Benchmarks
What we measured, on what, and what it does not prove. Every row so far comes from one laptop: Intel Core Ultra 9 285H with Arc 140T, Intel driver 32.0.101.9175, Windows 11.
Headline result
Game fps, Hyrule Field, fix on
22.83
14.75 with the fix off, same scene, frame generation off
Foliage draw, GPU time
5.8ms
15.7 ms with the fix off
How to read these numbers
- Game fps is how many frames the game itself renders each second. Frames made by frame generation never count.
- Verified scene means the run's scene check passed: the benchmark confirmed the game was where it claims before counting frames. A run in the wrong place is not a result, whatever its fps.
- One laptop, one scene. Other games, scenes, drivers, power limits and machines will give other numbers.
Verified scene
The run's scene check passed: the benchmark confirmed the game was at the scene it claims before counting frames. The fix was measured off and on in the same scene.
| Machine | Driver | Game and scene | Build or change | Frame generation | Game fps | Notes |
|---|---|---|---|---|---|---|
| Intel Core Ultra 9 285H · Intel Arc 140T | 32.0.101.9175 | The Legend of Zelda: Ocarina of Time, Hyrule Field | EDEN_NX2_GM=11 (global-memory routing fix) | Off | 14.75 → 22.83 | Foliage draw (GPU time): 15.7 → 5.8 ms. One laptop, one scene. Other games, scenes, drivers and machines may differ. |
Play configuration (not an A/B)
How the builds play with the settings a player would use, frame generation on. The two rows were measured at different spots, so they are not a comparison of the builds.
| Machine | Driver | Game and scene | Build or change | Frame generation | Game fps | Notes |
|---|---|---|---|---|---|---|
| Intel Core Ultra 9 285H · Intel Arc 140T | 32.0.101.9175 | The Legend of Zelda: Ocarina of Time, gameplay, a different spot from the v20 row | Previous build | On | 19.6 | Not the same spot as the v20 row: not an A/B. |
| Intel Core Ultra 9 285H · Intel Arc 140T | 32.0.101.9175 | The Legend of Zelda: Ocarina of Time, gameplay, a different spot from the previous-build row | v20 | On | ~26.7 | Approximate. Not the same spot as the previous-build row: not an A/B. |
Static copy of data/benchmarks.json (updated 2026-11-13).
Reports from others
Third-party numbers, on other builds. We list them because they help people with these GPUs, not as benchmarks of Guarana.
| GPU | Reported | Where | Verified by us |
|---|---|---|---|
| AMD Radeon 890M class, on a ROG Xbox Ally X | About 30 fps in Ocarina of Time | Lichii issue #1433 (upstream tracker) | No |
| Intel Arc 140V | 22–24 fps during gameplay | AlfaDev23's community build | No |
How new results are measured
New rows come from the benchmark harness in tools/bench. It runs one game scene the same way on any machine and writes one JSON row per run: it drives the emulator through its debug server, restores the same save for every run, plays the scene's timed inputs, checks the scene against your reference screenshot and refuses to count a run in the wrong place. It needs Python 3.9 or newer and nothing else.
Baseline and change run back to back on the same machine, at least two windows each. A change that doesn't beat run-to-run spread is not a gain.
Add your numbers
Run the harness with the benchmark contribution guide, or send a quick device report with your game fps. Untested devices need it most.