NVIDIA GPU3 Linux/Wine Headless Install Guide

Moderators: slegrand, Site Moderators, PandeGroup

Re: NVIDIA GPU2 Linux/WINE Headless Install Guide

Postby Hyperlife » Thu Nov 20, 2008 1:40 am

KE1HA wrote:
shatteredsilicon wrote:This is actually a waste of time. Loading the nvidia kernel module will create the device nodes anyway.

So once you modprobe nvidia, which could be done anytime during the init.d sequence, it creates the nodes in /dev ?

No, it doesn't work with modprobe. From a fresh reboot:

Code: Select all
$ ls /dev/nv*
ls: cannot access /dev/nv*: No such file or directory
$ sudo modprobe nvidia
$ ls /dev/nv*
ls: cannot access /dev/nv*: No such file or directory
$ sudo ./cudainit
$ ls /dev/nv*
/dev/nvidia0  /dev/nvidiactl

Starting X will create the nodes automatically, but my guide doesn't require X.
Last edited by Hyperlife on Thu Nov 20, 2008 2:31 am, edited 1 time in total.
Image
User avatar
Hyperlife
 
Posts: 438
Joined: Sun Dec 02, 2007 7:38 am

Re: NVIDIA GPU2 Linux/WINE Headless Install Guide

Postby shatteredsilicon » Thu Nov 20, 2008 2:00 am

Yes, "modprobe nvidia" will create the device nodes.
Grepping lspci for nvidia 3D and VGA devices is a good a way as any to find the GPUs. Do you really envisage ever being in a situation where you have a variable number of GPUs, though?
Image
1x Q6600 @ 3.2GHz, 4GB DDR3-1333
1x Phenom X4 9950 @ 2.6GHz, 4GB DDR2-1066
3x GeForce 9800GX2
1x GeForce 8800GT
CentOS 5 x86-64, WINE 1.x with CUDA wrappers
shatteredsilicon
 
Posts: 627
Joined: Tue Jul 08, 2008 2:27 pm

Re: NVIDIA GPU2 Linux/WINE Headless Install Guide

Postby Hyperlife » Thu Nov 20, 2008 2:26 am

shatteredsilicon wrote:Yes, "modprobe nvidia" will create the device nodes.

No. See the driver README file:

How and when are the the NVIDIA device files created?

Depending on the target system's configuration, the NVIDIA device files used to be created in one of three different ways:

at installation time, using mknod
at module load time, via devfs (Linux device file system)
at module load time, via hotplug/udev

With current NVIDIA driver releases, device files are created or modified by the X driver when the X server is started.
User avatar
Hyperlife
 
Posts: 438
Joined: Sun Dec 02, 2007 7:38 am

Re: NVIDIA GPU2 Linux/WINE Headless Install Guide

Postby Hyperlife » Thu Nov 20, 2008 2:30 am

KE1HA wrote:Question: Will the cudainit script count the number of cores or just physical PCI cards installed on a system, looks like just the PCI cards? I dont have a 9800 GX2 to test with.

I believe the script should work. Grepping lspci will locate the physical cards, and I think the device nodes are supposed to represent physical cards (not individual cores), but using the -gpu and -forcegpu flags should detect the second core.
User avatar
Hyperlife
 
Posts: 438
Joined: Sun Dec 02, 2007 7:38 am

Re: NVIDIA GPU2 Linux/WINE Headless Install Guide

Postby shatteredsilicon » Thu Nov 20, 2008 2:47 am

OK, my mis-explanation re: device auto-createion. I use a udev enabled distro so I take it for granted.
You get a device node per GPU, not per card. I get 6 device nodes for 3 GX2 cards.
shatteredsilicon
 
Posts: 627
Joined: Tue Jul 08, 2008 2:27 pm

Re: NVIDIA GPU2 Linux/WINE Headless Install Guide

Postby KE1HA » Thu Nov 20, 2008 3:12 am

shatteredsilicon wrote:Yes, "modprobe nvidia" will create the device nodes.
Grepping lspci for nvidia 3D and VGA devices is a good a way as any to find the GPUs. Do you really envisage ever being in a situation where you have a variable number of GPUs, though?

Thanks guys.

The reason I'm looking to identify the number of GPU's v.s. the number of GFX cards is for another purpose. I want to use a session manager to log into each GPU that's folding via dtach. The problem comes in when I have 3 physical cards installed, say they are 9800 GT's (Single Core) which would mean 3x Cores and 3x Sessions. Now I switch over to 3x 9800 GX2 cards, each have 2x cores. I know the -gpu 0, 1, 2, 3, 4 will work for starting the clients, but I need to create the session for each of the core, so If Im scripting it, I need a way to determine the number of physical cores present v.s the number of physical GFX cards.

