GTX 770 not detected - FAHClient 7.3.6

[Nicolas_orleans]

Hello

I built a new dedicated folding rig with 2x GTX770 running under Ubuntu 13.04 with NVIDIA drivers 319.32, in order to GPU fold with Core 17 and client-type=advanced.

FAHClient reports

Code: Select all

GPUs 0
CUDA 3.0
CUDA Driver 5050

FAHClient --lspci reports

Code: Select all

0x10de:0x1184: NVIDIA Corporation:
0x10de:0x1184: NVIDIA Corporation:

(and two similar entries for audio)

This appears similar to the update of June 5th in the whitelist
0x10de:0x1184:2:3:GK104 [GeForce GTX 770]

I know the whitelist is supposed to auto update in 7.3.6, however, in which linux directory can I find GPUs.txt to check if I have the correct version of the whitelist ?

Thanks

***

Full FAH log

Code: Select all

********************** Log Started 2013-07-16T21:36:28Z ***********************
21:36:28:************************* Folding@home Client *************************
21:36:28:    Website: http://folding.stanford.edu/
21:36:28:  Copyright: (c) 2009-2013 Stanford University
21:36:28:     Author: Joseph Coffland <[email protected]>
21:36:28:       Args: --child --lifeline 3410 /etc/fahclient/config.xml --run-as
21:36:28:             fahclient --pid-file=/var/run/fahclient.pid --daemon
21:36:28:     Config: /etc/fahclient/config.xml
21:36:28:******************************** Build ********************************
21:36:28:    Version: 7.3.6
21:36:28:       Date: Feb 18 2013
21:36:28:       Time: 07:24:08
21:36:28:    SVN Rev: 3923
21:36:28:     Branch: fah/trunk/client
21:36:28:   Compiler: GNU 4.4.7
21:36:28:    Options: -std=gnu++98 -O3 -funroll-loops -mfpmath=sse -ffast-math
21:36:28:             -fno-unsafe-math-optimizations -msse2
21:36:28:   Platform: linux2 3.2.0-1-amd64
21:36:28:       Bits: 64
21:36:28:       Mode: Release
21:36:28:******************************* System ********************************
21:36:28:        CPU: Intel(R) Celeron(R) CPU G1610 @ 2.60GHz
21:36:28:     CPU ID: GenuineIntel Family 6 Model 58 Stepping 9
21:36:28:       CPUs: 2
21:36:28:     Memory: 3.81GiB
21:36:28:Free Memory: 2.66GiB
21:36:28:    Threads: POSIX_THREADS
21:36:28:Has Battery: false
21:36:28: On Battery: false
21:36:28: UTC offset: 2
21:36:28:        PID: 3417
21:36:28:        CWD: /var/lib/fahclient
21:36:28:         OS: Linux 3.8.0-26-generic x86_64
21:36:28:    OS Arch: AMD64
21:36:28:       GPUs: 0
21:36:28:       CUDA: 3.0
21:36:28:CUDA Driver: 5050
21:36:28:***********************************************************************

Full FAH --lspci

Code: Select all

~$ FAHClient --lspci
VendorID:DeviceID:Vendor Name:Description
0x8086:0x0150:Intel Corporation:
0x8086:0x0151:Intel Corporation:
0x8086:0x0155:Intel Corporation:
0x8086:0x1e31:Intel Corporation:
0x8086:0x1e3a:Intel Corporation:
0x8086:0x1503:Intel Corporation:
0x8086:0x1e2d:Intel Corporation:
0x8086:0x1e20:Intel Corporation:
0x8086:0x1e26:Intel Corporation:
0x8086:0x1e44:Intel Corporation:
0x8086:0x1e02:Intel Corporation:
0x8086:0x1e22:Intel Corporation:
0x10de:0x1184:NVIDIA Corporation:
0x10de:0x0e0a:NVIDIA Corporation:
0x10de:0x1184:NVIDIA Corporation:
0x10de:0x0e0a:NVIDIA Corporation:

[bollix47]

I have seen a few reports that GPUs.txt does not download in Linux.

