Skip to content

Roadmap

Last revised 18 October 2026.

The goal is a stock Android replacement for the Retroid Pocket Nova: write an image, add games and any required BIOS files, and play. Every included emulator should work with the built-in controls and ship with settings chosen for this device. Low latency, reliable operation and faithful console timing are the priorities.

The immediate priority is getting every included emulator running reliably and configured properly. Latency experiments must also be tested across systems before they become defaults. The work below has a next action and a completion criterion; there are no release dates attached.

BUGS.md holds observed failures and reproduction details. Hardware test results should record the image version, game, settings and logs so that a later image can be tested against the same baseline.

Current baseline: Snes9x is the SNES core; NeoCD's device tuning is finished; PlayStation has an optional PRMPT switch again. VRR spans 65–120.198 Hz for RetroArch, the launcher and Steam's gamescope, with Steam capped at 60 fps. Platform sleep now works, measured at 23 mA over ten minutes. Black frame insertion was removed with vsync and is outside the plan.

Every included emulator runs without crashes

The README records successful tests for SwanStation, Snes9x and NeoCD, playable ARMSX2 with further work pending, and a broken Dolphin path: hangs and audio problems (#463). The rest of the included systems need an explicit test record, not an assumption that a successful build means a working emulator.

  • Make a test matrix for every emulator in the supported systems list. Cover launch, controls, video, audio, saving, loading, exiting and relaunching.
  • Fix Dolphin's hangs and missing audio. Work through the other emulators with the same matrix, retaining logs and reproducible cases for failures.
  • Test repeated launches, extended play, suspend/resume and return from game guides, including Vulkan cores that lose and recreate their graphics context.
  • Separate game compatibility limits from failures in our build, launcher, configuration or device integration.

Complete when: every included emulator passes the matrix on the Nova, with representative games and no crashes, hangs, missing audio or broken return to the launcher in those tests. Remaining game-specific limitations are documented with reproductions. Passing this matrix is the regression baseline for later changes.

Configure every system for plug-and-play use

Working emulators still need good defaults. A fresh install should not require a trip through RetroArch or a standalone emulator's menus.

  • Audit each system's controller mapping, hotkeys, save paths, BIOS handling, aspect ratio, scaling, shader, rendering resolution and performance settings.
  • Choose defaults using actual games on the Nova. Test demanding games as well as easy ones; avoid a visual preset that causes missed frames or audio underruns. Add per-game overrides where a system-wide setting cannot work.
  • Make save/load, exit and game-guide controls consistent where the emulator permits it. Explain missing BIOS files and unsupported formats clearly.
  • Test both a clean installation and an update with existing user settings.
  • Keep the installation guide and format table aligned with the launcher. CHD is the disc-library format; PlayStation, PS2, Dreamcast and Neo Geo CD already enforce it. Recommend PortareOS tools / nova-push and document cartridge formats and experimental emulator exceptions explicitly.

Complete when: games launch with usable controls, correct geometry, clean audio and suitable performance without manual emulator configuration. Required files and unavoidable exceptions are documented, and updates preserve settings the user deliberately changed.

MCU and input latency

The gamepad MCU now reports every 5 ms (200 Hz), versus the vendor's 9 ms (111 Hz). rsinput.frame_rate=3 sets the scan delay from boot, with a driver fix that makes command-line parameters safe before registration. Probe and resume send the setting; UART measurements confirmed 200 Hz across two suspend/resume cycles (#466). At 250 Hz, about 19% of reports never reached evdev; the cause is not yet investigated. Values 0 and 1 stopped the pad. Faster reporting is not yet evidence of lower button-to-screen latency.

  • Compare button-to-screen latency at the vendor setting and at 200 Hz. Establish a repeatable input trigger and camera measurement (#13).
  • Trace the lost reports at 250 Hz before considering a faster default. Measure report jitter, lost inputs and end-to-end latency; check buttons and analog sticks separately.
  • Check the rest of the path through rsinput, InputPlumber and the emulator to identify where additional delay is introduced. The existing per-system traces measure pad polling to first scanline, not physical button-to-screen latency or MCU improvement.
  • Extend the 200 Hz tests under load and through repeated suspend/resume; measure power cost and stick noise before further tuning.

Complete when: a measured configuration improves input latency without lost inputs, noisy controls or resume regressions, and the baseline and results are published. Keep the vendor setting as the comparison baseline.

Audio latency

RetroArch currently defaults to 8 ms. That setting is a buffer request, not end-to-end latency. Existing measurements and the limits encountered are in the audio latency notes.

  • Measure end-to-end audio latency and audio/video alignment on the current image, rather than inferring them from the configured buffer size.
  • Test 32, 44.1 and 48 kHz playback across all emulators, including standalone ones, on speakers and wired headphones. Record negotiated buffers, xruns and audible failures during extended play and rate changes between games.
  • Locate the remaining delay before reducing buffers further. Test any driver, DSP or scheduling change against the emulator regression matrix.
  • Measure idle wakeups, power use and suspend/resume with the tuned path.

Complete when: the shipped defaults have measured latency and stable playback across the tested systems and wired outputs. Any lower-latency change must show an improvement without crackle, dropouts or power/resume regressions.

Low-latency local audio and Bluetooth audio

Evaluate two audio profiles: a tightly tuned path for speakers and wired headphones, and a Bluetooth-compatible path with the buffering and rate handling it needs. Compatibility of Bluetooth with the current latency tweaks is an open question to test.

