Redhat Enterprise 4 drivers for NVIDIA quadro FX 570
 

I'm having trouble getting the drivers installed on a dell precision t7400 machine for a quadro fx 570 card. This is a dual boot machine running XP and RH enterprise 4.

I'm a linux noob, I've had a course on it last summer but haven't used it since. This machine is also running a tesla C1060 card as well that I'll need to install afterwards for cuda development. All hardware works fine in windows.

I've tried running the nvidia installer, and everything seems to install fine, but when I go to restart the x server, there is a problem and it has to reconfigure to using the vesa generic card in order to work.

The card is not recognized in the hardware browser before and after installing the driver, and under display settings the card shows up as the vesa generic.

I would install from RPM's but I'm also missing some dependencies and don't trust myself as much at getting that working.

Anyone have any suggestions?

Thanks,
Luke

Red Hat Enterprise Linux and a Tesla GPU? That sounds like a tough combination.
I will tell you this. If I were you, I would trust RPM's and up2date far more than anything I might install to satisfy dependencies without them.

The Card is supported by RHEL 4, I found an nice article on the availability of the drivers at Red Hat in DOC-19565, but they say what you already know, and that is that the driver is available from Nvidia... The C1060 is available for 64-bit and for 32-bit. I assume this is 32-bit?

I would like to help you further, but I am curious as to how you have installed the driver. I have some questions for you.

1) How did you install the driver?
2) What version of the driver did you try to install? Is it 32-bit?
3) Was an nvidia-installer.log created in your /var/log/ ?
3a) Is there any information on problems with dependencies?
3b) Does the log show you a failure of any sort?
4) Do you have missing dependencies for only the Nvidia driver?
If you have issues beyond the driver, what is missing?
Have you been attempting to add other applications and hardware
without using rpm or compiling from source?

5) Can you provide the output from lsb_release -a?

if you PAID for the license ( a MUST for red hat )
for the rhel4 ( about to hit end of life VERY VERY soon ?????????
why RHEL4 and WHY RHEL4 on a desktop ???

if you DID NOT buy a NEEDED AND MUST HAVE License
then install RHEL 5.5 - the current AND BUY THE LICENSE

then install the nvidia.run driver from the nvidia website

PS. to install the .run driver you MUST have the kernel source and gcc & Auto tools installed
To do that on red hat YOU MUST HAVE A PAID FOR LICENSE !!!

It is my understanding that Red Hat is Free as in Freedom. I am pretty sure that you can install Red Hat, but you cannot receive support or download pre-compiled binaries or security updates directly from the Red Hat Network. Red Hat Network even allows you to register your install without entitlements.

Organizations like CentOS make compiled(binary) versions of the packages available from Red Hat by downloading the freely available source rpm's and providing them without the Red Hat trademarks and "as is", i.e. with no hope of technical support directly from Red Hat, the company.

RHEL 4 still has a few years left on the support cycle, but RHEL 3 is coming up on end of life this month. The RHEL 4 end of production 3 phase is scheduled for February 29, 2012

If you are using a copy of Red Hat Enterprise Linux 4 without entitlements then you can expect some flack from the community because those like John who are more proficient know there are much easier ways to get what you need.

That said, if you still want to get some assistance on installing your NVidia driver, let's get started.

RHEL is NOT free
yes you can install a unlicensed OLD and NOT UPDATED version for free
-- BUT there will BE NO software updates . no installing anything using yum, not in 4, or "up2date" .
-- and a very big pain in the butt to get anything to install ( because -THERE ARE NO UPDATES TO IT, everything is a wrong version from the centos repos )
only for the VERY VERY expensive EXTRA long life support

if one has a very old server then the support is there but not for home desktop use

like installing the Correct kernel source WITHOUT using "up2date" and the RHN
and the gcc compiler and Autotools also without RHN

Thanks for the quick reply guys. As for the choice of RHEL 4, this is a school computer, I wasn’t involved in their purchase or the choice of OS. For the most part they are used in windows but they’ve also been used in linux to host random servers for grad projects etc. It might be possible to have it upgraded, but if I can avoid having to do that just yet it would be nice.

Starting with Dav since he responded first:

1_2) I downloaded the driver from nvidia’s site. The version of windows is 32 bit, so I downloaded the 32 bit driver first, but the installer said I needed the x86_64 driver so I got that.

Next I did ctrl+alt+f1, logged in as root, init 3, ran sh <driver>.run.

Hit yes through agreements etc, said yes to “install nvidia’s 32 bit compatibility opengl libraries”

When asked if I want to run nvidia-xconfig I say yes (since I don’t know much about configuring X I figure that I’m more likely to screw it up doing it manually).

