mirror of
https://github.com/tangerineturtle1/NX2navi.git
synced 2026-11-18 13:09:57 -04:00
A review of the DLSS and frame generation paths against NVIDIA's guides (the DLSS programming guide 310.6.0, the DLSS-FG guide 310.7.0, Streamline 2.14.16's guides and headers). What follows them: the texture bias (log2(render width / output width) - 1 on top of the game's own, at each change of the mode), the jitter (Halton, 8 x (output / render)^2 phases and 16 at least, in render pixels within half a pixel, the scene drawn with it and the motion vectors made without it), the creation flags (HDR, vectors at the render size, auto-exposure, depth not inverted) and the constants Streamline gets (unjittered matrices, the jitter in pixels, the vectors' scale of one over the render size). What did not: vk_dlssg.c holds a change of generation's state for FG_SETTLE_FRAMES frames, so for up to 12 frames after the depth, the motion vectors or the camera went away (a menu, a fade, a load) generation stayed on and nx2sl_frame set no constants and tagged nothing. Streamline's guide asks for the constants at every frame and for null tags where depth or motion vectors are not valid; it logged "Unable to find 'common' constants for frame" and "Failed to find global tag 'kBufferTypeDepth'" and used the frame before's (in every play log with frame generation). Such a frame now gets the last frame's constants with reset set and the three tags as null. Not run in a game yet: nx2navi_dlssg.dll builds and exports its 15 functions. What the guides leave open: the sign of the jitter offsets against the shift of the picture, which NVIDIA says to find by trying, and the bias reaches only the tiled, trilinear samplers, not every texture of the scene. For both there is a test switch now, read once: NX2NAVI_DLAA_JITTER_TEST=<x>,<y> tells the upscaler the jitter times two factors, and NX2NAVI_DLSS_LOD_BIAS_TEST=<bias> puts a bias in the rule's place. Neither was measured: the runs that would do it start the game. The line about the presents a second (the commit before) came at 59 a second under a limit of 60: it comes under 56 now. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
13 KiB
13 KiB
Install or update
- Already have 0.6.0 or newer: start the launcher; it offers 0.11.0, updates itself and keeps your saves and settings. This update translates the game's code once more (for the ocarina recorder's fix) and prepares every shader again with the fixed shader translator: a few minutes, once.
- On a 0.10.2 preview: the same; the launcher offers 0.11.0 on either update channel. 0.10.2 was this release's name while it was tested, and no final 0.10.2 follows.
- New install: download
NX2navi-Setup-0.11.0.exebelow, put it next to your.nupdump of the game and start it. Windows SmartScreen warns on the first start because the setup is not signed: choose More info and Run anyway. The launcher checks your game files, builds the game (10 to 25 minutes, once), prepares every shader and starts it.
Requirements: as for 0.10.0: Windows 10 or 11; an x86-64 CPU with SSE4.2; 16 GB of RAM; about 30 GB free for the
game and 1.5 GB for the launcher; a GPU with Vulkan 1.3 and VK_EXT_external_memory_host (current NVIDIA and AMD
cards).
Changed
- The Game tab's Testing settings keep 0.10.1's defaults: CPU helper threads Automatic
(
replay_workers = auto), Start next frame early On (tail_early = 1), and Helpers work ahead and Helpers take small jobs off (replay_pipe = 0,replay_small = 0). The last two stay experimental: a tester's measurement on an RTX 3090 found them slower at four of eight places and no more than 7 % faster at the others. - Automatic texture detail is more careful with video memory (
texture_detail = auto): Full now starts at cards of 12 GB (10 GB cards get Reduced, as 8 GB cards do), and integrated graphics never get Full, because their video memory is the PC's own memory. With a card filled to the room of an 8 GB one, a tester measured 29.6 fps with Full and 42.8 with Reduced on the same tour, and a handheld that got Full closed when a save was loaded. A texture detail you chose yourself stays as you set it. - Less video memory held in reserve. The game lets go of textures as you move and takes a few of them back; the
ones it let go of were kept for three minutes and could hold a quarter of the card's video memory (up to 8 GB), and
what they held at their fullest stayed in use for the rest of the session. They are now kept for one minute and may
hold a thirty-second of the card's video memory, 384 MB to 1 GB (
tex_released_hold,tex_released_mb). Measured on a run through Hyrule Field at 5120x2160 with DLSS Performance on an RTX 5090: 13.2 GB in use on average where it was 16.7, 13.8 at most where it was 17.9, at the same frame rate. An 8 GB card keeps 384 MB of such textures where it kept up to 1.9 GB. - Textures the game writes again with the same content are no longer decoded a second time. Running through
Hyrule Field the game rewrites texture memory with the bytes it already holds; textures over 64 KB were then
uploaded and decoded again each time, one of them 29 times in five seconds. On the same run that was 49,660 second
uploads in three minutes, now 214, and the graphics card's upload work fell from 0.41 to 0.14 ms a frame
(
tex_same_big = 0brings the old behaviour back for a comparison). - The tone mapped HDR picture has its settings back in the Screen tab, under HDR output on its tone mapped
choice (not Native) and the three brightness rows: Tone mapper and Gamma correction, then the rows of the tone
mapper chosen, with Exposure, Highlights, Shadows, Contrast and Saturation among them. Each change shows
at once. Left alone they give the picture HDR had from 0.9.0 to 0.10.1, true blacks included. Gamma correction now
counts with every tone mapper; its Off shows the dark tones lighter. Lines for these settings that your
nx2navi.ini still holds from before 0.9.0 (
hdr_tonemap,hdr_exposure,hdr_gammaand the othersnx2navi.ini.examplelists) count again: remove them for the default look.
