Memory access fault by GPU node-1 (amd rx6800 linux)

[daniij]

When generating several images, an error appears: Memory access fault by GPU node-1 (Agent handle: 0x589c1b587180) on address 0x7c29cb62b000. Reason: Page not present or supervisor privilege. webui.sh: line 304: 65316 Aborted (core dumped) “${python_cmd}” -u “${LAUNCH_SCRIPT}” “$@”.
Also, I may get a black square instead of a picture when generating. I have tried different stable and nightly versions of rocm pytorch 6.1, but the error persists. Solutions like adding the user to the render and video groups don't help. All this was running in ubuntu 22.04 container, host system arch linux. What can this error be related to? How can it be fixed? Please help me :(

[Oldcustard]

Having this same issue. RX 6800, Rocm 6.1, nf4 model. Running directly on an ubuntu 22 system. I can generate images 512x512, but not able to generate a test image at 896x1152.