After that I ran init 5 and logged back in. Everything seems fine but in the hardware browser, under video cards, all I see is nVidia corp: Unknown device 05e7. If I go to system settings>display, there is no resolution greater than 800x600 and the video card is listed as VESA driver (generic). If I hit configure and try to select a different card (the quadro fx 570 is not listed, so I’ve tried quadro fx generic, quadro fx 550, etc (I know the 550 is a different GPU but since the 570 wasn’t listed I figured I would try it).

If I try to select either, it tells me I have to restart the X server for the changes to take place, I hit ctrl+alt+backspace but the X server can’t restart (problem with the config) and it goes through steps to repair. Attached /var/log/xorg.0.log. It asks to run the X configuration program, I say yes. It sets back to VESA driver (generic), says "couldn't start X server with card 0", "couldn't start X server with old config, trying new configuration", then it goes back to the gui with the top half of the screen garbled and the bottom half split down the middle (mouse wraps around to the left). I thought I was able to get this fixed before letting rh reconfigure xorg.conf itself, but when I tried it today to make sure I was right about what happened, the only way I could get back to a desktop was to manually change the driver back to vesa from nv.

3) Attached nvidia-installer.log. 3a) I got no dependency errors that I can see. Certainly nothing during the installer. If you look at the bottom, it says I didn’t run xconfig, I accidently hit enter thinking it defaulted to yes, so I went and ran it right after (but I ran it during the installer yesterday). There seem to be some compile errors in the log file, perhaps thats part of the problem?

4) Besides what I mentioned before with the unrecognized video card, I haven’t noticed a problem (well, the TESLA card seems to be missing I assume, unless it sees that and NOT the 570. Though, the tesla card doesn’t actually function as a video card (no video output at all) so hopefully that’s not the case).

5)
[root@dino ~]# lsb_release -a
LSB Version: :core-3.0-amd64:core-3.0-ia32:core-3.0-noarch:graphics-3.0-amd64:graphics-3.0-ia32:graphics-3.0-noarch
Distributor ID: RedHatEnterpriseAS
Description: Red Hat Enterprise Linux AS release 4 (Nahant Update 6)
Release: 4
Codename: NahantUpdate6

@John: The license was bought, I’m not sure exactly when or what the specifics of that were (the university might have supplied the licenses). It is certainly not a pirated version or anything of the sort.
How do I ensure that I have the kernel source and auto tools? The .run file seems to work fine, would I have gotten an error? I know gcc is installed (running 4.2.2).

@Dav again, I would greatly appreciate further assistance in getting the drivers installed. I hope I have not made any fatal linux faux pas as just yet ;-)

As for RPM’s: I found this site - http://www.brandonhutchinson.com/NVI...Hat_Linux.html and I went to ATrpms (livna.org doesn’t seem to work) and found 260.19.12el4.x86_64 driver rpm, which requires the kmdl rpm which I also have (they are on this page, http://packages.atrpms.net/dist/el4/nvidia-graphics/). When I tried to install the kmdl rpm, it said I needed /boot/vmlinuz-2.6.9-89.31.1.plus.c4smp and nvidia-graphics-devices. I think I know where the graphics devices are (http://packages.atrpms.net/dist/el4/...phics-devices/) though

I’ve not tried to install that yet. I’m not sure what the vmlinuz dependency is, hopefully one of you can help with that. Maybe up2date can help me with these? Coming from windows, dependencies are a bit of a change.

I'm not sure if it matters whether I boot into ELsmp or EL, the bootloader has both options, I think smp has something to do with single processor? This machine is a dual quad core.

Sorry for the monster post guys, and thanks for being so willing to help a noob. This seems like a great community.

Luke

I work with the Dell C410x which houses many Tesla GPGPU's available to a group of blades from a C6100, but I have not installed RHEL 4 on any of them. I will have to do that.


That's a little confusing, you are definitely using i386 according to the output of your lsb_release, but you say it required the x86_64 download and then built successfully. The nvidia-installer.log says so too. I don't see how that could not fail to load.

I would like to see the output of this command:
That should verify the binary type of the installed version for the running kernel.

It is listed on the nvidia driver download that the driver is the same. Getting the driver in place is the most critical. After you have the correct driver in place the rest is just naming.

Here is the device section from my x.org.conf
After configuration, I would expect yours to be very similar, if not the same.

From reading your config, it sounds like you are not using the nvidia-settings to configure your display after the failure.

