007 First Light on AnyPS5

Screenshots

No screenshot of this game running has been posted in the project's threads yet. Testers who reach a new screen usually post one in the discussion linked on the right.

Where it stands

007 First Light starts on Linux and Windows test builds, creates its window and swapchain (1500x844), and keeps compiling shaders, but nobody has shown an image from it yet. On plain current main it does not present a single frame. On a hand-built Linux stack (current main plus many unmerged PRs and a personal fork), it presents 140 to 168 frames and then stalls waiting on the GPU. The testers have not checked what those frames contain. On Windows (RTX 5070 Ti) the window stays black.

The main work so far is a long list of start-up stops, tracked in one issue (#2523) by SleglJan, who counts about 19 stops in order. Early ones are now merged: missing network and SSL functions, a share-play call, an unusual shader. Others are still only in forks or open PRs: the 16-bit-float screen format the game uses (#3134), a file-loading call that deadlocked, a shader trick the translator rejected (cube-map instructions), and a very complex shader that needed more room to translate.

The game is very heavy on the shader translator. It uses very large 'bindless' image tables, so around 50 of its shaders still fail to translate on main because of size limits; a redesign of those tables is being tested on a separate branch. Translating shaders also eats CPU: one merged fix cut the time from launch to first flip from about 23.5 s to 15.6 s on a warm start.

More recent steps are on test branches. One run got past a depth-surface error (a memory area reused for a different kind of image) and continued compiling, but still showed black. Another cross-test ran out of memory, with a watchdog stopping it. A text-to-speech call that the game calls without starting up first is the next small stop.

What is stopping it

• No visible content: the window stays black on both Linux and Windows test builds. Thread
• On Linux the integration stack stalls after 140-168 frames at a GPU wait that times out (queue-0 WAIT_REG_MEM_64). Thread
• Around 50 shaders fail to translate on main because they use more bindless image tables than the translator allows; a redesign is only on a test branch. Thread
• The game's 16-bit-float display format is not yet merged (open PR). Thread
• Start-up spends huge CPU time translating shaders; one speed fix is merged, more work is needed. Thread

How it got here

1. 9 Oct 2026First fix PRs opened for the game's stops (async file-loading submit, cube-map shader instructions, bindless image stores); they were closed during the project's PR intake pause and carried on as branches. Source
2. 10 Oct 2026Tracking issue opens, listing each start-up stop in order and the PR that passes it. Source
3. 10 Oct 2026Missing SSL and network functions are added, so the game loads at all. Source
4. 10 Oct 2026A share-play start-up call that the game makes with a null heap is accepted. Source
5. 10 Oct 2026Speed fix merged: time from launch to the first flip drops from about 23.5 s to 15.6 s on a warm start. Source
6. 10 Oct 2026On an integration stack, the Linux build creates the window and presents 140-168 frames before stalling. Source
7. 10 Oct 2026A Windows cross-test passes the depth-surface stop and keeps compiling shaders; the window is still black. Source
8. 10 Oct 2026PR opened to present the 16-bit-float display buffers the game registers. Source

Tested hardware