Maybe they will create an MPI / SMP version of the GPU client and solve all my problems :P ... 82 second would be come 2 / 4 / 6 seconds, then we run into Network and File System I/O binding issues :mrgreen:
.
Image
2x - Q6600 1 with 2x 9800GT's / 1 with 2x 8800GTS G92 512's UB-8.04 CLI Only
2x - AMD 9850 1 with 3x 9800 GT's / 1 with 1x 9800 GT's UB-8.04 CLI Only
1x - QX9770 GTX-280 WinDoze Vista 64
User avatar
KE1HA
 
Posts: 73
Joined: Fri Dec 21, 2007 12:03 am
Location: San Jose, CA

Re: NVIDIA GPU2 Linux/WINE Headless Install Guide

Postby jcfuller » Thu Nov 20, 2008 5:46 pm

Hyperlife,
How much disk space does the complete install take?
Would it be possible to fit it on a Usb Flash Drive?
I had Ubu64 cli desktop installed on a 4gig Corsair Flash Voyager last year, before I took folding break, but I assumed the server version with ssh and somba would not fit so I used the installed hd.


James
jcfuller
 
Posts: 28
Joined: Thu Oct 30, 2008 11:22 am
Location: Fort Edward, NY

Re: NVIDIA GPU2 Linux/WINE Headless Install Guide

Postby Hyperlife » Thu Nov 20, 2008 5:54 pm

The total install is a little over 1GB using the Server version. It should easily fit on a 4GB flash drive (in fact, I tested this on an old 4GB hard drive I had lying around).

I'm now running this system headless and diskless. It netboots from my FreeBSD work server using PXE and NFS.

I don't have samba installed, but it wouldn't take up much additional room if you add it.
User avatar
Hyperlife
 
Posts: 438
Joined: Sun Dec 02, 2007 7:38 am

Re: NVIDIA GPU2 Linux/WINE Headless Install Guide

Postby KE1HA » Fri Nov 21, 2008 7:36 pm

Here's my disk space usage.

UB 8.04 Sever (CLI) Install: Base + the following packages:

openssh-server
samba
ndiswrapper
ntp
dtach
Nvidia Driver
Cuda Tool-Kit

Code: Select all
Filesystem            Size  Used Avail Use% Mounted on
/dev/sda3             9.3G  1.2G  7.7G  13% /
varrun                502M  212K  502M   1% /var/run
varlock               502M     0  502M   0% /var/lock
udev                  502M   28K  502M   1% /dev
devshm                502M     0  502M   0% /dev/shm
/dev/sda1              93M   24M   65M  27% /boot
User avatar
KE1HA
 
Posts: 73
Joined: Fri Dec 21, 2007 12:03 am
Location: San Jose, CA

Re: NVIDIA GPU2 Linux/WINE Headless Install Guide

Postby jcfuller » Sat Nov 22, 2008 1:32 am

Well that didn't work very well. Grub just hangs
I don't have time to research it further. The heat is actually quite nice with it below freezing outside. Maybe I can figure it out by summer :D

James
jcfuller
 
Posts: 28
Joined: Thu Oct 30, 2008 11:22 am
Location: Fort Edward, NY

Headless Install, GPU2 client hosed, can't get it going agai

Postby tmoble » Tue Dec 02, 2008 7:19 am

Rats, this was working fine for a couple weeks, then the GPU2 client quit, now it won't start. I've even rebooted the machine, still no good. here's what I'm getting, anybody know what this means?

Code: Select all
Could not release constant ring memory.
Could not free constant ring memory.
Could not release constant ring memory.
Could not free constant ring memory.
Could not release constant ring memory.
Could not free constant ring memory.
Could not release constant ring memory.
Could not free constant ring memory.
Could not destroy cal context.
Could not close cal device.
Unable to initialize CAL runtime, aborting
[07:04:35] CoreStatus = FFFFFFF6 (-10)
[07:04:35] Client-core communications error: ERROR 0xfffffff6
[07:04:35] This is a sign of more serious problems, shutting down.
Application tried to create a window, but no driver could be loaded.
Make sure that your X server is running and that $DISPLAY is set correctly.
tom@ubunfold:~/fahgpu2$
tom@ubunfold:~/fahgpu2$



