AMD Drivers for Debian Jessie - Installing not successful

Message boards : GPUs : AMD Drivers for Debian Jessie - Installing not successful
Message board moderation

To post messages, you must log in.

Previous · 1 · 2

AuthorMessage
Profile Agentb
Avatar

Send message
Joined: 30 May 15
Posts: 265
United Kingdom
Message 73479 - Posted: 21 Oct 2016, 15:58:53 UTC - in response to Message 73478.  
Last modified: 21 Oct 2016, 16:13:26 UTC

BINGO!!!!!!!!
It is odd... In order for it to recognize and do any GPU work, I must still stop and restart the daemon.

One thing I didn't mention before was the pesky message I received from the BIOS before boot:
The system failed to boot several times."
Never saw that before installing the GPU. I reset the BIOS to defaults and now no message on startup.

And, something that I really do NOT like... X appears to stutter occasionally when the client is running with a GPU work unit running. The screen freezes temporarily and overall is non-responsive. Not sure what is happening...


Goodo.

OK not finished yet... it is worthwhile getting to the cause.

a) xhost commands do not function during the init startup for most recent Debian based distros. xhost must be started from within an X session. xhost WILL work because you are starting it from within an X window. If you wish this to be automated, then there is a little more work to do, to get around the problem.

b) making changes to the boinc-client script will get overwritten if a distro update occurs. If it is the delay (i somewhat doubt it as you have the gpu running by the time you run the script) then the correct place to add that and be free of updates clobbering it, is in my early post in /etc/defaults/boinc-client

It is important to identify what exactly fixed the problem.

I have asked for the xhost output, because the xhost local is a bit open, now any user such as guest etc will have full access (could read screens etc).

The command xhost local:boinc is actually badly formatted (very old format) it should be xhost local: - meaning all local hosts have access. To remove access xhost -local:

To specify just boinc it should be xhost +si:localuser:boinc (see man xhost) the plus(+) is optional.

Any chance you can provide the xhost output and retry with the standard boinc-client script?

Regarding freezing you might want to consier setting the use GPU when idle etc and see if that fixes the problem.
ID: 73479 · Report as offensive
Ken Linder

Send message
Joined: 27 Jul 11
Posts: 18
United States
Message 73480 - Posted: 21 Oct 2016, 16:56:43 UTC - in response to Message 73479.  

Thanks again!

I will get back to you soon. Was up all night last night for work and promised my wife I would work on our house remodel project. :-)
ID: 73480 · Report as offensive
Gordon Haverland

Send message
Joined: 25 Oct 16
Posts: 3
Canada
Message 73558 - Posted: 25 Oct 2016, 0:54:47 UTC - in response to Message 73478.  

And, something that I really do NOT like... X appears to stutter occasionally when the client is running with a GPU work unit running. The screen freezes temporarily and overall is non-responsive. Not sure what is happening...

But, it is indeed processing GPU work units now.


Greetings. It seems I have never logged in to BOINC itself, but I've been doing SETI, ClimatePrediction and others for a while. Or, it didn't recognize my email.

Anyway Ken, I have 4 Debian/stable AMD64 machines here, all with AMD GPUs. Smallest GPU is a HD5450, which is smaller than yours. I have 2 HD6450s (same), an A10 APU (R7 class), a R7-250 and now a RX460.

The stuttering is normal. That GPU is small enough, that stuttering will happen. You can set in your preferences to not run GPU jobs while the console is "active". Active being generating key scan codes from the keyboard or mouse movements. Normally this is a good definition, but if you hit Ctrl-Alt-F1 to get to a text UI session (console 1), this is still seen as activity on the keyboard, and will put your GPU task to sleep for as long as you indicated. Most people don't run text consoles and GUI consoles, so they don't run into this.

Looking for advice about fglrx (Debian package family) a couple of years ago, probably got better advice than now. Nobodies fault in BOINC.

The first proprietary driver was fglrx (Fire GL Radeon X, or something like that). Probably still when ATI owned it. After a while, a new proprietary driver came out (Catalyst). New driver, and so we will give it the name, ... fglrx. Catalyst will do a lot of things, but not everything (for example, it won't let my A10 with R7-250 both work). The new proprietary driver which will let both GPUs work is called Crimson, and its package name is, ... fglrx.

ATI and AMD have had a bunch of changes in hardware technology. R7-250 is about the start of GCN. The A10 APU I have also has this R7 type GPU in it, which is GCN. In any event, it would seem that yet another software technology got started, which seems to be called AMDGPU, which is meant to support GCN. As I understand it, there is an open source part called AMDGPU, and there is a proprietary part called AMDGPU-PRO. Vulkan is another name which pops up in this.

In any event, all non-GCN GPUs are somewhere in the process of being deprecated. And hence, bug fixes and documentation fixes, and things like getting Crimson to install just don't get much work any more. Or so it seems. I am wading through the Crimson install 'run' file setup, trying to figure out why it can't make a driver set for my machine. I am going cross-eyed. :-)

---

The reason for me to visit BOINC itself, is that the BOINC manager is getting flaky. I am guessing that it is trying to manage the generation of work units. CPUs and HD5450s and HD6450s and A10-GPU and probably RX460 generate work units for SETI at much different timescales. If I am generating miniscule units per time with the CPU and big units per time with a GPU, it causes problems for work with other projects that don't have GPU work. I can run out of work at WorldCommunity Project (and units are available), but it won't ask for units because it isn't the most important project on the machine (or something like that). So, I have to manually suspend the other projects so that World Community is the only project left, and then it will download more work.

