Linux users of open source AMD drivers, are you happy?

[prakashmane]

I am thinking about replacing an aging video card in one of my Linux computers and I really like one of those small RX560 cards that do not require additional power other than PCIe. I have tried one a little over a year ago with Ubuntu 14.04 LTS. The functionality of the open drivers was awesome, but power management - not so much.

Specifically, the fan control was completely manual. It would never turn off (as it did on Windows) and run at a constant speed. The temp was 31C and that never-ending z-z-zzzzzz... was driving me nuts. I followed a bunch of blog posts and random Github repos and had to fiddle with a bunch of scripts writing/reading into sysfs. So I ended up ripping the fan off and throwing it away (I am not a gamer) which probably contributed to the card dying 3 months later. It got replaced with an old nVidia card with a proprietary driver which behaves exactly like it should - the fans are always off, unless I try to play a 4K video or something...

[bitnick]

On my Sapphire RX 5500 XT the fans stop completely under low load conditions. Before that I had a Sapphire RX 560 - on that card the fan ran slowly under low load conditions: quiet, but it could still be heard as a low "fluttering" sound. Silent enough that I mostly noticed it when I turned the computer off and the sound disappeared. It is nicer with a completely silent card though.

BTW, my first RX 560 killed itself by never turning on the fan, and the replacement did the same thing in that computer. I still don't know why. That the cards don't thermal throttle enough to survive a stopped fan is a big bug in the cards' firmware IMO.

[chithanh]

I am rather happy with AMD's open source Linux drivers. But there are a number of annoying little things that don't work.

I have four AMD systems: A personal netbook with a Kabini APU (A4-1200), a personal Ryzen desktop with a Tahiti GPU (R9 290), and at work a laptop with a Raven Ridge APU (2700U) and a work desktop with a Bristol Ridge APU (A12-9800).

• With the Kabini APU and open source radeon drivers, the essential things mostly work. With amdgpu, suspend-to-RAM doesn't work properly[1] and the VGA output doesn't work[2]. I cannot use ROCm/amdkfd with either radeon nor amdgpu, despite Kabini having a Sea Islands generation GPU.
• With the Ryzen desktop, I have to choose between using the radeon driver which supports the Ryzen memory encryption (SME, which is a good defense against Rowhammer/RAMbleed[3]) and but does not support ROCm/amdkfd, or the amdgpu driver which doesn't support memory encryption[4] but supports ROCm/amdkfd. Before I had the R9 290, I used a HD6870 which had GPU crash during video decode[5], this issue remained unfixed until I retired that card.
• The Raven Ridge laptop has a somewhat buggy BIOS (this is very common on AMD, compared to Intel they seem to enforce much less quality here) and so ROCm/amdkfd works only if I neuter the IOMMU[6]. Also, two recent kernel releases were broken in very obvious ways[7, 8], and mainline kernels did not support important functions of the Raven Ridge APU for over a year after release[9]. AMD seems not to test new Linux kernels on a wide range of hardware.
• The Bristol Ridge desktop failed to resume from suspend-to-disk for years[10]. That one got fixed in early 2019.

[1] https://gitlab.freedesktop.org/drm/amd/issues/205
[2] https://gitlab.freedesktop.org/drm/amd/issues/341
[3] https://rambleed.com/
[4] https://gitlab.freedesktop.org/drm/amd/issues/832
[5] https://gitlab.freedesktop.org/drm/amd/issues/630
[6] https://gitlab.freedesktop.org/drm/amd/issues/4
[7] https://gitlab.freedesktop.org/drm/amd/issues/842
[8] https://gitlab.freedesktop.org/drm/amd/issues/934
[9] https://www.phoronix.com/scan.php?pa...-I2C-Linux-5.2
[10] https://bugs.freedesktop.org/show_bug.cgi?id=100979