Removed
- The low-end GPU mode is gone from the settings menu's Game tab, with the rows under it: it no longer gained
frame rate, and the mesh shader fallback now covers the cards it was written for.
low_end_gpu = 1in an existing nx2navi.ini does nothing, and the log says so once; the game then runs as it does for everyone else. The settings the mode turned on can still be set one by one in nx2navi.ini (nx2navi.ini.examplelists them); a line for one of them that is already in your file keeps counting.
Fixed
- See-through ground (issues #569 and #824). 0.10.0 and 0.10.1 shipped an older shader translator, whose shaders could not reach the newest texture slots, so ground and walls could turn see-through after some play. 0.11.0 ships the translator built from its own sources, prepares the shaders again with it, and no longer uses translations kept from the old one. The release tools now refuse a translator that was not built from the release's sources.
- AMD Radeon RX 5000 and 6000 cards losing the device (the game ending with code 72): on AMD's driver these cards
now take the mesh shader fallback by default (
mesh_fallback = auto), and a lost device is noticed and reported sooner. Not yet confirmed by players with these cards: please report how it goes. - Sky-coloured holes in the ground where grass is still loading, on cards that take the mesh shader fallback: a
patch of grass or leaves is now left out whole until it is ready, where its depth alone was drawn before
(
mesh_fallback_hold = 0brings that back). Not yet confirmed by players with these cards either. - The game not starting after a settings change on a laptop with two graphics cards ("no usable Vulkan device", code 70, until nx2navi.ini was edited by hand): with DLSS Frame Generation on and the Graphics card setting on the card that is not NVIDIA's, the device could not be made. The game now starts on the NVIDIA card, and where there is none, or frame generation cannot be set up, it starts without frame generation and the log says why.
- A crash on GeForce GTX 10 cards that had the low-end GPU mode on (the game closing inside NVIDIA's driver at the first grass or foliage of an area, reported on a GTX 1050 Ti, a GTX 1070 and a Titan X): the mode turned on an older way of drawing the game's mesh shaders that this driver does not survive. NVIDIA cards without mesh shaders now always take the mesh shader fallback, and the mode itself is gone (see Removed).
- Flashing or white foliage on cards that take the mesh shader fallback (reported on a Radeon RX 580): the numbers each such draw goes by were written in an order the graphics card was free to change, and about half the draws got the wrong ones. Each is now written exactly once (by Tobim6).
- The whole scene flashing for single frames with Helpers take small jobs on (
replay_small, with Helpers work ahead; the interface stayed as it was), under DLSS or FSR 3.1: a helper thread could name an earlier image of the frame as the scene after a later one had been named, and the upscaler then took that image for one frame. At one place in Kakariko with both settings as a tester had them, 51 and 52 frames flashed in two runs of 100 seconds, and none with the fix. Both settings stay off by default. - One long frame about 30 seconds into every session (55 to 82 ms, the longest frame of play in the two sessions
it was found in): the first memory report asked Windows for a summary of the game's heap, and that summary holds
every allocation of the game while it is taken. It is now taken before play only (
mem_census = 1asks for it in play too, for a diagnosis). - Video memory staying over the card's budget once the frame rate falls under 60 fps (reported on an RTX 4070
SUPER at 2560x1440 with DLSS and frame generation, where the frame rate went from 63 to 21 fps): a texture the game
has just let go of is kept out of the video memory pressure's reach for 2 seconds, counted in frames. Under 60 fps
the count did not shrink, so the 2 seconds were 4 at 30 fps and nearly 6 at 21, and the textures waiting them out
held 2.1 GB on that card. The wait is 2 seconds at any frame rate from 15 fps up now. On our own PC at a 30 fps
limit those textures held 209 MB on average where they held about 400. Not yet confirmed on a card that is really
short of video memory: please report how it goes (
tex_rate_low = 0brings the old count back). - DLSS Frame Generation on frames without a scene (the first frames of a menu, a fade or a load): generation stayed on for up to 12 such frames without the camera data and the depth and motion buffers NVIDIA's library needs, and the library logged errors and went on with the data of the frame before. Such a frame now tells it to start anew and gives it no buffers, as NVIDIA's guide asks. Found in a review against NVIDIA's guides for DLSS and frame generation, which the texture mip bias and the jitter were found to follow; not yet seen in a game by us.
