Settings & performance
The core ships about forty options. Almost none need your attention. Two of them interact in a way that can make a slow game slower.
Leave the defaults alone
Accurate blitter, stock clocks, every experimental timing model off, HLE BIOS, and per-title enhancement defaults on, which gives roughly twenty recognised games a visual upgrade that was verified safe for that specific title. If a game boots and plays, you are done here.
Four settings are worth knowing about: Blitter, Internal Resolution, DSP Idle-Loop Fast-Forward, and for CD games CD Boot Mode. Everything else is a diagnostic, a controller, or networking.
Where frame time goes
Profiling put the cost in one place, and not where intuition puts it.
| Subsystem | Share of frame time |
|---|---|
| DSP | 33-67%, the top cost on every title measured. 50-67% on the DSP-heavy ones (Iron Soldier 67%, Alien vs Predator 63%), about 33% on Skyhammer |
| GPU | large, title-dependent. Around 24% on Skyhammer |
| Blitter (fast) | 5-17% |
| Blitter (accurate) | up to about 34% on Alien vs Predator |
| Object Processor | about 1% |
| 68000 | 0.7-2.6% |
The DSP sits in a 3-5 instruction wait loop for 90-99.7% of every frame on the titles measured, while the emulator interprets 416k-590k opcodes per frame doing it. That is more than real silicon could retire in a frame (about 443k cycles at 26.59 MHz over a 60.05 Hz field), because pipeline hazards are not modelled. So it is interpreter work, not a hardware budget, which is exactly why skipping it is free. The 68000 and the Object Processor are noise.
Host profiles on desktop arm64, not device measurements. No on-device Apple TV capture exists yet.
When a game runs slow
In order. Stop as soon as it is fast enough.
- Turn off Per-Title Enhancement Defaults. If your game is recognised, it is rendering at 2x and/or full-precision Gouraud right now. 2x and true color together measured about 30% of the frame on Alien vs Predator; its shipped preset is 2x only, so disabling it recovers that share rather than the whole 30%. On a low-end device this is step one, not step three.
- Internal Resolution 1x, True Color off. The manual version of step 1, if you want to keep per-title defaults elsewhere.
- Turn on DSP Idle-Loop Fast-Forward. The biggest lever, and it is off by default. Nobody gets it by accident.
- Blitter to Fast. Measured 1.21x typical, up to 1.58x on the title that gains most. It breaks some games, with wireframe artifacts and missing geometry, so switch back the moment anything looks wrong.
- Only then consider overclocking, and read the next section first, because it can make things worse.
There is no frame-skip option and no internal speed limiter. Pacing is entirely your frontend's job.
What cancels idle-skip read before overclocking
DSP Idle-Loop Fast-Forward disables itself when any of these is in effect:
| Setting | Value that disables idle-skip |
|---|---|
| RISC (GPU/DSP) Clock Scale | anything other than 1x |
| DRAM Timing | enabled |
| GPU Pipeline Timing | enabled |
| Blit Memoization | enabled or verify |
It stays active under M68K Clock Scale at any value, and under Blitter Bus Timing. Neither of those turns it off.
Overclocking the RISC to make a slow game faster can make it
slower. Setting RISC Clock Scale to 2x gives the
GPU and DSP twice the cycles, and simultaneously forfeits a 66-87%
reduction in DSP work. On a DSP-bound title you have traded the bigger
win for the smaller one.
Since v3.5.1 the core tells you. Enable idle-skip
alongside any of the above and one [perf] line lands in
your frontend log naming the exact setting suppressing it. It warns
only. Your settings are always honoured, never overridden.
Jaguar CD
CD Boot Mode defaults to HLE, the fastest and most broadly compatible path, and it needs no BIOS files. Switch to real BIOS if a disc misbehaves: it is a different implementation of the same boot sequence, and a handful of titles work under one and not the other.
Full guide
This page is the short version. The complete guide lives in the repository: every option, a restart-required table, a symptom to setting troubleshooting table, and the evidence behind each figure above. settings-and-performance-guide.md.