If you installed without changing anything that would change the default location then look in /var/lib/fahclient.

Code: Select all

21:36:28:        CWD: /var/lib/fahclient

I have found that one easy way to find a file in Linux is open a terminal and type the following commands:

Code: Select all

sudo updatedb
locate GPUs.txt

Now here's a bit of strange info ... I have a GTX 670 running on Ubuntu 12.04 and it's recognized and working fine. However, I cannot locate GPUs.txt anywhere on that file system and it's certainly not in the folding data directory.

I also have a GTX 660ti running on Ubuntu 13.04 and there is a GPUs.txt file in the folding data directory which in my case is ~/fah where I manually start the client running.

[Nicolas_orleans]

Bollix

The locate GPUs.txt returns nothing, such as my searches in the 3 folding directories I know are used by the client :
/var/lib/fahclient
/etc/fahclient
/usr/bin

So I believe in my install such as one of yours... there is no GPUs.txt

How can I update the whitelist ???

[bollix47]

viewtopic.php?p=212029#p212029

You can download it from the above link. It will open a page in your browser which you save as GPUs.txt and then copy it to /var/lib/fahclient using sudo. Restart the client.

One other possibility is your drivers. Does nvidia-settings work and show the correct drivers? I don't see 319 in the regular Ubuntu repositories. Did you install them manually or use another source like xorg edgers? If so you could try 325.

[Nicolas_orleans]

Thanks bollix, the manual download to the /var/lib/fahclient followed by init.d stop and start scripts worked.

First two Wus downloaded, folding at 78-80 degrees celsius ! 319.32 were downloaded from xorg edgers and appear to work fine with my 3.8 kernel.

Do you know if there is an opened ticket to fix the GPUs.txt download at install and autoupdate feature for linux folks ?

Nicolas

[7im]

There is no ticket like that. What version of of fah software did you download and from which source?

[HaloJones]

How are the points per day totals working out so far?

[bollix47]

I see the following message in the log when I restart the client on one of my GPU Linux installations running on Ubuntu 12.04:

02:49:10:WARNING:Attempting to update GPUs.txt: Failed to connect to assign-GPU.stanford.edu Connection refused

I don't see it on a couple of others using 13.04 but Nicolas is using 13.04 so I doubt the version has anything to do with getting the above message.

All of mine are running 7.3.6 and are setup to run folding from my home directory in a sub-directory called fah. Since Nicolas is using the same version of the client and used a default service install no clues as to why it happens there either.

Others have reported seeing the same message in Linux and their only solution has been to download the file manually. Some, including me, have reported they found no GPUs.txt on their Linux systems but that GPU folding was working fine.

Although it was setup to allow GPU folding it could be that the client needs a bit of fine-tuning when it comes to Linux.

[7im]

I recently had the same problem on a Windows Vista computer. FAHClient didn't download the GPUs.txt file, so my GPU was not recognized. I had to download the GPUs.txt file manually. I assumed it was because the system hadn't had any Windows Updates in 4 years and was out of date. (PC was in storage.)

When I formatted the drive and installed Win 7, and then installed FAH, it worked just fine. Very difficult to recreate the problem, so will be difficult to find the solution. But, yes, we'll need a bug ticket on this problem if there isn't one already.

[Joe_H]

There is Ticket 1035 which appears related to this problem. I did not see another ticket on a quick search that was exactly the same as reported here, but solving whatever is causing the issue described in #1035 could solve both issues.

[Nicolas_orleans]

There is no ticket like that. What version of of fah software did you download and from which source?

fahclient_7.3.6_amd64.deb and fahcontrol_7.3.6-1_all.deb, downloaded from http://folding.stanford.edu/ then "start folding"
As stated by bollix, default service install for my setup

Although it was setup to allow GPU folding it could be that the client needs a bit of fine-tuning when it comes to Linux.

