Troubleshooting Ubuntu 20 and a fresh install of Amd drivers

mountkidd
mountkidd
Joined: 14 Jun 12
Posts: 175
Credit: 10892097506
RAC: 5537516

AMD are expecting to release

AMD are expecting to release amdgpu 20.50 on Mar 18.  It will have support for the 5.8 kernel.

Wedge009
Wedge009
Joined: 5 Mar 05
Posts: 117
Credit: 15650033764
RAC: 7422449

Indeed? I can't remember if

Indeed? I can't remember if they've supported HWE kernels before.

I wonder if ROCr will play nicely with BOINC GPU processing...

Soli Deo Gloria

robl
robl
Joined: 2 Jan 13
Posts: 1709
Credit: 1454480658
RAC: 8819

mountkidd wrote:AMD are

mountkidd wrote:

AMD are expecting to release amdgpu 20.50 on Mar 18.  It will have support for the 5.8 kernel.

I am waiting on two events to occur:  1.  Mar 18 and the release of 20.50, and 2.  some of you brave participants to install 20.50 and report on the success/failure of that effort.  I really don't have time to pursue empty promises with AMD drivers and Ubuntu.  I recently switched to win10 and like they say:  tah tah!!!!!

San-Fernando-Valley
San-Fernando-Valley
Joined: 16 Mar 16
Posts: 260
Credit: 6911481637
RAC: 21626072

robl wrote: mountkidd

robl wrote:

mountkidd wrote:

AMD are expecting to release amdgpu 20.50 on Mar 18.  It will have support for the 5.8 kernel.

I am waiting on two events to occur:  1.  Mar 18 and the release of 20.50, and 2.  some of you brave participants to install 20.50 and report on the success/failure of that effort.  I really don't have time to pursue empty promises with AMD drivers and Ubuntu.  I recently switched to win10 and like they say:  tah tah!!!!!

 

+1

mountkidd
mountkidd
Joined: 14 Jun 12
Posts: 175
Credit: 10892097506
RAC: 5537516

Success on 2 installs with

Success on 2 installs with amdgpu 20.50...

First install was on my "test mule", a vintage 2012 Dell laptop, w/i7-3612QM, AMD Thames 7500M/7600M video, running kernel 5.8.0-45 & Ubuntu 20.04.2.  Install 20.50 command:  sudo ./amdgpu-install -y --opencl=legacy,rocr --no-dkms --headless.  Video card doesn't have fp_32 support, so Boinc finds no usable gpu and won't do anything other than startup.  Previously w/amdgpu 20.45, opencl=rocr broke the install for every possible combination.  Principal video driver on this host is MESA & Clover (opencl) via the oibaf/ppa, as this video card is fglrx material.  Adding the amdgpu/opencl's is a bonus that serves no useful purpose other than the 5.8 kernel test.

The second install was on my prime host, 033 - i7-3770k, AMD RX480 running Ubuntu 20.04.2 over a 5.4.0-67 kernel.  Uninstalled amdgpu-pro 20.45 w/ sudo ./amdgpu-pro-install --uninstall and then sudo shutdown -r now.  Install 20.50 command: sudo ./amdgpu-pro-install -y --opencl=legacy.  None of the cpu/gpu hardware on this host is supported by ROCm, so I didn't bother adding it to the opencl install, but I do have another very similar host, 745, and may try rocr with it just because AMD says you can install both.

I do realize there are a number of other HW combinations, especially with newer AMD cards that require rocr,  other kernels, ie, 5.10/11/12 and other versions of Ubuntu 20.10, 21.04 that are non-LTS that other users may want to try with amdgpu 20.50. Will it still work?  Maybe.  Maybe not.

For now, AMD's claim of supporting 20.04 LTS on kernels 5.4 & 5.8 seems to be valid.

 

edit:

Updated host 745 as well, using opencl=legacy,rocr.  Installed without any problem!

Wedge009
Wedge009
Joined: 5 Mar 05
Posts: 117
Credit: 15650033764
RAC: 7422449

Thanks for the update. '18th'

Thanks for the update. '18th' was yesterday for me, so I forgot to check today.

As we discussed previously, looks like none of your GPUs actually uses ROCr, so everything is working fine with 'legacy' OpenCL, which is what I would expect. It's up to me to have a go at testing if ROCr is still broken (at least for BOINC processing) - I suspect it will be.

Soli Deo Gloria

Wedge009
Wedge009
Joined: 5 Mar 05
Posts: 117
Credit: 15650033764
RAC: 7422449

Tom M wrote: And then "it"

Tom M wrote:

And then "it" fell into the dreaded logon loop while I was migrating Rx 5700 gpus onto the box.

I swapped out the HD for a "test" version of Windows 10 and am up again (finally).

So it sounds like Tom wasn't successful with ROCr after all.

I just tested against Ubuntu kernel 5.8.0-45 with amdgpu-pro 20.50: no apparent difference - BOINC GPU processing still fails with ROCr-based OpenCL so I reverted to kernel 5.4.0-54 with amdgpu-pro 20.40 (last one with PAL-based OpenCL).

https://community.amd.com/t5/opencl/amdgpu-pro-20-45-rocr-vs-pal-opencl-breaks-boinc-gpu-processing/m-p/458229

Soli Deo Gloria

robl
robl
Joined: 2 Jan 13
Posts: 1709
Credit: 1454480658
RAC: 8819

mountkidd wrote: Success on

mountkidd wrote:

Success on 2 installs with amdgpu 20.50...

o

o

o

Thanks for the effort.  It gave me hope that this might be the fix we are looking for.  Longstory short:  It worked for me also.  I will post some additional stuff.  My proedure follows:

  • download Radeon™ Software for Linux® 20.50 Release from here:  
  • https://www.amd.com/en/support/kb/release-notes/rn-amdgpu-unified-linux-20-50
  • install ubuntu 20.04.2 desktop 
  • be sure to check the "do updates" during install
  • reboot after install completes
  • login in at a terminal and "uname -a" to check your kernel - mine was 5.8.0-45
  • install boinc ----> sudo apt install boinc-client boinc-manager
  • install openssh --> sudo install openssh-server
  • update OS again if necessary
  • reboot
  • install amd drivers:
  •     cd to where driver tar file is:  mine was in $HOME/Downloads/amd
  •     tar -Jxvf amdgpu-pro-20.50-1232447-ubuntu-20.04.tar.xz
  •     cd to $HOME/Downloads/amd/amdgpu-pro-20.50-1232447-ubuntu-20.04
  •     sudo ./amdgpu-pro-install -y --opencl=legacy
  •     you might be asked for a secure boot passwd.  This was new for me and might              have been requested because of my bios settings.  Not certain but write it down.  
  •     you will need it at next reboot.
  •     after the driver install completes reboot (you will most likely be asked for the secure boot password.

    
join the E@H project 
your pc should now properly display the gpu in the list of your PCs on E@H
 

 

mountkidd
mountkidd
Joined: 14 Jun 12
Posts: 175
Credit: 10892097506
RAC: 5537516

Robl, Well, that was

Robl,

Well, that was easy!

 

Wedge009
Wedge009
Joined: 5 Mar 05
Posts: 117
Credit: 15650033764
RAC: 7422449

I thought you were the one

I thought you were the one who smugly said 'ta ta'...

Anyway, looks like you're running GPUs older than Vega so you would have no problems.

Soli Deo Gloria

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.