Astro Bot on AnyPS5

Screenshots

Posted by testers in the project's threads; the date links to the post. Click to enlarge.

Where it stands

Astro Bot is the furthest along of the games tracked here. On a testing branch (called `astrobot`, maintained by oneandonlydean, with extra Windows fixes from other contributors) the game shows the PlayStation Studios logos, the title screen, the save menu and the intro cinematic. It then plays the Crash Site tutorial, flies out to the galaxy map and loads levels such as Sky Garden, Gorilla Nebula and Aerial Garden. A DualSense controller works, including over Bluetooth on Windows. Sound plays.

Speed is the big problem. Reports range from about 3 FPS in most levels to 11 FPS in the Crash Site on a very fast Windows PC with a locally modified build, and 17 to 19 FPS on a Radeon RX 9060 XT with stutters. Title and menu screens run faster, from 9 up to about 45 FPS depending on the PC. The game's main graphics work is done one draw call at a time on a single thread, and that thread is the limit. Cards with 8 GB of video memory can also run out of memory in heavy levels. A tester on an 8 GB card ran about 80 of the game's 125 levels in short tests, and the thread lists out-of-memory crashes and a few other level-specific stops.

Plain current main is behind the branch. Many of the branch's graphics fixes have been merged into main one by one, and recent main reports show startup problems being cleared quickly: a Windows crash in the audio thread, a corrupted command buffer on AMD/Linux (fixed 2026-10-10) and a shader that could not be prepared on NVIDIA/Linux. Even so, the latest main reports on Linux (RTX 3090) show the game still compiling shaders with no window or frame at 5 minutes, and an RX 9070 XT run hits a graphics-card hang (a mesh-shader draw that never finishes) after 70 to 140 seconds. A Windows RTX 3060 Ti user hit an early shader crash at startup (reported 2026-10-09, closed as a duplicate of a problem the shader-deferral fix addresses). No one has shown gameplay on plain main, so for now the branch is the only way to see the game running.

There are two regional releases in the reports: PPSA21564 (version 01.007.000) and PPSA21567 (01.018 and 01.019). Both reach the same point on the branch. The Linux builds of the branch needed a special libc build option for PPSA21567 because that release ships its own C library. You also need to supply your own legally dumped game files.

What is stopping it

• A mesh-shader draw (40 workgroups) never finishes and hangs an AMD RX 9070 XT (RADV) 70 to 140 seconds in, forcing a GPU reset. Cause not found yet. Thread
• On NVIDIA/Linux plain main, a small vertex shader that uses a barrier fails validation at its first draw and aborts the game. Thread
• Speed: the single graphics worker thread is the bottleneck, giving about 3 FPS in most levels on many PCs; around 1000 render-pass breaks per frame were measured on one Windows PC. Thread
• 8 GB graphics cards run out of video memory in heavy levels; mitigations (smaller texture caches, skipping the top texture detail) were proposed but are not on main. Thread
• Sampler border colours from a table are not implemented; this ended both Armadillo challenge levels in testing. Thread
• A shader with a 2D comparison sample on a plain 2D image has no prepared mode, so the game stops about 80 seconds in on plain main (RADV); a one-line workaround was tested locally. Thread

How it got here

1. 29 Sep 2026First Astro Bot fixes land in main; the PlayStation Studios intro video renders for the first time on Linux (RTX 4090). Source
2. 2 Oct 2026First Windows reports on a testing branch: the game starts, loads its data, shows the Sony logo and the opening cinematic at under 2 FPS. The same day the branch author posts a Linux video from the intro through the Crash Site tutorial to Sky Garden. Source
3. 3 Oct 2026First Windows run in a real level: the Crash Site tutorial is finished and the first Gorilla Nebula level loads, at 5 to 10 FPS (about 0.1 FPS once the level appears). Source
4. 4 Oct 2026A second release (PPSA21567) gets from the logos through the intro cinematic to the Crash Site tutorial on Linux with an RTX 3070; the 8 GB of video memory is found to run out in heavy levels. Source
5. 5 Oct 2026A fix for the Windows audio-thread crash is proposed (#637). With it, a Radeon RX 9060 XT PC goes past the intro into gameplay at 17 to 19 FPS, and a Windows Intel PC reaches the Crash Site. Source
6. 5 Oct 2026Linux user tests 82 of the game's 125 levels in 5-minute runs on an 8 GB card (details of which fail are in the discussion). Source
7. 8 Oct 2026Windows tests of a level (Aerial Garden) reach about 3 FPS; the bottleneck is traced to render-pass breaks and the graphics worker thread. Source
8. 10 Oct 2026A fix for a corrupted command buffer that stopped the game about 20 seconds in on AMD/Linux (#1458) is merged into main. Source
9. 10 Oct 2026Main now defers shaders it cannot translate instead of crashing at startup (#1836); NVIDIA/Linux runs no longer abort there, but the next stops appear. Source
10. 10 Oct 2026A Windows contributor reports raising the Crash Site level from 3.2 to about 11 FPS with a local stack of speed-ups; parts are offered as pull requests. Source

Tested hardware