Graphic issues on RX 7900 XT #9

Activity

1. rigaz29 commented on Nov 12, 2026

   

   rigaz29

   on Nov 12, 2026

   Owner

   More actions

   Thanks for the two reports and the screenshots, they helped a lot.

   This is not specific to your RX 7900 XT. Right after the Delphi temple loads, the game's own video memory manager starts printing errors (invalid size, CopyLocalMemory ... offset error): its bookkeeping got damaged, and from then on textures are read from the wrong places. That is the striping and noise in your screenshots, the HUD text included. In your second session the same damage later sent the game through broken pointers, and it froze.

   In both sessions it starts right after a cache of the port's own fills up for the first time: the one that keeps EDGE's vertex results on the GPU. So the bug is very likely ours. I could not make it happen with the saves I have, though. It seems to need the temple.

   Two things would help me find it:

   1. Your save. Press Report a problem in the launcher and tick Include my saves, so the problem can be loaded where it happened, then attach the new zip here. It holds your save data, no game files. With it and the pad recording already in your reports, I can play your session back here.
   2. If you have a minute, a test with that cache off. Put a file named test.bat next to Ascension-R.exe, containing:

      @echo off
      set GOWA_EDGE_RESIDENT=0
      start "" "%~dp0Ascension-R.exe"
      

      Start the game by double-clicking it, load the same save, and tell me whether the temple still breaks. It can run a few frames slower; the setting lasts only for games started from that file.

   Please don't attach any of the game's own files.
2. rigaz29

   added a commit that references this issue on Nov 12, 2026

   EDGE fills of skinned meshes: the SPU's rounding

   ...

   8b6bd74
3. soldiermxdeath commented on Nov 12, 2026

   

   soldiermxdeath

   on Nov 12, 2026

   Author

   More actions

   This was before test.bat
   report-20261111-205055.zip

   

   And when I used test.bat, the game froze. I had to close it with alt + f4 again.

   report-20261111-210027.zip

   
4. rigaz29

   added 2 commits that reference this issue on Nov 12, 2026

   Temple note: the reporter's save does not break it here

   ...

   67e4ba9

   Guard the game's VRAM manager against negative block sizes

   ...

   523875d
5. rigaz29 commented on Nov 12, 2026

   

   rigaz29

   on Nov 12, 2026

   Owner

   More actions

   Thanks for the saves and the two sessions, they made the difference.

   A correction first: my earlier guess was wrong. Your session with GOWA_EDGE_RESIDENT=0 shows the same errors, so the vertex cache is not the cause.

   What goes wrong: at the temple, a block in the game's own video memory manager ends up with a negative size. Its allocator then hands out space that overlaps other textures or lies past the end of video memory, and its defragmenter later moves that block 16 MB at a time: over the textures after it (the stripes and noise) and, because the port let those copies run past video memory into main memory, over the game's own code pointers (the freeze). On a PS3 the game's own check would have stopped there; the port did not.

   With your saves, settings and input recording the temple plays clean here, on my build and on the 0.1.0-alpha.6 Kit itself, so something on your PC brings it out, and I don't know what yet.

   0.1.0-alpha.7 keeps it from spreading: an allocation or a move of that kind is refused, and the log gets a [vram] line saying which texture it was and what it held. The launcher builds the game's code again for the new version, and its first start builds the vertex shaders before the game, a couple of minutes.

   Could you play to the temple once more with it and send a report (Report a problem in the launcher), whether it looks right or not? Those lines are what I need to find the cause. As before, please don't attach game files.
6. soldiermxdeath commented on Nov 12, 2026

   

   soldiermxdeath

   on Nov 12, 2026

   Author

   More actions

   I can confirm the graphics issue with the most recent version still there, until now I didn't have a freeze but there's missing textures.

   report-20261112-132934.zip

   
7. missingnoXXVI commented on Nov 12, 2026

   

   missingnoXXVI

   on Nov 12, 2026

   More actions

   Was planning to wait for the next release before I included my own logs and footage of graphical glitches that I have gotten from the start (Ryzen 5 5600X and a RTX 5060 on an nVME SSD)

   Or would you prefer that I just go ahead now? Or create my own post? I had to stop at the Manticore of Delphi fight because the performance instabilities were testing my patience on hard mode. And now that I know that the saves are broken until the next release, I am concerned I will just have to start all over 😂
8. rigaz29

   added a commit that references this issue on Nov 13, 2026

   VRAM guards: check the texture mover's size; precompile finds the arc…

   ...

