01 / Method
The same load. Two APIs.
KiscPort Probe 1.3.0 compared M3G and Mascot Capsule Micro3D v3 with the same mesh, screen area, rotation and measurement window. Test 7 measures Mascot; test 9 measures M3G. Full-frame FPS is used because part of the W995's M3G work occurs during the screen flush.
Hardware · data reported from Probe screenshots| Load | W995 / M3G | W995 / Mascot | W900i / M3G | W900i / Mascot |
|---|---|---|---|---|
| 768 triangles | 27.6 | 49.0 | 52.9 | 29.4 |
| 3,072 triangles | 10.7 | 21.4 | 22.4 | 10.9 |
| 384 triangles · 16 px texture | 30.6 | 54.6 | 62.8 | 33.6 |
| 384 triangles · 256 px texture | 25.9 | 50.0 | 62.5 | No data |
| Screen filled 8 times | 7.7 | 21.5 | 51.0 | No data |
Mascot was approximately twice as fast on W995 and twice as slow on W900i for the documented loads. The phones were identified by the owner; confirmation using the report's platform= field is still pending.
02 / Decision
Fidelity is a requirement, too.
The port uses vertex color multiplied by textures, partial alpha, atlases with more than 256 colors and per-pixel depth. Mascot v3 has different constraints: per-face color or its own lighting, per-polygon blending and indexed textures.
M3G remains the renderer for the faithful version. Mascot is a candidate for a possible visually adapted variant on devices where its results are favorable. That variant is neither finished nor published.
03 / Hypotheses and limitations
What these figures do not establish.
- These are synthetic scenes, not playthroughs of the port. A 2× advantage does not automatically carry over to the game.
- FreeJ2ME-Plus data cannot compare physical Mascot drivers: its implementation is different.
- Fidelity was studied using API capabilities and resources; no pixel comparison was performed on a phone.
- Possible affine texture mapping in Mascot remains a hypothesis; it was not verified on hardware.
- Test 8 in Probe 1.4.0 proposes a paired comparison; its physical execution is pending.