work with core_11 and not 15.

Could you be more specific? Is it failing at 100%? Do you have a log for that?

It failed at terminating the core properly. The unit itself was completed successfully as far as I can see.
Don't have the logs anymore.
Only core_11 worked. I've never been able to get core_15 to work in either V6 or V7

Are you getting any stubs or other Wine debug information to the console?

Yes but nothing I've been able to understand

Have you tried running the Windows client on WINE?

Yes that won't work.

[Sidicas]

It failed at terminating the core properly. The unit itself was completed successfully as far as I can see.
Don't have the logs anymore.
Only core_11 worked. I've never been able to get core_15 to work in either V6 or V7

Hmmm... Core_15 should work... Are you using the GPU3 wrapper, CUDA 3.0 and some reasonably recent proprietary nvidia drivers along with a stable version of wine (1.2.2 or earlier)?
Check out these guides and see if you can get Core 15 working.
http://foldingforum.org/viewtopic.php?f=54&t=6793
http://linuxfah.info/index.php?title=Main_Page

Are you getting any stubs or other Wine debug information to the console?

Yes but nothing I've been able to understand

Can you post them?... Stubs means that some library mappings are incomplete.. That's pretty much expected as the wrapper only connects what's needed and a new client uses more libraries.. Though of course, make sure you're using the GPU3 wrapper as noted above.

[johnT89]

This is what I get.

