[polackangry]

With Frame generation by at Auto, nothing on screen said whether NVIDIA DLSS Frame Generation or the built-in interpolation made the frames. It falls back to the built-in one silently: without nvngx_dlssg.dll (v2.1.0 and the current nightly copy only DLSS's runtime), without a temporal Upscaler, in the game's own menus, or at Frame generation's Auto. A player on an RTX 5060 Ti couldn't tell DLSS FG wasn't running until the log said not available (0xbad00004).

What it shows

• Frame generation by, while frame generation is on: Auto: DLSS FG, Auto: Built-in FG or Auto: FSR 3.1 FG (the FSR 3.1 interpolation at 2x). The choice itself is now Built-in FG. Off: Auto, as before.
• Its line, when another than DLSS FG makes the frames at Auto: why, for example Not DLSS FG now: DLSS Frame Generation makes a fixed 2x to 8x, not Frame generation's Auto., or no picture from a temporal Upscaler (also the game's menus), nvngx_dlssg.dll / an RTX 40 missing, more frames than its runtime makes, NGX refusing. Every line it can have is a variant, so the help box keeps its height.
• The FPS counter: its second line was game 60; now Base 60, DLSS FG (or FSR 3.1 FG, Built-in FG). "Base" for the game's own rate, the name frame generation tools use.

One name per engine, " FG", on the row and the counter. When #2665 adds FSR 4 as a frame generator, FSR 4 FG fits the same way (FgMaker and framegen_maker_name).

How

• vk_framegen.cpp: framegen_maker(). Each framegen_record notes who made its set (DLSS-G's dlss_fg_frames, fsr3fi_frame, the renderer's own passes, or none) in a ring of the last 64 sets; the menu's thread reads the counts and takes the most. Nothing noted for half a second (Off): none.
• vk_dlss.cpp: dlss_fg_skip(), the reason as a code (FgSkip in vk_dlss.h). Each return of dlss_fg_frames that leaves the set to the renderer's own names its reason (skip(FG_SKIP_...)); a set DLSS-G makes clears it; a set with nothing to blend from (a loading hitch) leaves it as it was. The build without the NGX SDK says so.
• vk_menu.cpp: tx_framegen_by, hn_framegen_by (the reasons' texts in framegen_skip_text, so they can be translated) and the FPS counter's line.
• menu_i18n.h: every new text in all 13 languages (Auto: %s, Base %.0f, Base %.0f, %s, Built-in FG, Not DLSS FG now: %s and each reason); Built-in and game %.0f, no longer written, let go. Product names (DLSS FG, FSR 3.1 FG, Upscaler) stay as the tables already keep them.
• The FPS counter's second line now goes through i18n::tr: it draws with its own Tab, past ui::'s translation, so main's game %.0f was never shown translated. Its buffer is 96 bytes, for UTF-8.
• readme.md: the row and the counter.
• Tests: test_dlssg_runtime followed the new returns and now checks that the reasons are named and cleared; present_source_crop_test checks the maker noted for FSR 3.1's set and for the built-in fallback.

Screenshots

3440x1440, Fire Temple. Frame generation 2x: the row and the counter say DLSS FG.

Frame generation Auto: the built-in one makes the frames, and the line says why.

In Polish (top, 2x) and Russian (bottom, Frame generation Auto): the counter's line and the row's line translated.

Testing

In game, native Windows 11, RTX 5060 Ti 16 GB, driver 617.42, a build of this branch on main (51da046) with NVIDIA's SDK and nvngx_dlssg.dll 310.9.16, DLSS Quality, a save in the Fire Temple. The row's text read through the control bridge (option get frame_generation_by), the screenshots of the window:

Case	Row	FPS counter
2x, in play	Auto: DLSS FG	Base 63, DLSS FG, 126-129 FPS
4x, in play	Auto: DLSS FG
Frame generation Auto	Auto: Built-in FG + why	Base 71, Built-in FG, 240 FPS
4x, the game's pause menu	Auto: Built-in FG
The title screen	Auto: Built-in FG
Built-in chosen, 3x	Built-in FG
Off	Auto

No fault through 6x, 4x, 2x, Auto and Off. Menu language Polish: Auto: Wbudowany FG, Bazowe 63, DLSS FG, Teraz bez DLSS FG: ...; Russian: База 69, Встроенный FG and its line. The log agrees: of the last 600 presents 600 made by DLSS-G at 2x.

Tests: test_menu_i18n passes (it failed on the first push: the tables still had Built-in and game %.0f). The checks that read vk_menu.cpp, vk_dlss.cpp, vk_framegen.cpp, menu_i18n.h or the readme, run here: all pass but one; test_input_background fails the same on main on this PC (it was in my earlier PRs' list of machine-only failures). A row for the run is in docs/tested-configurations.md.

Not covered: FSR 3.1 FG on screen (no AMD GPU here; its maker is checked in present_source_crop_test), the reasons for an RTX 30 card or a missing runtime on screen (read from the code paths, and the missing runtime from this PC's earlier log). The race when stepping the multiplier up is #2658's, not touched here.

🤖 Generated with Claude Code

[Joey-Magnus]

I merged this into current main and ran every check that reads vk_menu.cpp, vk_dlss.cpp, vk_framegen.cpp or menu_i18n.h (33 checks, test_menu_i18n, test_dlssg_runtime and present_source_crop_test among them). All pass, and CI is green.

The threading is fine. note_maker only runs on the recording thread, at vk_framegen.cpp:898 and in the Auto branch. The ring itself isn't shared, and the menu thread only reads the relaxed maker_count atomics and the timestamp. Each early return of dlss_fg_frames now names its reason, and the !blend and closed returns deliberately leave the old one, which matches the description. The screenshots and the table of cases on an RTX 5060 Ti cover what a player sees.

One heads-up for the maintainers: this conflicts with #2665 (FSR 4 / FSR 3.1 as Frame generation by choices) in vk_framegen.cpp, vk_menu.cpp, readme.md and tests/test_framegen_preset.py. Whichever lands second needs a rebase. FgMaker and framegen_maker_name are already shaped for an "FSR 4 FG" entry, so it should be mechanical. It merges cleanly with #2658.