Repository navigation
Releases
PS5_Lichii v1.10.00 (pre-release)
Fritchii v5.5, our PS5 renderer work back in place, and Switch 2 games started from their own files, with no .eden file to write.
| Mario Kart World race | Average | Median | 10% low | 1% low | Worst frame |
|---|---|---|---|---|---|
| Fritchii v5.5 alone | 35.4 fps | 37.7 fps | 22.8 fps | 9.7 fps | 1,157 ms |
| v1.10.00 | 34.7 fps | 36.4 fps | 24.5 fps | 15.3 fps | 290 ms |
Pre-release. It installs as ProsperoEden (
PPSA99008) and replaces a ProsperoEden already installed; settings, saves and game files under/data/prosperoedenare shared.
No piracy. No games, keys or firmware are included, and they never will be. Use only games, keys and firmware you dumped yourself from hardware you own.
New
- No
.edenfile needed. Copy a Switch 2 game's.xci,.nupor.nspintoroms/(or a folder of its own there) and it is in the Library. A.nupis always a Switch 2 game; an.xcior.nspis one when its title ID says so, and Switch 1 games keep starting as before. A.edenfile is still read, for a loose dump or for options a file alone cannot carry. - Fritchii v5.5 (from v5.3): AMD drivers now prepare all of a game's recorded pipelines, and several Windows-side fixes. Frame generation (MihawkGen) keeps working: v5.5 switches it off in Fritchii builds, and the PS5 build keeps it under its own setting.
- Our PS5 renderer work is back. Since the move to Lichii, a moved line in a source file made the build drop whole files of our changes without a word. They are re-targeted to the current code: guest writes no longer wait for the GPU thread's cache locks, pages ahead of a write are marked at once, exact conditional rendering, a barrier after each compute dispatch, and the per-draw timers. Where upstream now does the same, ours gave way (its compute barriers, draw submission policy, flush-area fast path and blitter memory release).
Fixes
- Fewer long stalls in races: Mario Kart World's 1% low 9.7 → 15.3 fps, worst frame 1.16 s → 0.29 s.
- Ocarina of Time looked too bright with an earlier attempt of this work (a read fast path that missed GPU-written buffers); the picture now matches Fritchii's.
Known issues
- Ocarina of Time draws no flowers on the PS5.
PS5_Lichii v1.00.00 (pre-release)
The first pre-release of PS5_Lichii: Switch 2 games on the PS5 Pro, on Fritchii v5.3. It is built from the code and settings of the measured run 50: Ocarina of Time in Hyrule Field, moving, frame generation off.
| Window | Average | Median | 10% low | 1% low | Worst frame |
|---|---|---|---|---|---|
| Whole gameplay stretch | 29.4 fps | 28.2 fps | 18.9 fps | 11.5 fps | 624 ms |
| Last minute | 32.7 fps | 33.9 fps | 22.8 fps | 18.8 fps | 58 ms |
Pre-release. Expect hitches (250–590 ms at area changes) and games that do not start. It installs as ProsperoEden (
PPSA99008) and replaces a ProsperoEden already installed; settings, saves and game files under/data/prosperoedenare shared.
No piracy. No games, keys or firmware are included, and they never will be. Use only games, keys and firmware you dumped yourself from hardware you own.
New
- Fritchii v5.3 is the emulator (Lichii, Eglantine and the Switch 1 Eden base are gone from the build). Its AMD ASTC decoder no longer writes past an image's edge, which hung the GPU.
- Switch 2 games from a single file: XCI game card images read in place, NUP containers (newest update over the base), NSP packages, or loose dumps, each named by a
.edendescriptor. - Memory kept free for area changes: cold guest memory packed into a compacted slab store, a background scanner that keeps 2.5 GiB free, two-tier emergency packing, ASTC conversion in 64 MiB batches, allocator recovery, and the heap in 32 MiB pieces.
- Compute pipelines rebuilt on a worker thread, from Nolar-Hack/Lichii.
- Defaults from the measured run: Fast GPU, Compact memory and Block list on. MihawkGen frame generation is available in Settings > Video and off by default.
- No update check: homebrew.page's entry for
PPSA99008is upstream ProsperoEden, whose releases would replace this build.
Fixes
- No out-of-memory stops on the measured route (they ended every earlier long run in Hyrule Field).
- Worst frame 0.62 s instead of 2.3 s.
- A race between two threads mapping the same page of guest memory.