9070 and 9070 XT -- Setup Notes for Linux

Maybe, but you can’t objectively state that at some point the errors will become uncorrectable. It is constantly spamming dmesg to the point that it hides anything else it’s useful to hide it. The pcie slot is fine, works fine with other cards and has been reseated multiple times. Mobo bios is updated. Changing pcie speed did nothing, the preferred speed is whatever uses the least power as pcie bandwidth isn’t a factor (compute workload 99.9% kernel bound).

I’m chalking it down to early-adopter issues and my fix is to wait. If AMD wants to pay me to debug then they can DM me :wink:

I got the cable and it’s working fine. So far I havent updated the firmware and it connects with:

60Hz
VRR
3840x2160p@60
RGB 12b 4L6 SDR

Will update the firmware and try again. I will test with Doom 2016 under Vulkan from Steam. I would like to have HDR and VRR working.

Edit: linux-fiirmware git has updated dcn*.bin files released within the last 24h. I’m updating.

Updated the firmware for the cable and it works at a fixed 60Hz. I get a choice of up to 120Hz but if I select that the screen goes blank and doesn’t come back.

Try the new AMD driver instead AMDVLK 2025.Q1.3, if you have an AMD card that is… I been using it, I only have some minor glitches like shaders randomly taking longer when launching a steam game just because I have it set to compile the shaders before launch, since I hate shader stuttering.

Got some new issue after updating to kernel 6.14 and mesa 25.0.2-2.

[ 5570.744566] amdgpu 0000:03:00.0: amdgpu: [gfxhub] page fault (src_id:0 ring:24 vmid:4 pasid:32798)
[ 5570.744570] amdgpu 0000:03:00.0: amdgpu:  in process GameThread pid 13334 thread vkd3d_queue pid 13635)
[ 5570.744572] amdgpu 0000:03:00.0: amdgpu:   in page starting at address 0x0000008001326000 from client 10
[ 5570.744574] amdgpu 0000:03:00.0: amdgpu: GCVM_L2_PROTECTION_FAULT_STATUS:0x00401431
[ 5570.744574] amdgpu 0000:03:00.0: amdgpu:      Faulty UTCL2 client ID: SQC (data) (0xa)
[ 5570.744575] amdgpu 0000:03:00.0: amdgpu:      MORE_FAULTS: 0x1
[ 5570.744576] amdgpu 0000:03:00.0: amdgpu:      WALKER_ERROR: 0x0
[ 5570.744576] amdgpu 0000:03:00.0: amdgpu:      PERMISSION_FAULTS: 0x3
[ 5570.744577] amdgpu 0000:03:00.0: amdgpu:      MAPPING_ERROR: 0x0
[ 5570.744577] amdgpu 0000:03:00.0: amdgpu:      RW: 0x0
[ 5570.744584] amdgpu 0000:03:00.0: amdgpu: [gfxhub] page fault (src_id:0 ring:24 vmid:4 pasid:32798)
[ 5570.744585] amdgpu 0000:03:00.0: amdgpu:  in process GameThread pid 13334 thread vkd3d_queue pid 13635)
[ 5570.744586] amdgpu 0000:03:00.0: amdgpu:   in page starting at address 0x0000008001326000 from client 10
[ 5570.744590] amdgpu 0000:03:00.0: amdgpu: [gfxhub] page fault (src_id:0 ring:24 vmid:4 pasid:32798)
[ 5570.744591] amdgpu 0000:03:00.0: amdgpu:  in process GameThread pid 13334 thread vkd3d_queue pid 13635)
[ 5570.744591] amdgpu 0000:03:00.0: amdgpu:   in page starting at address 0x0000008001326000 from client 10
[ 5570.744598] amdgpu 0000:03:00.0: amdgpu: [gfxhub] page fault (src_id:0 ring:24 vmid:4 pasid:32798)
[ 5570.744598] amdgpu 0000:03:00.0: amdgpu:  in process GameThread pid 13334 thread vkd3d_queue pid 13635)
[ 5570.744599] amdgpu 0000:03:00.0: amdgpu:   in page starting at address 0x0000008001326000 from client 10

hardlock crash in Marvel Rivals for me :<

Pageflip timed out on RX 9070 (#4095) · Issues · drm / amd · GitLab - appears to be this issue in the bugtracker.

Well, I tried installing Win10 to see if I could get 4k@120 working via the DP to HDMI cable linked above with that special firmware. No cigar. I just got a desktop with non-functioning mouse and partial keyboard support.

I also feel slightly violated that Windows touched my actual hardware.

2 Likes

Weird, I bought this thing and it seems to work fine:
Gets 4k 120 on Linux Amazon.com: CalDigit DisplayPort 2.0 to HDMI 2.1 (DP to HDMI) Active Adapter 8K 60Hz UHD (7680 x 4320), High Refresh, 10 bit Color, HDR and Eyefinity Support (DP 2.0 to HDMI 2.1 (8K)) : Electronics

Welp, that kinda sucks. Decided to try and fire up some PCSX2 with the 9070, and uh… It doesn’t work.