Now, as I understand it we don't start X on this deal. I did run cudainit as root. I checked to see the 2 nvidia entries are in /dev, they are present with the correct time stamp for when I ran cudainit after the reboot:

tom@ubunfold:~/fahgpu2$ cd /dev
tom@ubunfold:/dev$ ls -la | grep nvidia
crw-rw-rw- 1 root root 195, 0 2008-12-02 00:01 nvidia0
crw-rw-rw- 1 root root 195, 255 2008-12-02 00:01 nvidiactl
tom@ubunfold:/dev$


starting from the fahgpu2 directory with this command line, which should look familiar:

nice -19 wine Folding@home-Win32-GPU.exe -verbosity 9 -forcegpu nvidia_g80

What am I doing wrong?
tmoble
 
Posts: 118
Joined: Mon Dec 03, 2007 7:19 am
Location: Glendale, Az

Re: NVIDIA GPU2 Linux/WINE Headless Install Guide

Postby uncle_fungus » Tue Dec 02, 2008 11:26 am

Trash your queue and the work folder and try again. The errors you're getting seem to indicate that the client is trying to launch the ATI core (CAL) rather than the NVIDIA core. The rest of the log would help confirm this.
User avatar
uncle_fungus
Site Admin
 
Posts: 1682
Joined: Fri Nov 30, 2007 9:37 am
Location: Oxfordshire, UK

Re: NVIDIA GPU2 Linux/WINE Headless Install Guide

Postby codysluder » Tue Dec 02, 2008 10:20 pm

KE1HA wrote:
shatteredsilicon wrote:The reason I'm looking to identify the number of GPU's v.s. the number of GFX cards is for another purpose. I want to use a session manager to log into each GPU that's folding via dtach. The problem comes in when I have 3 physical cards installed, say they are 9800 GT's (Single Core) which would mean 3x Cores and 3x Sessions. Now I switch over to 3x 9800 GX2 cards, each have 2x cores. I know the -gpu 0, 1, 2, 3, 4 will work for starting the clients, but I need to create the session for each of the core, so If Im scripting it, I need a way to determine the number of physical cores present v.s the number of physical GFX cards..


From an old discussion, Windows counts cards, not gpus. I'm not sure how to use that information in Linux/Wine, but you might.
codysluder
 
Posts: 2128
Joined: Sun Dec 02, 2007 12:43 pm

Re: NVIDIA GPU2 Linux/WINE Headless Install Guide

Postby shatteredsilicon » Tue Dec 02, 2008 10:27 pm

It seems irrelevant. CUDA counts "CUDA resources", which are discrete GPUs available. If you have a GX2, it'll have access to devices 0,1. If you have 2 GX2s, you'll have devices 0-3 available. With 3 GX2s, 0-5, etc. I know this because I was running 3 GX2s in one system until last week.
shatteredsilicon
 
Posts: 627
Joined: Tue Jul 08, 2008 2:27 pm

Re: NVIDIA GPU2 Linux/WINE Headless Install Guide

Postby tmoble » Wed Dec 03, 2008 6:54 am

Well, after deleting the work folder and other relevant files3 times it decided to be fixed. Whatta PITA. WHen I delete the files here's what I have left:

tom@ubunfold:~/fahgpu2$ ls -la
total 4104
drwxrwxrwx 2 tom tom 4096 2008-12-02 12:12 .
drwxrwxrwx 5 tom tom 4096 2008-11-13 23:41 ..
-rw-r--r-- 1 tom tom 1380352 2008-07-01 16:44 amdcalcl.dll
-rw-r--r-- 1 tom tom 2170880 2008-07-01 16:44 amdcalrt.dll
-rw-r--r-- 1 tom tom 188416 2008-07-01 16:44 cudart.dll
-rw-r--r-- 1 tom tom 435200 2008-08-16 06:42 Folding@home-Win32-GPU.exe
tom@ubunfold:~/fahgpu2$

uncle, I saw that you posted about the cal error having to do with it trying to run the ATI files on an nvidia card, but those two amd* files have cal in the name too. Since it's working now without changing anything but re-downloading I'm going to figure it had something to do with the amdcal* files. The nvidia force argument is in the command line anyway, hard to see how it could trying to start the ATI client.
tmoble
 
Posts: 118
Joined: Mon Dec 03, 2007 7:19 am
Location: Glendale, Az

PreviousNext

Return to unOfficial Linux GPU (WINE wrapper) (3rd party support)

Who is online

Users browsing this forum: No registered users and 1 guest

cron