One common factor among the readers is that they have multi-GPU systems, typically an on-die Intel GPU (HD Graphics 4000 or something) integrated into their Intel CPU alongside a PCI Radeon GPU. My original hypothesis regarding the problem, therefore, was that the driver installation application was detecting the integrated Intel GPU and failing before locating the AMD GPU. However, mizoguchi666777 pointed me to a solution involving the disabling of the open source radeon driver’s power management facility.
The idea behind the solution appears to be that the dynamic power management feature introduced in kernel 3.13 is causing the PCI Radeon GPU to be powered down and therefore undetectable by the Catalyst installation application. Starting the kernel by adding radeon.runpm=0 to its arguments in GRUB should disable the power management feature and keep your card powered on and available to the Catalyst installation application.
To do that, simply boot your system and,
- Upon reaching the GRUB menu (where you select the kernel to boot), press ‘e’.
- This allows you to edit the default GRUB entry (or whatever menu entry you had highlighted when you pressed ‘e’).
- Append a space followed by the radeon.runpm=0 value to the end of the kernel arguments on the line which looks something like:
linux /boot/vmlinuz-linux root=/dev/mapper/fedora-root ro rd.lvm.lv=fedora/swap vconsole.font=latarcyrheb-sun16 rd.lvm.lv=fedora/root rhgb quiet LANG=en_GB.utf8
That line above will now look something like:
linux /boot/vmlinuz-linux root=/dev/mapper/fedora-root ro rd.lvm.lv=fedora/swap vconsole.font=latarcyrheb-sun16 rd.lvm.lv=fedora/root rhgb quiet LANG=en_GB.utf8 radeon.runpm=0
3. Press F10 to start your system with the edited line.
If this really does solve your problem, you should now be able to install the Catalyst driver using the instructions provided in the previous post. You shouldn’t have to modify GRUB to make this a permanent boot option since you won’t be using the radeon driver after a successful Catalyst installation.
Fabio seems to have demonstrated that this solution overcame the error described in the subject of this post, but he encountered subsequent errors. So, this is a solution which needs more testing. I don’t have a machine with the necessary configuration, so if anyone out there has this issue and can report back regarding this potential solution, please do so!