Ghost of Yotei in AnyPS5: open-world evidence and frames-per-ten-seconds context
ItsOOGWAY · PC guides & recorded tests
Ghost of Yotei in AnyPS5: open-world evidence and frames-per-ten-seconds context
Ghost of Yotei has reached the open world in AnyPS5 contributor test-branch reports. This update explains the milestone, the reported roughly 0.5-0.6 FPS world performance, and the newer queue and stability notes.
The key checkpoint: The reported comparison is frames presented over ten seconds, approximately 0.419 to 0.499 FPS. Preserve that unit and the October 22 test-branch context.
How to read the evidence
- Keep the date, source revision, scene and measurement unit with each claim.
- Contributor reports and earlier local gameplay are attributed evidence, not a new benchmark.
- Use the primary documents below to check later changes.
Research date: 22 October 2026. The walkthrough discusses attributed contributor reports and experimental branches as checked then. It is not a current compatibility verdict.
Resources from this video
Links, downloads & documentation
Primary documents, attributed reports and every source URL from the video description are collected here. Preserve the date and measurement context when comparing later updates.
Watch the complete video
Open this video on YouTube ↗. Screenshots below link to their exact recorded moments.
Walkthrough 01
Read the dated branch evidence
I'm reviewing the developer discussion and the branch notes checked on October twenty-second. These are contributor results, not a benchmark I ran myself. The first thing to understand is that this is a work-in-progress branch called Yotei best. It is based on upstream AnyPS5, with additional work and diagnostics. The branch name is not a guarantee that it is suitable for normal play.
Walkthrough 02
Understand the open-world milestone
The history in the recording shows why the open-world captures matter. The reports move through blocked startup, black images, the introduction and early screens before reaching world scenes. An error disappearing does not mean the game is finished; it exposes whatever comes next. Seeing the actual landscape is evidence that more of that chain has been crossed than a successful build oRAMenu alone could establish.
Walkthrough 03
Convert the reported time unit correctly
Now for the speed claim, because this is very easy to misread. One same-session comparison in the discussion improved from four point one nine to four point nine nine presented frames over ten seconds. That is about nineteen percent more frame output. It is not forty-nine FPS. Dividing by ten gives roughly zero point four two becoming zero point five FPS. A real relative improvement can still leave the absolute performance extremely low. Both parts of that statement belong on screen together.
Walkthrough 04
Keep warm-cache observations separate
The same report says an early world-frame milestone arrived sooner with warm caches. That is encouraging for repeated testing, but a warm-cache run is not the first launch on a fresh setup. The notes warn that a cold run can spend thirty to sixty minutes reaching the world. A shader-cache version change can make that preparation necessary again. We should not compare those two loading conditions as though they are the same test.
Walkthrough 05
Distinguish older warnings from newer branch notes
And there is a newer development after the older warning visible in this recording. That warning says the extra Vulkan queue was hanging and should be left off. The later branch notes describe a change to which queue family handles that work. They report four clean world runs, up to sixteen minutes each, using a compute-only queue family. The latest overview puts world performance around zero point five to zero point six FPS on the tested RTX thirty-eighty laptop.
The new queue comparison is explicitly provisional. It is progress in a small test set, not proof that the hang is solved everywhere. Why does that distinction matter? Because letting work overlap only helps when the results remain correct. If one operation reads something before another operation has finished writing it, extra parallelism can produce a wrong image or a hang. The goal is not simply to turn more work on. It is to make the ordering correct, then demonstrate that the added overlap is useful.
That is why a stability result can be as important as a percentage gain.
Walkthrough 06
Assess reproducible progress without a normal-playability claim
Memory is another part of the investigation. A separately tracked graphics issue describes a resident color target storing data back over CPU writes made after the draw. In plain language, a later copy can overwrite newer information if ownership and ordering are wrong. It illustrates why changing a graphics path can affect correctness, not just speed. It does not prove that this one issue explains every problem in Yotei. There is also work on the shader translation itself.
A merged shader change restructures certain address-lookup functions through one return. Details like that can look small next to a game screenshot, but the translated program must preserve the original behavior before its output or performance can be trusted. A larger checklist and a faster counter do not replace that requirement. So how far are we? The branch reports reaching the open world from a save, with the HUD and a quest, on its specified test machine.
It also reports very low world frame rates and unresolved limitations. That is enough to say the project is producing important new evidence. It is not enough to announce a complete game, a thirty-FPS breakthrough, or an upcoming official release. The next useful milestones are sustained movement through more of the world, correct rendering, controlled performance comparisons, and fixes that can be reviewed independently and carried upstream. Any one capture is a moment. Repeating the result across builds and machines is what turns it into reliable progress.
Checks for interpreting the report
| Situation | Next useful check |
|---|---|
| A current source differs from the recording | Keep the recorded snapshot date. Check current revision and measurement rules before comparing. |
| A screenshot appears to prove more than the report | Read the full source, the test environment and the untested limitations. |
| A headline number sounds decisive | Check its numerator, denominator, time unit and the specific outcome it measures. |
Comments
Post a Comment