   4dcaf60
9. rigaz29 commented on Nov 13, 2026

   

   rigaz29

   on Nov 13, 2026

   Owner

   More actions

   @soldiermxdeath thanks for the new report, it narrowed this down a lot. 0.1.0-alpha.8 is out with two fixes that came from it:

   1. Your shader precompile never ran. Your game folder has only GOWA.SELF, so the launcher decrypts GOWA.elf into the port's own folder, and the precompile looked for the game's .PSARC archives next to that copy instead of in the game folder. The temple was building its shaders while you played. Fixed in this version.
   2. The broken textures start with one streamed texture whose data range ends before it starts. The game moves it with a negative size, which also lifts the limit on how much video memory it rearranges that frame, and hundreds of bad moves follow. This version stops it at the first one: that move becomes empty, so at most one texture is wrong, and the log gets a [vram] streamed texture line saying which texture it was.

   I still can't make it happen on my PC, so please play the temple again on 0.1.0-alpha.8 and send a report from the launcher, even if everything looks fine. To update, unpack it over your old folder; the first start builds the game's code again, once.
10. rigaz29 commented on Nov 13, 2026

    

    rigaz29

    on Nov 13, 2026

    Owner

    More actions

    @missingnoXXVI thanks for asking first. 0.1.0-alpha.8 is out now: https://github.com/rigaz29/Ascension-R/releases/tag/v0.1.0-alpha.8

    Please open an issue of your own: your PC and what you see differ from this one, and one issue per problem keeps them apart. Play a bit on 0.1.0-alpha.8 first and report what is still there. Report a problem in the launcher's sidebar makes the zip to attach, and with Include my saves ticked I can load the game where it happened. Footage helps too, as a link or a short clip in the issue.

    Two things in 0.1.0-alpha.8 may already change what you saw:

    • If the launcher decrypted your copy (your game folder has GOWA.SELF but no GOWA.elf), the shader precompile never ran on 0.1.0-alpha.7. The game then built its shaders while you played, which can show as stutter. That is fixed. Your alpha.7 log would have a line starting [precompile] the vertex shaders were not built.
    • The save bug (Cannot load save game #10) only hit PCs where Windows placed the game's memory below 4 GB. If the launcher's ready check shows your profiles and the game loads them, your saves are fine and carry over. If it says "No saves yet" and the save folder (Open save folder in the launcher) has files like DAT-1 and SAVEDATA loose in it, Cannot load save game #10 has the steps to put them back in their folders, so you should not have to start over.

    To update, unpack 0.1.0-alpha.8 over your old folder; the first start builds the game's code again, once.
11. soldiermxdeath commented on Nov 13, 2026

    

    soldiermxdeath

    on Nov 13, 2026

    Author

    More actions

    Thanks for your effort. I used the new version, the game didn't freeze, but the graphics issues still there.

    report-20261112-204800.zip

    
12. rigaz29 commented on Nov 13, 2026

    

    rigaz29

    on Nov 13, 2026

    Owner

    More actions

    @soldiermxdeath thanks for the report and the screenshots, they narrowed it down a lot. The textures break before anything else goes wrong: your 20:45:10 screenshot already has a broken pillar, while the game's video memory bookkeeping only goes wrong after 20:46:42. I replayed your session here many times, with far more texture streaming than yours, and every texture stays correct on my PC. So I need two short tests from yours, each showing which half of the pipeline breaks them:

    1. In the launcher's Graphics tab, set Render thread to Off. Play the temple until the textures would usually break (or about 15 minutes).
    2. Set Render thread back to On. Make a file named test.bat next to Ascension-R.exe containing:

       set TEX_NOCACHE=1
       start "" Ascension-R.exe
       

       Start the game with it and play the same place. It sends every texture to the GPU again every frame, so it will run slower; that is expected.

    One report after both tests is fine (it keeps the last three sessions); just tell me in which test, if any, the textures still broke.