AMDNR-v0.3.4.2

GitHub release — AMDNR v0.3.4.2

Title: AMDNR v0.3.4.2 — hotfix: Assetto Corsa menu input, runtime chooser, the Ray Regeneration section, Wine / Proton text
Video: Hotfix video
Tag: Alpha0.3.4.2

---

Hotfix for 0.3.4.1. Neural Rendering is byte-identical to 0.3.4.1 except one accepted-layout row: the neural pass,
both runtimes and the pak are unchanged; only OptiScaler.dll and the documents change. That row is what makes this
build accept one more danielblnc runtime layout (see below).

Join the Discord (support, bug reports, test builds): https://discord.gg/AMDNR

The danielblnc runtime is Daniel Blanco's work. The AMD neural runtime in the *Runtime.zip files
(dlssnr_amd_pass1..3.dll) is DLSS-NR on AMD by Daniel Blanco (danielblnc) -
https://github.com/danielblnc/DLSS-NR-on-AMD. Copyright (c) 2026 Daniel Blanco, all rights reserved.
AMDNR ships it unmodified, with his permission; it is not AMDNR's work. Please support his project.
The lmxxf runtime is built on lmxxf by Kien (MIT).

What's fixed

• Assetto Corsa: the menu (the second report, at a low frame rate: the menu opened by itself on the runtime
  chooser, clicks did nothing, and Insert hid the menu only while it was held).
  • The menu key toggles once per physical press: it acts on the press, never on the release, and a key message that
    arrives late is ignored.
  • Clicks and menu keys shorter than a frame are no longer lost: both edges are kept and replayed inside one frame.
  • The runtime chooser answers to 1 / 2 (pick), Enter (use) and Esc (Decide later), has a title-bar X, and
    closing the menu with Insert counts as Decide later, so it can never trap you.
  • Found in review and fixed before release: with [Menu] OverlayMenu=false and frame generation on, a replayed
    click could arrive twice; a menu key press sent from another thread could look dead; the held menu key is now
    hidden from the game's DirectInput reads and keyboard hooks too, until it is released.
  • To get 0.3.4.1's menu key and clicks back (no new key): add DiagInputHooksSkip=presslatch,clickreplay under
    [Hotfix] in your OptiScaler.ini yourself; the shipped ini only describes the line in a comment.
  • The fix is built and covered by this release's tests; confirmation from the Assetto Corsa player who
    reported it is still pending.
• The runtime chooser no longer opens the menu by itself. With files for both runtimes present and no choice
  made, the menu stays closed and one notice says which runtime runs until you pick one and which key opens the menu; the chooser
  appears the first time you open the menu, and Decide later still works. What runs while unchosen is 0.3.4.1's rule.
• The Ray Regeneration settings could not be found. An RX 7000 player went looking for the Ray Regeneration
  section in the Neural tab and found nothing: it was drawn only while FSR Ray Regeneration had denoised a frame in
  the last 3 seconds, or when Ray Regeneration had given the game up. The section is now always there, with one dim
  line for each of four cases: it is denoising now, the game has not turned Ray Reconstruction on (that line gives
  the three steps in the game), Ray Regeneration gave this title up and why, or it last ran N seconds ago. A card
  that needs a word about itself gets a third dim line: on RX 7000 it is offered by default but the driver may refuse
  it; on RX 6000 and older it is not offered, because AMD ships it for RDNA 4, and
  [FSR-RR] FfxDenoiserAllowPreRdna4=true offers it anyway. The controls stay on screen, greyed out, while it is not
  denoising. No setting and no default changed with this.
• Wine / Proton text: the Neural tab and the [RR_DIAG] log line no longer claim that FSR Ray
  Regeneration works there. They now say what the 0.3.4.1 notes say: FSR upscaling works under Wine / Proton; FSR Ray
  Regeneration is a known issue there (pink / magenta patches in some games). Nothing else changed on Proton.
• Diagnostics for support: the log names the overlay DLL a game loads and the caller of D3D11CreateDevice (INFO,
  once per distinct name; an overlay AMDNR blocks keeps its line at DEBUG); menu input lines at INFO.

Ray Regeneration

• The sharpening AMDNR adds after Ray Regeneration is an ini key now: [Sharpness] RrDefaultSharpness, default
  unchanged at 0.25 - the number used when the game sends no sharpness of its own, applied by FSR to the denoised
  picture. Sharpening after a denoiser lifts the grain it left, so grain on faces or textures can now be answered with
  0.15, 0.10 or 0 in the ini instead of a new build. A game's own value keeps priority, an explicit Override wins over
  both, Wine / Proton still adds nothing, and Save Settings keeps your number.
• A [Sharpness] Sharpness left in the ini while Override is off is named as waiting - it does nothing until you
  tick Override and then applies at once (one report carried 1.00, four times the RR default). Image > Sharpness shows
  a dim tag on the Ray Regeneration path, and the [RR_POST] log line says the same.
• No denoiser tuning changed and no default moved: every Ray Regeneration number is 0.3.4.1's.

Neural Rendering

• This build accepts one more danielblnc runtime layout. AMDNR recognises a danielblnc pass DLL by the file
  itself (size and SHA256) and reads its knobs from a table of accepted layouts; a build that is not in that table is
  not driven. This release adds one more row to that table, so a newer danielblnc build can be driven here without
  waiting for an AMDNR update: the Neural tab names the file and its version as usual, and Network style, Tone curve,
  Black lift, Game exposure and Fast mode work with it. The layouts accepted before are untouched
  and no default moved - that row is the only change on the neural side of OptiScaler.dll, and the runtimes and the
  pak in this zip are the 0.3.4.1 files, byte for byte.

