AMD Delivers Many Fixes For Polaris GPUs On Linux - Finally Enables ZeroRPM Fan Mode

[perpetually high]

Just finished compiling and testing these on the amd-staging-drm-next branch, as they did not patch cleanly on the latest 5.9 mainline kernel.
(Also big thanks to Michael for the heads up, and EQ@amd for the patches!)

I can confirm Zero RPM is working on my RX 480:



It did, however, introduce flickering issues on my multi-monitor setup (two 24" 144hz FreeSync screens) when overclocking the video card. Without overclocking, no flickering issues.

Running sudo cat /sys/kernel/debug/dri/0/amdgpu_pm_info showed VDDGFX fluctuating between 800 mV to 975 mV, and SCLK and MCLK both set to 300.

By running the following commands I was able to fix the flickering by changing the mclk value to a higher clock, and then setting it back to 300 with a fixed 975 mV. It appears 800 mV is too low of a setting for dual-monitors at 300/300.

rocm-smi --setmlevel 0 2200 1075 --autorespond y
sleep 2
rocm-smi --setmlevel 0 300 975 --autorespond y

Not sure why that works. (the "sleep 2" command was necessary; sleep 1 or omitting sleep doesn't allow the change to "settle in" if used in a startup script). Interesting to note that both rocm-smi commands above are necessary to fix the flickering. Simply setting mclk to 300 with 975 mV does not fix it. Also, if I don't use 2200 but use 600 (just something higher than 300), the flickering comes back after a few seconds. So I'm a bit stumped why this workaround fixes the flickering, but it does.

So after the above, I have no other problems with the patches.

A couple other things:

Prior to these patches coming out, --setmlevel via rocm-smi did not work in changing the mclk frequencies.

Also prior to these patches, although I was able to overclock my memory (say to 2200, stock is 2000), it was stuck on that frequency and no longer able to scale down (say to 300Mhz) which resulted in using far more power at idle or light use than I needed, as shown below:



Last thing I want to mention, using Feral GameMode's config to set a script to run at game start and game stop, I can set the mclk to 2200 (overclocked value) and 300 (powersave value) to save on power when idle and still enjoy the overclocking when gaming. Probably too much tinkering for the average user, but wanted to mention it.

[M@GOid]

Despite the praised AMD drivers on Linux, I did have multiple problems with my RX 590. Any app going fullscreen appears to cause a crash after a few seconds, and my system freezes reliably after a few seconds of high load on the H265 hw encoder.

Great that they are still working on it though. Hope they fixed these.

If your unit is not defective, maybe your PSU is under-powered for that card, or it is defective? Many years ago, I bought a factory overclocked R9 290, to use with a 550W Corsair PSU. I got the type of problems you are describing. The thing boots and open the desktop, but any serious load will make it crash. It was resolved with a more powerful PSU (a 750W EVGA one, but the longevity of that one turned out not being as good as the 550W Corsair).

[Tumm]

If your unit is not defective, maybe your PSU is under-powered for that card, or it is defective? Many years ago, I bought a factory overclocked R9 290, to use with a 550W Corsair PSU. I got the type of problems you are describing. The thing boots and open the desktop, but any serious load will make it crash. It was resolved with a more powerful PSU (a 750W EVGA one, but the longevity of that one turned out not being as good as the 550W Corsair).

It works fine during gaming though (on both Linux and Windows). On Windows, there are no issues with fullscreen and H265 hw encode with Handbrake is stable.

[pal666]

Nope. In the last 10 years I buy only PSU's with good reviews from Corsair or Seasonic

good psus from seasonic have warranty longer than 10 years, why do you keep buying them?

[fenixex]

... the 580 was super popular among crypto miners. So the reality is a bit skewed.
Seems to me the GTX 1060 and 1660 (and their Ti/Super variants) were otherwise the most popular among gamers.

I assume you're referencing the Steam charts. The one which list 5 different models of the 1060 (1 which in only available in China) as one card.
It's also the card which is purchased in bulk for internet cafes throughout Asia.
Lets talk skewed.

[Guest]

Since we're talking about Polaris; is there anything known about potential HDMI issues? I've had a RX 560, and 4 different RX 580s, and all of them had intermittent instability at 4K@60Hz over HDMI across 6 different HDMI 2.0-certified cables.

The RX 560 and 2 RX 580s were from XFX, and the last 2 RX 580s from SAPPHIRE. This happens in Windows 10, macOS (Mojave and Catalina), and Linux. Tried about 5 different motherboard with both PCI-E 2.0 and 3.0, at least 2 power supplies, and in the case of macOS, I was using a TB2 enclosure. On all operating systems, the instability went away when I created a CVT-RB resolution.

I was thinking it was my display, but I used a GTX 1060 that worked without issue.

Some reports from others leads me to believe Polaris has some flawed HDMI implementation, but I don't hear a lot about this I'm assuming because it's not a popular set-up (I imagine most people use DisplayPort or resolutions that aren't 4K).

[theriddick]

Very good, maybe one day soon they can use this driver on Windows also and just have the whole directx components as a add-on.

[timrichardson]

polaris will be around for years to come. someone getting a cheap machine together could quite easily get a polaris GPU and have a fantastic experience under Linux!

Linux Torvalds has a polaris card in his threadripper beast. I have a RX570 in my 3900X. If you only game a bit or not at all, Polaris cards are fine. Perhaps there is a payback on energy savings of a more modern card, I'm not sure, but the other advantage is that the drivers seem to give no problems. It was my first AMD card on a machine that only has Linux so I wanted to minimise risks. So for low cost and low risk, these cards were still pretty viable only a few months ago, I think.

[grung]

Since we're talking about Polaris; is there anything known about potential HDMI issues? I've had a RX 560, and 4 different RX 580s, and all of them had intermittent instability at 4K@60Hz over HDMI across 6 different HDMI 2.0-certified cables.

...

Some reports from others leads me to believe Polaris has some flawed HDMI implementation, but I don't hear a lot about this I'm assuming because it's not a popular set-up (I imagine most people use DisplayPort or resolutions that aren't 4K).

I used to have overclocked XFX RX480 8GB and I was not able to use it for Windows Mixed Reality with default HDMI cable. To mitigate this I bought DP-to-HDMI converter and my WMR started working like a charm. So I think you are right there is some problem with HDMI on Polaris.
Last year I upgraded my GPU to Vega 56 and now I don't have to use the converted which clearly means that there is something wrong with the hardware.

[skeevy420]

Some reports from others leads me to believe Polaris has some flawed HDMI implementation, but I don't hear a lot about this I'm assuming because it's not a popular set-up (I imagine most people use DisplayPort or resolutions that aren't 4K).

I run 1080p HDMI to my TV and have had the same issue on Windows with my RX 580 since the day I bought it. Oddly enough, the exact same setup works flawlessly on Linux.

With all the posts I've seen around HDMI issues, there does seem to be something up with AMD's HDMI implementation. My guess is something to do with HDCP handshakes due to how my TV behaves with my RX 580 on Windows but not Linux.