I would try this after you get your system booted into graphical mode:
Once you have the graphical config you can verify your settings under the X Server Information section. From there you can go to the X Server Display Configuration and set your display settings.
Then save your configuration using the Save to X Configuration File.
Once you have that done, restart X.

did you run /usr/sbin/nvidia-config-display? that will be specific to nvidia. The other(system-config-display) will attempt to insert the gpl/supported driver.

The TESLA card has an entirely different device, but it should show up as one of your GPU's
You can search for them in the rpm database
that is the general search method I use.
I am not convinced you require an x86_64 version of anything just yet, but I have never even dreamed of configuring an i386 machine to use a TESLA GPGPU.
You are not using an x86_64 kernel yet, so I don't see how this requirement got started. It seems to me that it can only snowball into a long list of dependencies that you may not need at all!

if you have access to RHN through up2date and can update, you are 3 quarterly updates behind. Be sure and upgrade the kernel too, that usually requires a force for up2date.

ELsmp is the multi-core variation. That is the one you want. Keep them both because the single-proc version(EL) is important for troubleshooting.
[/QUOTE]

So let's try this:

• find the binary type of the nvidia driver (command above)
  
• determine why we need the x86_64 driver. Is this a requirement of the TESLA?
  
• Upgrade to RHEL 4.9 if you are allowed.
• Upgrade the kernel using up2date -fu if that is allowed.
  
• run the nvidia-config-display tool to ensure we are using the appropriate binary driver while configuring the xorg.conf.

how does that sound?

That sounds great! I'll try that the first chance I get when I get access to the machine next (probably later today or tomorrow). In the meantime I'll get permission to do the updates.

EDIT: What is it thats inconsistent with x86_64? Wouldn't core-3.0-amd64 indicate a 64bit AMD processor? Perhaps I'm reading that wrong? It seems that its at least not a i386?

EDIT EDIT: I think I might see what you mean, the processor might be x64, but I'm not running a 64 bit kernel?

Thanks again

Luke

yes. Let's be certain on what kernel you are running:
uname -a will give us the kernel and the arch.
like so.

This shows that I am using x86_64 or 64-bit kernel.

This is the result of uname -a, I'm just now working on the rest of the stuff but I figured I would go ahead and give you that.

Yay! You do require the x86_64 drivers. Everything is alright with the driver. so I was wrong about you running 32-bit. So we should continue troubleshooting the config.
Use the tools included with your nvidia driver. stay with that. I think you will get to the bottom of this very soon.

I get an error unless I take out the leading $, that might have just been your prompt? If I take it out, I get no output at all. I tried slocate nvidia.ko and get nothing, slocate nvidia pulls up a lot of things. Am I messing that up somehow or does that just mean the file isn't found?

I had to run updatedb before I could use slocate or locate though. It seemed to have completed, so that could possibly be part of it.

EDIT: I think thats probably because I uninstalled the driver when I was trying to install from rpm's because I didn't want to have them conflicting. I don't want to reinstall the driver while its updating so I guess I'll wait (I booted into single user mode before to do it and I probably don't want to do that right now).

I never did run that, I didn't even know it existed (no installation instructions I saw anywhere said anything about it). I'm running up2date right now, so after its done I'll reinstall the driver and try it.

I talked to the CS dept system admin and got him to start the update (he had all the information). Hopefully it will be done shortly, but there were a LOT of updates.

I'm getting there!

Thanks again for the help, I'll post an update as soon as up2date is done and I can start doing things again.

Luke

up2date just finished running, and I'm now running version 2.6.9-89.31.1.ELsmp. Since the kernel is x86_64, do you think any more updates are required? I'm going to install the driver and try the configuration tools you suggested. I should be able to run any updates that need to be done now without going through the SysAdmin.

EDIT: Success!!! It seems to be working!! I think it was the

Commands that did it. I really don't know for the life of me why the documentation that came with the drivers didn't mention either (well, they might have but not in an obvious place, I read many pages). The nvidia-xconfig seemed to be working against me too, configuring the xorg.conf file in a way that wouldn't even work at all.

In nvidia-settings, I can see both video cards and change their settings, and the drivers seem to be functioning correctly. I've not had a chance to install CUDA (I got excited by getting it installed that I almost missed class), but tomorrow morning I will work on that.

Im curious if you still think the kernel needs updating or not, or if you have any other last minute tips or suggestions, but aside from that, thank you VERY much for your help. The problem wasn't as bad as it seemed but it would have stumped me for a good while nonetheless.

At this point, you should work with the kernel you require for your software.
I am so glad!

I am happy I could help. I would upgrade only when you are ready to rebuild the driver or you need to fix a problem.

I noticed you haven't marked this as solved. Is there anything else you needed?