When buying used Polaris cards, check /sys/class/drm/card0/device/pp_od_clk_voltage to see if there’s something funky with the Pstates for memory and core clock. If there is… It has a mining VBIOS on it and Mesa is hiding that fact until you get crashes in games…
Original thread:
Finally have a Powercolor RX 580 8GB Red Dragon for Mesa testing and the immediate thing I noticed running Oibaf is catastrophic failures of the driver stack if you leave a RADV game running too long. (arguably, it was on ACO, but the failure was catastrophic to the point it rebooted the entire system, causing a green/purple screen of death) I downgraded to padoka stable and I still had image corruption issues… Then I remembered I had enabled Freesync the entire time.
Anyone else with experience that Freesync flakes out when you try to use it on a factory OC card or apply increased power targets? This would prove Freesync is still not in a satisfactory state and definitely not capable of being used on cards with factory OCs.
If it’s just me, I got a lemon. Though the weird thing is 1080p stress testing is fine. It’s only 1440p and 4K. (1440p and 4K are my equivalents to Furmark using Valley or Superposition, 1440p had visual corruption of the entire screen when the benchmark was in windowed mode)
It seems without Freesync on and without ACO, things are a lot more stable, but image corruption pops up whenever I engage Freesync.
All I want to know is if there are more issues engaging Freesync in Mesa at higher resolutions while at the same time tweaking the power limit. The desktop resolution is 4K and the processor is the Ryzen 5 3600X on a B450-I. I double made sure processor and memory instability was not the cause.
Was it mostly ACO? or mostly Freesync?
The only other thing I can think of was I was polling the monitoring stats of this:
Mesa “WORKING” hid the fact it was a mining BIOS. Unlike Windows where drivers won’t even install. I had to dig into the Pstates to find out it indeed was a mining BIOS.
Flashed a BIOS from Techpowerup using atiflash-linux and I’m good again. Still weary of the memory so I’m giving 1000mV rather than 950mV for the memory.
Got another problem with this former mining GPU. There’s a purple line that scans down when the GPU reaches the American Megatrends BIOS screen with the boot logo disabled. Memory problem?
What’s weirder is it seems only DisplayPort at 4K is affected.
Edit: Mobo UEFI issue apparently. With the Boot Logo enabled the lines don’t appear and with CSM disabled the problem doesn’t crop up when the boot logo is disabled.
Like, using passthrough? If so, then Mesa would not be able to even lay a finger on the card, since Linux itself has no access to it. If not, then yes, as the virtual machine would simply get a virtualized card, which has no barrings on the physical card.
I don’t want this to come off in a bad way but have you considered just making yourself a blog and using it to explore your issues… Tag those who can help (sparingly)?
Just an observation from you liking to figure stuff out and tinkering around… Seems more blog tailored idk that’s me I’m not saying in anyway this post is bad and don’t mean this in any negative conatation
There’s a blog category here, but this initially started as a question so I’m leaving the context of a question.
I have a Video series called Jack’s Hardware. I’m gonna give a PSA that Linux can mask the effect a mining BIOS was put on a used RX 400/500 series card. I’m just getting thermals on this PowerColor sorted out trying to find 2.5mm thermal pads.