  • Pair a Bluetooth headset and establish which current settings work and which cause failures. Record routing, codec, rate and buffering.
  • Determine whether separate PipeWire/WirePlumber profiles are sufficient or whether separate service configurations are needed.
  • Switch profiles when the output changes, including connecting and disconnecting a headset during a game. Restore local latency settings when returning to speakers or wired headphones.
  • Test pairing, reconnect, volume control, suspend/resume and failure recovery.

Complete when: local outputs retain their tuned latency, Bluetooth plays reliably, and moving between them requires no manual service restart or configuration edits. Document Bluetooth's measured latency separately.

Reduce host cost without increasing latency

NeoCD's completed tuning exposed a system-wide cost: real-time audio and GPU threads keep the X3 at its maximum clock while using about 1% of a core. The global RT clamp at zero and two coarse placement changes were measured and rejected. See NeoCD and the performance plan.

  • Measure per-thread utilization clamping or a kernel policy against the existing capture, keeping audio latency, xruns and frame delivery fixed.
  • Compare the 401 MHz gaming GPU floor with lower available frequencies on light cores; include demanding games before changing a shared default.
  • Use validated captures, three repeated runs, separate intro/gameplay windows and current configuration records for each candidate.
  • Validate Steam's VRR and 60 fps cap with a real game on the final kernel and gamescope, including focus changes and return to the launcher.

Complete when: a change reduces measured cost beyond run-to-run spread, with stable audio, frame delivery and latency across the affected systems.

Reduce the image further

Use the package inventory as the starting point (#333). Its size figures come from an older image. Recheck unused shader families, Python's remaining runtime users and GStreamer before treating them as savings.

  • Measure the current compressed image and installed package sizes.
  • Keep only the shaders in use and their transitive includes; launch every affected preset to check that the trimmed set is complete.
  • The Bluetooth agent is already C. Check remaining Python users, including gamecapture and Steam's dependency path, before proposing removal.
  • Trace GStreamer's users and remove it if the retained features can run without it. Continue removing unused packages.

Complete when: each removal has a measured before/after size and passes its affected runtime checks. Publish the resulting image size; do not trade working emulators, Bluetooth or required tools for a smaller headline number.

ARMSX2 through libretro

PS2 runs ARMSX2's libretro core in RetroArch; the standalone SDL frontend is gone. BIOS images in bios/armsx2 are copied to bios/pcsx2/bios on update, and the memory cards stay in the ps2 folder.

  • Get back what the standalone had. The first Nova test (Time Crisis II, upstream's generic CI build of the core) ran slower and with more visible glitches. Leads, from the two configurations side by side:
  • Texture barriers: the standalone forced them off (OverrideTextureBarriers = 0); the core's automatic turns them on, and a barrier is expensive on a tiler.
  • Pacing, done: the core now takes RetroArch's frame rate as the host refresh rate and lets the hand-off to retro_run pace it, one clock (patch 001).
  • VsyncQueueSize, done: 0 as in the standalone (patch 001). Step to 1 only if 0 cannot hold full speed.
  • Clocks and core layout: schedutil clocked the big cores down because the core's threads take turns within a frame (40 fps against 59.7 on performance). PS2 now runs on performance, with EE pinned to the X3 and VU and GS to the A715s. Research and measurements in documentation/emulators/ARMSX2.md.

Texture barriers are not a core option; the core keeps them in <system>/pcsx2/inis/armsx2-libretro.ini. Measure on its own. - Compare it with what the standalone did on the same games: compatibility, speed, frame pacing, input/audio latency, controls and save behavior. - RetroAchievements: the core exposes no memory to RetroArch, so PS2 has none. The standalone had them built in.

Complete when: the core matches the standalone's Nova results, or each gap has an issue.

Fix external display support

  • Reproduce the current failures with USB-C displays and adapters. Record connector detection, available modes, DRM ownership and emulator logs.
  • Make the launcher and each rendering path select and release the external output correctly, with suitable resolution, refresh rate and aspect ratio.
  • Test connection before boot, connection during use, disconnection during a game and return to the internal panel. Handle unsupported modes cleanly.
  • Verify external audio routing where available and recovery to local audio.

Complete when: the launcher and included emulators work on tested external outputs, and connection/disconnection recovers without a black screen, crash or reboot. Publish tested adapters, displays and remaining limitations.

Other work to retain

  • Suspend and resume: #658 enables CX, AOSS and DDR sleep. The 18 October test measured 23 mA over ten minutes and Wi-Fi back within a second. Measure an overnight run, repeated short cycles, an open playback PCM at suspend, charger attachment asleep and working audio after wake; retain sleep-ledger and counter evidence in power notes (#62).
  • Implement all remaining panel modes: inventory the missing modes for supported systems, including 32X, Dreamcast 240p and arcade board rates. Add the modes and any core reporting or mode-selection fixes they need; verify every mode on hardware against the core's output, including switching between games and returning to the launcher (#284).
  • Launcher: test current navigation and selection persistence, error reporting, game-guide handover and update behavior. Coordinate changes with portarelauncher; the image currently ships 0.8.3.
  • Color profiles: fix crushed dark greys and verify the result on the panel before recommending a calibrated profile again; see BUGS.md.
  • A game session that is not root: move game sessions to a dedicated user and verify access to saves, devices and launcher handover (#204).

Scope

Retroid Pocket Nova only, NTSC focused, one emulator per system and direct KMS presentation. Steam uses gamescope as its compositor exception. Other devices, PAL panel modes, black frame insertion and a desktop environment are not planned.