DLSS 5 Autopilot 2.1.1
Updating from 2.1.0 with update now downloads this build, checks its
SHA-256 against the SHA256SUMS.txt published with the release, and offers
a one-click restart. Your settings, library, hidden games and profiles carry
over.
Added
- run DLSS in a helper process (feeder route, 64-bit games). DLSS5-Feeder's
helper mode: the game hands its frames tohost64\dlss5-feed-host64.exe,
which runs DLSS outside the game's process. It is for games whose own
process cannot start DLSS - the diagnosis names those. Needs feeder 1.18.0
or newer; the work area is 100 % there. On the command line:
--feed-helper. - Built-in motion vectors on the feeder route. DLSS5-Feeder 2.0.0
estimates motion vectors insideDLSS5_Feed.fxitself. motion vectors
offers it as Built in: nothing else is downloaded and there is no second
technique to switch on. It needs feeder 2.0.0 or newer (an older build is
refused before anything is written), and it is not offered for Direct3D 10
and Vulkan games, where the feeder's author has not verified it.
LumeniteFX Kernel stays the default. - jlrouzies' fork in optiscaler build: wilsjo2's build with the
neural-rendering settings inOptiScaler.iniapplied while the game runs
(DLSS5-Feeder 2.0.0's menu drives it). Offered, never picked automatically. - Presets L and M in dlss preset. From feeder 2.0.0 auto means preset L
(it was K). - The game page says that Alt+C opens the feeder's own menu (feeder 2.0.0
and newer).
Changed
- An Unreal game whose NVIDIA plugins ship DLSS Frame Generation or Ray
Reconstruction is read as Direct3D 12 even when its executable imports
d3d11.dll, so the routes for Direct3D 12 games are the ones offered
(#1014). - Before an install, a game that asks Windows for a Direct3D 12 Agility SDK
it does not ship, on a Windows whose own Direct3D 12 is older, is named
with the routes it affects (#1035). - Reinstalling and update all keep a motion-vector provider switched in
the feeder's own Alt+C menu, and a recorded generic provider (0) stays 0. - The tool writes the motion-vector provider into the
[DLSS5_Feed.fx]
section ofReShadePreset.iniwhen that section exists. The effect's own
setting wins over the global one, so a provider written only globally was
ignored. - Arknights: Endfield counts as anti-cheat (ACE): the install warns
before it writes anything, and the game is left out of update all
(#991, #992). - The bug report carries an
anti-cheat:line when one sits beside the
game.
Fixed
- A game that never started and whose process Windows would not open
(protected), or that ships an anti-cheat, is answered with its anti-cheat
and uninstall, instead of "undo the dlss page" or "not run yet"
(#992, #1047). - A game read before it had loaded any graphics DLL (a large process with no
d3d9/d3d11/d3d12/opengl/vulkan in it) is answered "press did it work?
again while the game is drawing", instead of "try another proxy name"
(#975). D3D12_ERROR_INVALID_REDIST(0x887E0003) is named: no Direct3D 12 device
can be made inside the game, every route that needs one fails the same
way, and the feeder's helper mode is the way to try (#1035).- A bridge install on a game that drew with Direct3D 12 is answered with
what that means (on Direct3D 12 the DLSS 5 add-on takes the game's own DLSS
call, and native is the route for it), instead of "the bridge delivered no
frame" (#1014). - A fault in a Windows or Visual C++ runtime library (
ucrtbase.dll,
ntdll.dll,msvcp140.dll...) no longer blames "an overlay": the event
names who was called, not who called (#987, #289). - On the feeder route, a game that rebuilt its graphics device during
start-up, where ReShade then refused the add-on, is named, with what to try
(start in fullscreen or borderless at the display's resolution) (#987). - Feeder 2.0.0's add-on and its helper refuse each other when they come from
different releases. The install writes the helper first, so a helper still
held by the last session stops the install before the add-on is replaced,
and did it work? names a mismatch. - did it work? reads feeder 2.0.0's built-in motion-vector line; an
earlier session's LumeniteFX line used to answer for it. - Uninstalling a feeder install that used no downloaded provider (generic,
iMMERSE or built-in) took the person's own LumeniteFX techniques out of
the preset. - GITC-Uplift: did it work? reads
[Uplift] Enabledfrom
ReShade.ini, so a session with Enable ticked and no neural frame is not
answered "tick Enable". On Vulkan and OpenGL, a cached ReShade older than
6.8 (when reshade.me and GitHub cannot be reached) is refused before the
previous route is taken out. - The RTX 40 multi-frame generation extra asks for its v1.2.1 release by tag
on the project's new name (MFG-Unlocker). With GitHub's API rate limited,
the release list fell back to pages that no longer reached it. - DXVK and REFramework downloads use the cached release list and github.com's
pages when the API cannot be reached (#965). - A video download that saved its file is no longer reported as failed
because yt-dlp exited with an error afterwards (#1034). The video
processor's whole output is kept beside the result as a.log(#1042).