At some point (hopefully soon), two things may happen. I will be too busy to micromanage BOINC, I will have computation intensive stuff of my own running, using up CPU and GPU (and maybe FPGA) resources. So this isn't a complaint, it is just an observation.
ID: 73558 · Report as offensive
Ken Linder

Send message
Joined: 27 Jul 11
Posts: 18
United States
Message 73605 - Posted: 28 Oct 2016, 18:52:35 UTC - in response to Message 73558.  

Thanks Gordon,

This is my first attempt at Linux + ATI/AMD GPU so I was stumbling a little. I activated the "Suspend GPU Computing when computer is in use" and there is no more stalling or stuttering. Other than the typical window drag lag with LXDE, response is fairly smooth.

The embedded GPU on the C-70 is a 6290; I believe it lacks the floating point capability required to handle SETI workunits so never really tried to make it work.

If I get another AMD/ADI video card for processing, I will definitely dig into the info you posted.

-Ken
ID: 73605 · Report as offensive
Juha
Volunteer developer
Volunteer tester
Help desk expert

Send message
Joined: 20 Nov 12
Posts: 801
Finland
Message 73606 - Posted: 28 Oct 2016, 19:28:31 UTC - in response to Message 73605.  

The embedded GPU on the C-70 is a 6290; I believe it lacks the floating point capability required to handle SETI workunits so never really tried to make it work.


Single precision floating point is what all GPUs have. Double and half precision floats and integers of various widths are then extra. Seti doesn't need more than single precision floats.
ID: 73606 · Report as offensive
Ken Linder

Send message
Joined: 27 Jul 11
Posts: 18
United States
Message 73607 - Posted: 28 Oct 2016, 19:43:45 UTC - in response to Message 73606.  


Single precision floating point is what all GPUs have. Double and half precision floats and integers of various widths are then extra. Seti doesn't need more than single precision floats.


Ah, thanks. I will need to revisit that when I have time. My main goal was to use the external GPU to run headless for BOINC projects while I used the C-70 GPU for normal operation. Was thinking about making the whole thing headless; then I will see about making that 6290 process work units.

For now my main goal is to make it run headless; may need to make that dummy monitor plug that was mentioned earlier.
ID: 73607 · Report as offensive
Ken Linder

Send message
Joined: 27 Jul 11
Posts: 18
United States
Message 73608 - Posted: 28 Oct 2016, 20:43:53 UTC - in response to Message 73479.  


It is important to identify what exactly fixed the problem.


OK, WHEW!!! I have identified what fixed the problem. Just like debugging an odd bug for my work, there isn't a single factor that caused the problem.

#1 - Install the proper up-to-date AMD/ATI GPU driver. I initially installed an old driver that didn't work. I found a posting somewhere that indicated I needed to install fglrx separately. That just added to the problem. Completely removing fglrx and the old driver, then installing the new driver was necessary.

#2 - Install BOINC from the Debian Jessie back-port repository. The version in the standard repository did not recognize the GPU even thought it was there and the driver was up-to-date.

#3 - Maybe the key that was confusing things... Reset MOBO BIOS to factory default, thus allowing it to properly identify & handle the new GPU. After installing the GPU card, the BIOS had been complaining "The system failed to boot several times." before booting into Linux. Resetting the BIOS to factory settings likely played some part in this.

I have removed my modifications to boinc-client and still need to do a restart so that the GPU is recognized. I can repair it if manually restarting the client becomes too much of an irritation.

Thanks so much again!
ID: 73608 · Report as offensive
floyd
Help desk expert

Send message
Joined: 23 Apr 12
Posts: 75
Message 73625 - Posted: 29 Oct 2016, 14:45:06 UTC - in response to Message 73608.  

I'm afraid I don't fully understand how you got it working, but good to hear you did.

Install BOINC from the Debian Jessie back-port repository. The version in the standard repository did not recognize the GPU even thought it was there and the driver was up-to-date.

Older BOINC versions expect to find the OpenCL library under a name that's not provided by current OpenCL packages. You would have needed to install the OpenCL development package. Recent BOINC versions, as included in jessie-backports, have a workaround for this. If you encounter a suggestion to manually add a link somewhere, ignore it.

I have removed my modifications to boinc-client and still need to do a restart so that the GPU is recognized. I can repair it if manually restarting the client becomes too much of an irritation.

If "repair it" means editing /etc/init.d/boinc-client, I strongly recommend against it. This file is not meant to be modified and unexpected things can happen if you do so. In general, don't make manual changes to system files or directories that aren't explicitly meant for it. This includes the link mentioned above.

As a side note, the BOINC packages in jessie-backports have been updated and a new /etc/init.d/boinc-client has been installed today. Not sure what would have happened to your modified file. Probably you would have been asked if you wanted to keep your modified version or install the new one. That's both not exactly what you would have wanted, but some predefined default action could have been really problematic.

You seem to be running lightdm. Open /etc/lightdm/lightdm.conf, find the line
#display-setup-script=
and change it to
display-setup-script=/usr/bin/xhost +si:localuser:boinc
That's not "xhost local:boinc" as has been mentioned before.

Now for the dummy plug. I've built quite a few GPU crunchers running Linux, both nVidia and AMD, including headless or dual GPU with one monitor, and I have never needed a dummy plug. Have you verified that it's still not working for you? If so, please show your xorg.conf.
ID: 73625 · Report as offensive
Previous · 1 · 2

Message boards : GPUs : AMD Drivers for Debian Jessie - Installing not successful

Copyright © 2022 University of California. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.