I completely agree, some fine-tuning may be required to offer the same feature list among OSes, especially if they could help new donors to fold flawlessly after install

How are the points per day totals working out so far?

Slightly off topic (more FAH Hardware subsection), but after having finished almost 4 work units, the system (2 cards) generates around 180-190k PPD on 8900s, that's 90-95k per card. Based on ambiant temperature in my home I see the turbo boost 2.0 being limited by the 80°C limit built in the Nvidia drivers, so I would say it's not the full turbo potential.

[bollix47]

Nicolas, here is a step-by-step way to adjust your fan speed in Linux and bring those temperatures down:

NB: I've used this method on single card installations only so you may need to adjust one of more steps to account for multiple cards.

See 2/ in this post for an alternate method to the following(down to ****):

Open nvidia-settings.
Select "X Server Display Configuration" from the options on the left.
Click on "Save to X Configuration File".
The save screen should show /etc/X11/xorg.conf as the file to merge with.
Click on Save.
Quit nvidia-settings.

Open a terminal and type(replace gedit with you favorite text editor):

Code: Select all

gksudo gedit /etc/X11/xorg.conf

Scroll down to Section "Screen"
Within that Section and before EndSection add the following:

Code: Select all

Option    "Coolbits"  "4"

Save and quit.
You need to pause folding at this point and reboot in order for the setting to take effect. (logoff and logon may also work but I prefer reboot)

****

After your o/s comes back up you again start nvidia-settings and this time select Thermal Settings.

You should now see a box next to "Enable GPU Fan Setting". Check that box and change the fan speed to your new setting ... press enter and click on Apply. (You can also use the slider in which case pressing enter is not necessary)

You might need to quit and restart nvidia-settings before it will properly show the new speed. Assuming the new speed is higher than it was before performing the above your temperature should be lower.

If you find your new fan setting does not survive a reboot of your system you can do the following:
Create a file in your home directory called gpu_fan.sh and add the following to it:

Code: Select all

#!/bin/bash
nvidia-settings -a [gpu:0]/GPUFanControlState=1
nvidia-settings -a [fan:0]/GPUCurrentFanSpeed=70

The above is for one GPU so you would add two more lines for gpu:1 and fan:1 ... and may need to adjust the Speed from 70 or create two separate files ... one for each GPU and add both to Startup Applications.

Now you need to find Startup Applications and add gpu_fan.sh so that it runs when Linux starts. Use the full path in the command line:
/home/yourusername/gpu_fan.sh

You may also need to make gpu_fan.sh executable although I'm not sure if it's necessary.
Open a terminal and type:
chmod +x gpu_fan.sh

Here's a related link but I have changed quite a few steps:
http://ubuntuforums.org/showthread.php?t=2114764

[7im]

I talked with one of the developers on this topic already. Both the Windows and Linux client are configured to download the GPUs.txt file when the installer detects a GPU. And detection of the GPU depends on having installed a recent version of the driver from the OEM (AMD or NV). Open source drivers in Linux are not supported at this time. Plus, the reports on this issue are very few. If the installer was broken, we'd get a lot more reports.

That said, there does seem like another rare issue that seems to popup now and then.

[Nicolas_orleans]

Thanks bollix, this is some very interesting information, will get a closer look

[Quisarious]

I talked with one of the developers on this topic already. Both the Windows and Linux client are configured to download the GPUs.txt file when the installer detects a GPU. And detection of the GPU depends on having installed a recent version of the driver from the OEM (AMD or NV). Open source drivers in Linux are not supported at this time. Plus, the reports on this issue are very few. If the installer was broken, we'd get a lot more reports.

That said, there does seem like another rare issue that seems to popup now and then.

I've set up 4 machines to GPU fold so far, and not one has automatically downloaded the GPUs.txt file. In each case I had proprietary drivers loaded (3xNvidia, 1xAMD). It's not a big deal, but it's also not a rare issue. Linux GPU folding is rare. I know several other people that have had to manually download also. I suspect Linux users are more used to manually setting things up, so you don't hear about problems as much.