Does your failure match the intro-to-3D problem?
The AMD device-loss tracker, issue #871, describes Ocarina of Time losing the Vulkan device on some AMD setups. Many reports fail shortly after the intro movie ends and the first 3D scene begins, or after a save loads. Other reports fail later in gameplay.
On Windows the report may involve a display-driver timeout. Linux reports can involve an amdgpu reset and, in some cases, loss of the desktop session. This does not mean every crash on every card is the same bug. Match the actual timing and log before using a workaround intended for this tracker.
If the emulator never opens, use platform setup. If the game is not listed, use game not showing. Neither problem is diagnosed by a tracker for the first 3D scene.
Record the details before changing settings
Capture:
- Lichii build number and whether it is standard, PGO or another build.
- Exact GPU model, operating system and graphics driver or Mesa version.
- Game version and any enabled add-ons or changed settings.
- Whether it fails after a new-game intro, after loading a save or later in play.
- The affected session’s log and the time of any graphics-driver timeout.
The tracker asks for eden_log.txt and, if the app has been restarted, eden_log.txt.old. Check files for private details before sharing. Do not upload games, keys, firmware, shader dumps or private account information.
Does build 117 fix this?
The build 117 release fixes a particular build 116 regression during Ocarina of Time shader building on RX 6000 and older Radeon families. The remaining device-loss tracker describes failures on builds containing earlier fixes, so “AMD crash fixed” in one note should not be generalized to every AMD failure.
Check the current state of issue #871 and newer release notes. This guide reflects the sources checked on 9 November 2026 and cannot guarantee that an old comment describes your current build.
What about the mesh fallback workarounds?
Issue #871 records different outcomes for experimental environment variables. At the time reviewed, forced mesh fallback had positive reports on some Linux RADV setups, with a possible performance cost in foliage-heavy scenes. Another no-mesh workaround could remove terrain, plants, doors or push blocks. Results on Windows were mixed or incomplete.
That evidence is useful for a carefully controlled test requested by the maintainer; it is not a universal recommended preset. Follow the current issue instructions for a matching build and system instead of copying a command from an unrelated post. Back up your data first and keep one unchanged comparison run.
If the machine repeatedly resets, preserve what you already observed and consult the tracker rather than repeatedly forcing the same crash. A report with accurate conditions is more useful than a claim that you tried every setting.
What should you do if a newer build changes the result?
Keep the older and newer build numbers, reproduce the same scene if it is safe to do so, and note the one changed variable. Report a regression if an unchanged setup previously worked. An issue may already consolidate similar reports, so search and contribute there before opening a duplicate.
After a crash is addressed, test rendering and saving separately. Getting past the intro is progress, not proof of a complete error-free session. Continue with the Ocarina of Time guide or graphics and audio diagnosis if another symptom remains.