It doesn’t see it at all. Trying to run it in Vulkan errors out, stating that no compatible adapters are found. This is regardless of using mesa-git or the AMDVLK drivers. Popped a B580 in and got some interesting results from it, including being able to run PCSX2 in Vulkan under either the B580 or the 265k’s onboard graphics. (with the 9070 installed, PCSX2 doesn’t see the iGPU or the 9070 with display via the 9070) Maybe oddly the onboard graphics did better in terms of average framerate than the B580 with the graphics upscaling turned down, but this generally sucks because my only “good” solution for PCSX2?

Yeah it involves an RX 580. The 6700 is out due to the HDMI audio problems, the 9070 has just dropped from mY pErFeCt GpU to “I still have to keep an RX 580 around”, and the B580 is “the Linux drivers are still terribad”.

Using PCSX2 v2.3.212-1 from Arch AUR on my 9070XT vulkan Rendering works in the settings it allows me to pick between “default” and “AMD RADEON GRAPHICS”, my IGPU is turned off

Tested with Ridge Racer V and GT4

PCSX2 v2.3.212-1
Kernel 6.13.8
MESA 25.1.0
AMDVLK 2025.Q1.3

1 Like

Manually grabbed the PCSX2 Nightly flatpak after seeing I was alone in that, and it’s functional.

Although Vulkan options still don’t show it as an option, and it only allows “default” for OpenGL, my experience has been that OGL works better for most things in PCSX2 either way. Kinda odd that I must’ve been out of date on PCSX2 though; normally Fedora was pretty good about updating it in previous installations.

I’ll order this one as well and try. Can you check what mode it connects with? There is some problem with HDMI and Linux if I understand it correctly.

Not sure what you mean by mode, it’s a DP to HDMI cable that can handle DP2.0 and HDMI 2.1 and has HDR and can work at 8k 60/4k 120 and capable of 10bit color. I Tested it on Linux Distro OpenMandriva Kernel 6.14 using the AMDVLK 2025.Q1.3 driver. I also tested it in Windows and can turn on 10bit and 12bit without issues. Tested it on Linux with an ASRock 1440p Ultrawide and it runs at 100hz which is not my ultrawide monitors default with the DP to HDMI cable I linked, when usually HDMI by itself without the adaptor it runs at a locked 120hz on Linux and HDR works perfectly also.
My second monitor is a 4k 3840x2160 BENQ monitor but the monitor itself is limited to 60hz but HDR works on that one perfectly also. As far as 4k 120, I cannot test that, the comments on the adapter itself on Amazon are saying that it works on Linux at 4k 120hz perfectly without issues.

I think it also may depend on what HDMI cord you are using also, im using just a generic one, so once you get the adapter you might need to also get an HDMI cord that has good specs. Also the audio pass through seems to work on it.

1 Like

That’s what I mean, I couldn’t get VRR or anything above 60Hz working with the other adapter linked above.

If I connect directly over HDMI I can get 100Hz but then it’s YCbCr 420 and I can definitely tell the difference.

For my use case 4k@100Hz fixed 10bit would be perfectly fine. Ubuntu 25.04 is supposedly HDR enabled.

I’m installing Bazzite on a spare drive to test as well.

Interesting new issue, my left screen the 4k BENQ screen seems to be stuck in HDR mode now and wont come out of it when un-checking the box, and I can’t turn it off with the HDR button on the monitor or in the monitor menu…lol

Just following up on the setup of Stable Diffusion type tools… I’ve installed torch, torchvision and their dependencies from the pytorch nightly repos for 6.3 … and image generation works, sort of.

At the moment, I can only generate at 512x512 even on SDXL based models, even though these should usually be used at 1024x1024 or similar, otherwise I get errors about running out of VRAM, even with the same settings that worked just fine on my 16GB RX6800.

However, it does at least work, so hopefully with improvements in ROCM this will improve.

Anyone have any recommendations for stability? I’ve been having nothing but hangs on my DP-1, but then I’ll load up a game and it’ll be fine randomly.

Thought of downclocking or something to help but again no clue.

Not really had any instability unless messing with overclocking, but you can undervolt and drop the power limit, was playing around with -80 uv and 230-250w power limit, playing The last of us part II and deadspace remake, very cool what these cards are capable of efficiency wise.

1 Like

I’ll give it a shot, thanks mate.

How’s that working for you so far? Any instabilities? Could you share your hardware setup and settings?

hm, that means I could get an xt and use my 650w rig for now? (gimp down a 9070xt to an non-xt)

what are the rough perf differences of a ~230w xt vs a 220w non-xt? any ideas?