Code: Select all

08:32:02:Slot 01 paused
08:32:04:Slot 01 unpaused
08:32:05:Connecting to assign-GPU.stanford.edu:80
08:32:05:News: Welcome to Folding@Home
08:32:05:Assigned to work server 171.67.108.11
08:32:05:Requesting new work unit for slot 01: READY gpu:0:"G84 [GeForce 8600 GT]" from 171.67.108.11
08:32:05:Connecting to 171.67.108.11:8080
08:32:07:Slot 01: Downloading 46.04KiB
08:32:08:Slot 01: Download complete
08:32:08:Received Unit: id:00 state:DOWNLOAD project:5768 run:7 clone:94 gen:1151 core:0x11 unit:0x00a1deeb4d9d7688047f005e00071688
08:32:08:Starting Unit 00
08:32:08:Running core: /var/lib/fahclient/cores/www.stanford.edu/~pande/Win32/AMD64/NVIDIA/G80/Core_11.fah/FahCore_11 -dir 00 -suffix 01 -lifeline 32486 -version 701 -checkpoint 15 -gpu 0
08:32:08:Started core on PID 17560
08:32:08:FahCore 0x11 started
08:32:16:FahCore running Unit 00 returned: FILE_IO_ERROR (117)
08:32:16:WARNING: Unit 00 Fatal error, dumping
08:32:16:Sending unit results: id:00 state:SEND project:5768 run:7 clone:94 gen:1151 core:0x11 unit:0x00a1deeb4d9d7688047f005e00071688
08:32:16:WARNING: Work server too old for dump report
08:32:16:Cleaning up Unit 00
08:32:16:Connecting to assign-GPU.stanford.edu:80
08:32:17:News: Welcome to Folding@Home
08:32:17:Assigned to work server 171.67.108.11
08:32:17:Requesting new work unit for slot 01: READY gpu:0:"G84 [GeForce 8600 GT]" from 171.67.108.11
08:32:17:Connecting to 171.67.108.11:8080
08:32:18:Slot 01: Downloading 44.81KiB
08:32:19:Slot 01: Download complete
08:32:19:Received Unit: id:00 state:DOWNLOAD project:5769 run:4 clone:8 gen:2027 core:0x11 unit:0x64fcb9754d9d769407eb000800041689
08:32:19:Starting Unit 00
08:32:19:Running core: /var/lib/fahclient/cores/www.stanford.edu/~pande/Win32/AMD64/NVIDIA/G80/Core_11.fah/FahCore_11 -dir 00 -suffix 01 -lifeline 32486 -version 701 -checkpoint 15 -gpu 0
08:32:19:Started core on PID 17584
08:32:19:FahCore 0x11 started
08:32:27:FahCore running Unit 00 returned: FILE_IO_ERROR (117)
08:32:27:WARNING: Unit 00 Fatal error, dumping
08:32:27:Sending unit results: id:00 state:SEND project:5769 run:4 clone:8 gen:2027 core:0x11 unit:0x64fcb9754d9d769407eb000800041689
08:32:27:WARNING: Work server too old for dump report
08:32:27:Cleaning up Unit 00
08:32:27:Connecting to assign-GPU.stanford.edu:80
08:32:28:News: Welcome to Folding@Home
08:32:28:Assigned to work server 171.67.108.11
08:32:28:Requesting new work unit for slot 01: READY gpu:0:"G84 [GeForce 8600 GT]" from 171.67.108.11
08:32:28:Connecting to 171.67.108.11:8080
08:32:28:Slot 01: Downloading 44.85KiB
08:32:29:Slot 01: Download complete
08:32:29:Received Unit: id:00 state:DOWNLOAD project:5771 run:2 clone:210 gen:1292 core:0x11 unit:0x3d18dcd44d9d769e050c00d20002168b
08:32:29:Starting Unit 00
08:32:29:Running core: /var/lib/fahclient/cores/www.stanford.edu/~pande/Win32/AMD64/NVIDIA/G80/Core_11.fah/FahCore_11 -dir 00 -suffix 01 -lifeline 32486 -version 701 -checkpoint 15 -gpu 0
08:32:29:Started core on PID 17606
08:32:29:FahCore 0x11 started
08:32:30:FahCore running Unit 00 returned: FILE_IO_ERROR (117)
08:32:30:WARNING: Unit 00 Fatal error, dumping
08:32:30:Sending unit results: id:00 state:SEND project:5771 run:2 clone:210 gen:1292 core:0x11 unit:0x3d18dcd44d9d769e050c00d20002168b
08:32:30:WARNING: Work server too old for dump report
08:32:30:Cleaning up Unit 00
08:32:30:Connecting to assign-GPU.stanford.edu:80
08:32:31:News: Welcome to Folding@Home
08:32:31:Assigned to work server 171.67.108.11
08:32:31:Requesting new work unit for slot 01: READY gpu:0:"G84 [GeForce 8600 GT]" from 171.67.108.11
08:32:31:Connecting to 171.67.108.11:8080
08:32:32:Slot 01: Downloading 44.84KiB
08:32:33:Slot 01: Download complete
08:32:33:Received Unit: id:00 state:DOWNLOAD project:5772 run:2 clone:324 gen:1487 core:0x11 unit:0x17d0ea0e4d9d76a105cf01440002168c
08:32:33:Starting Unit 00
08:32:33:Running core: /var/lib/fahclient/cores/www.stanford.edu/~pande/Win32/AMD64/NVIDIA/G80/Core_11.fah/FahCore_11 -dir 00 -suffix 01 -lifeline 32486 -version 701 -checkpoint 15 -gpu 0
08:32:33:Started core on PID 17620
08:32:33:FahCore 0x11 started
08:32:34:Slot 01 paused
08:32:34:Slot 01: shutting core down
08:32:41:FahCore running Unit 00 exited

FILE_IO_ERROR (117) when the client tries to start the core

[Sidicas]

Sounds like a permissions issue on the folder where it saves workunits....

chmod -R u=+rwx <Folding@Home FolderName>


Also, Make sure the files and folders are all owned by the user that's running it.

[johnT89]

Did that, the problem is still here. Note that SMP client runs ok and the permissions are the same for all the main /var/lib/fahclient subfolders..

Might it be because it tries to run GPU client as a service? (https://fah-web.stanford.edu/projects/F ... ticket/594)

EDIT: I am not sure it applies to linux too