- Crashes and hangs found in a review of the whole runtime; none was seen in our own play, and each fix has a check
of its own:
- a crash when the game read memory the lazy write-back was just handing to the GPU (NVIDIA and Intel cards);
- a game thread reading a frame's record another thread had just freed, when two threads submitted at once;
- a race between a system event's signal and its destruction, and full event tables (vsync events after 16, host events), each of which ended the game;
- a movie ending the game when memory for a picture ran short;
- game sessions closed while another thread still used them;
- after 15 to 66 hours in one session, textures the game writes stopping to update (the counters behind them are now 64-bit);
- audio settings the game sends that our audio renderer did not accept ended the game; now the one effect is left
out and one line goes into the log (
audio_strict = 1brings the old behaviour back for bug reports); - a stuck shader translation no longer holds the game for four minutes;
- a save file could be closed while another thread was still reading it.
- Crash reports and exits: a second crash no longer cuts the first one's report short; a stack overflow writes a
report; an error or a lost device ends the game without running every library's shutdown, where DLSS's could crash;
HD rumble stops at every end; the end of the log is written even past its 64 MiB cap. A hang watch writes a report
to the log when no frame is shown with the window in front, or the GPU does not finish, for 20 seconds
(
hang_watch_s, 0 turns it off). The launcher explains code 72 (the graphics device was lost) and shows the game's error line after a restart and from the Steam shortcut.
Frame pacing and RivaTuner Statistics Server
- RivaTuner's profile is put back when the game closes, before the slow save of its pipelines, and no frame drawn
after that writes the game's limit into it again. After a crash or a Task Manager kill, the next start puts it back
from
rtss-profile.txtnext to the game, unless you changed the profile since. If you never made a RivaTuner profile for the game yourself, deleteProfiles\ocarina.exe.cfgin RivaTuner's folder once: earlier versions could leave a frozen copy of the global profile there. - The game hands the limit to RivaTuner only once RivaTuner is seen holding it, and takes it back while a movie plays, while the window is minimised or no picture goes out, and for the run when RivaTuner lets frames through faster than the limit or never shows its hold. A limit set in RivaTuner itself on top of the game's is named in the log and in the menu's frame rate row.
- A limit change reaches RivaTuner's profile once it stays, not at every switch of low latency;
displayon a 59.94 Hz monitor is rounded down for RivaTuner versions without fractional limits. - With DLSS Frame Generation, Reflex now holds the frame limit (
reflex_limit = 1), so the game waits once per frame instead of twice; the game's limiter takes over during movies and when Reflex does not hold it. - Fewer stutters: turning DLSS or FSR on or off, changing its mode, pausing and resizing no longer stop the graphics card until it is idle; fewer memory freezes right after an area load; pipelines are remembered as soon as they are first built; the shipped list of pipelines built before play grew to 10,230.
Fixed above 60 fps
- The ocarina's song recorder keeps its timing at any frame rate (
ocarina_timing, on by default). Above 60 fps its clock stood still: a song taught to the scarecrow lost its pauses and the recorder's time limits never came. Songs recorded above 60 fps in earlier versions keep their missing pauses. - The shake key swings at the game's own pace at any frame rate.
- Dynamic resolution (
dyn_res = 1) aims at the frame limit you set, not at 60 fps.
Settings
- In
nx2navi.ini, a comment after a value (pace_smooth = 0 ; off) and the wordsoff,falseandnonow mean what they say for on/off settings; before, such a line could turn the setting on. Paths and names keep their whole text. - New:
reflex_limit,ocarina_timing,audio_strict,hang_watch_s,mesh_fallback_hold, and switches that bring back the earlier behaviour of this release's larger changes should one cause trouble:upscaler_idle_wait,present_order,wb_lazy_unlocked. Seenx2navi.ini.example.