AMD Is Hiring Another Lead Linux Kernel Developer To Work On Their Graphics Driver

[NotMine999]

I wonder how many people AMD has employed for Linux development pertaining to AMD CPUs. If past performance is any indication of the future, the auguries are not good.

[dwagner]

I'm running a Polaris card on Linux and I'm having the best open-source gaming experience of my life.

I'm "running" a Polaris card on Linux and I've had my worst experience of driver instability for (non gaming) computer use in my entire life, across all the many computers I owned or worked with. So much so, that I am now only desperately waiting for any discrete Xe GPU to arrive on the shelves.

I'm afraid somebody competent enough to turn the amdgpu driver into something that does not crash the system frequently might just shriek hysterically and hand in his notice after having had his first glance on the hardware documentation that he is supposed to base his work on. Or maybe even earlier, after looking into the history of commit messages.

[tuxd3v]

So now is the time for a small thank you in return - a Linux GUI.

Agree with that, AMD should had already unified Windows and Linux "Control Settings" or "AMD settings"..
They can do it for example with QT Framework..

There seems to be out there a Gui, for AMD, QT based, I just don't know the name.. but when I saw it it was very nice, and they can use that work as a base, to integrate the features they want..

The AMD team really needs more developers, to catch up, in the Linux Market..

[Venemo]

I'm "running" a Polaris card on Linux and I've had my worst experience of driver instability for (non gaming) computer use in my entire life, across all the many computers I owned or worked with.

I'm sorry to hear about your experience.
Can you please let us know which card is it and what sort of issues you have with it? Also can you open bug reports about your problems, if you haven't yet?

[phrix]

The issue with their drivers has probably little to do with the available manpower, but with the culture that remains from before AMD bought ATI.

Exactly my opinion. They should change the managers there and change how things are going.

One developer is way too little, I should say a team of 5 to 10 people. Don't forget the QA team. Or the customers are doing the QA part?

[illwieckz]

My primary complaint with AMD now is that the entire idea of a settings GUI for Linux seems to have been lost to the wind. There's no legitimate reason for them not to create an Adrenalin equivalent for Linux any longer.

The legitimate reason is that it's a niche market in a niche market. There is probably more people doing crane RGB ligthing than people wanting a GPU tweaking GUI on Linux.

Anyway, have you tried radeon-profile ?

[illwieckz]

There seems to be out there a Gui, for AMD, QT based, I just don't know the name.. but when I saw it it was very nice, and they can use that work as a base, to integrate the features they want.

https://github.com/marazmista/radeon-profile ?

It works. It's based on the hard work AMD developers do on the kernel. I don't know what people are asking for, an AMD badge on it?

[dwagner]

Can you please let us know which card is it and what sort of issues you have with it? Also can you open bug reports about your problems, if you haven't yet?

The device is a XFX RX460, original BIOS of course and not overclocked at all (which probably does not matter because the crashes occur most often when there is low load, anyway).

The most devastating, show-stopping bug I filed a report for 2 years ago - https://gitlab.freedesktop.org/drm/amd/issues/226 - I also filed other reports, and after realizing that the frequency of these crashes increased, rather than decreased, with newer kernel versions over a period of 1.5 years, I gave up hoping for any solution.

If you look into just the code that determines which memory and shader clock frequencies to use, it is hilarious how often it was changed back and forth, with commit messages indicating how these were more like random shots in the dark, hoping that the Nth iteration of 5 lines of code may possibly end up being correct without knowing why they should. To this day, if you e.g. decrease the screen refresh frequency, you will experience weird increases in shader clocks and vice versa - all indications that the code is not based on knowledge but mere guesses what the hardware does.

[Jabberwocky]

The device is a XFX RX460, original BIOS of course and not overclocked at all (which probably does not matter because the crashes occur most often when there is low load, anyway).

The most devastating, show-stopping bug I filed a report for 2 years ago - https://gitlab.freedesktop.org/drm/amd/issues/226 - I also filed other reports, and after realizing that the frequency of these crashes increased, rather than decreased, with newer kernel versions over a period of 1.5 years, I gave up hoping for any solution.

If you look into just the code that determines which memory and shader clock frequencies to use, it is hilarious how often it was changed back and forth, with commit messages indicating how these were more like random shots in the dark, hoping that the Nth iteration of 5 lines of code may possibly end up being correct without knowing why they should. To this day, if you e.g. decrease the screen refresh frequency, you will experience weird increases in shader clocks and vice versa - all indications that the code is not based on knowledge but mere guesses what the hardware does.

This is very interesting. It seems you're not the only one.

I've been using a rx480, rx580, and 4 different rx 560 cards in 4 different computers. All the way from Ivy Bridge i3 to Ryzen 7 2700X. Not a single problem.

Give your situation and where I have been in the past, I do hope that improved debugging or fallback/recovery gets attention.

Good luck!

[aht0]

Understood; doesn't really change my point that their Windows team really needs help.

+1, have to agree. Newest AMD GPU's on Windows.. drivers are horrible. Discrete Vega's seem fine but Navi's - too many problems to be worth buying.