Docs

• README: the rename list now names every proxy name the mod accepts (dxgi.dll, d3d12.dll, winmm.dll,
  version.dll, dbghelp.dll, winhttp.dll, wininet.dll); d3d11.dll is not one of them.
• README FAQ: Uncharted: Legacy of Thieves Collection crashes at the first NR frame on RX 9000 (a stack overflow
  at HIP's first initialisation on the game's 192 KiB job fiber): keep [DlssNr] Enabled=false in that game until
  0.3.5; switching runtime does not help.
• README FAQ: hybrid PCs (integrated graphics plus a Radeon): if the game renders on the integrated GPU, NR never
  runs; set the game to the discrete GPU in Windows Settings > System > Display > Graphics.
• README: [DlssNr] AmdLmxxfTierSnap=true on RX 9000 (opt-in, off by default; some sizes drop a tier, 1440p Quality
  1707x960 -> the 900 size: network 14.08 -> 9.96 ms measured outside a game, slightly softer); the lmxxf cost hover
  says the same on RX 9000.
• README FAQ: "You cannot find the Ray Regeneration settings?" - where the section is, what its dim line means
  while Ray Regeneration is not running, and the three switches a game needs (DLSS as the upscaler, ray tracing or
  path tracing, Ray Reconstruction). All nine languages.
• README FAQ: "Ray Regeneration looks grainy or noisy?" - judge it with Neural Rendering switched off, because
  the neural pass runs after Ray Regeneration on its output; the setting to try per kind of grain is in
  the Ray Regeneration settings guide RR-BEST-SETTINGS.md
  (not in the zip); and a value under [Sharpness] Sharpness does nothing until
  you tick Override, then it applies at once. No denoiser default and no sharpening default changed in 0.3.4.2.
  Said plainly there as well: a part of the grain is the game's own ray sampling, which AMD's denoiser is not built to
  repair, and the grain that crawls in a still scene has a structural cause on our side that no slider removes
  completely - a known issue, worked on for 0.3.5.
• Known issue, unchanged: Ray Regeneration on Proton can show magenta patches in some games; use plain FSR (FSR 4 on
  RX 9000) and send a Save report.

Update

• AMDNR Launcher users: it updates for you (press REPAIR / UPDATE on a game that shows "Update available").
• No launcher yet? AMDNR-Launcher.exe 0.3.5 ships on this release (signed, 71,952,104 bytes, SHA256
  fdf26eff70c9ee0b9732ccbcc79570ca13a39ddfc8f228de8248f90adc4d9bfe):
  https://github.com/3zwr1/AMD-NR---OptiScaler/releases/download/Alpha0.3.4.2/AMDNR-Launcher.exe. It updates
  itself, speaks nine languages, and can run under Wine / Proton on Linux (experimental, not tested by us yet).
• By hand, coming from 0.3.4.1 or 0.3.4: replace OptiScaler.dll only (the file you renamed, e.g. dxgi.dll, with
  the new one renamed the same way). LmxxfNrRuntime.dll, LmxxfNrRuntime.pak and your danielblnc runtime files are
  unchanged: keep them. Your OptiScaler.ini works as it is.
• On this release: AMDNR-v0.3.4.2.zip, SHA256SUMS.txt and AMDNR-Launcher.exe.
• danielblnc 0.4.3 (published 2026-09-28) is supported day one - its layout was already accepted in the host, so
  AMDNR 0.3.4, 0.3.4.1 and 0.3.4.2 all drive it with no AMDNR update. It is the recommended danielblnc runtime on
  RX 9000 and RX 7000: v0.4.3-Runtime.zip on the
  Alpha0.3.4.1 release, which the AMDNR
  Launcher offers as the recommended runtime; v0.4.1-Runtime.zip and v0.4.0-Runtime.zip on the same release are
  still accepted, and 0.3.3 is retired (no longer recommended for RX 7000). The lmxxf runtime needs no runtime zip.
  His own numbers for 0.4.3: +20% Reference / +18% Fast vs 0.4.2, his measurement.
• **New install, or older th...

AMDNR-v0.3.4.1

GitHub release — AMDNR v0.3.4.1

Title: AMDNR v0.3.4.1 — hotfix: Ray Regeneration sharpening, Control Resonant popup, Linux / Proton, AMDNR Launcher 0.3.4.1

Tag: Alpha0.3.4.1
Video: HOTFIX Video

Hotfix for 0.3.4, with AMDNR Launcher 0.3.4.1. Neural Rendering is unchanged from 0.3.4 (same runtime and
pak).

Join the Discord (support, bug reports, test builds): https://discord.gg/AMDNR

The danielblnc runtime is Daniel Blanco's work. The AMD neural runtime in the *Runtime.zip files
(dlssnr_amd_pass1..3.dll) is DLSS-NR on AMD by Daniel Blanco (danielblnc) -
https://github.com/danielblnc/DLSS-NR-on-AMD. Copyright (c) 2026 Daniel Blanco, all rights reserved.
AMDNR ships it unmodified, with his permission; it is not AMDNR's work. Please support his project.
The lmxxf runtime is built on lmxxf by Kien (MIT).

What's fixed

• Ray Regeneration is less soft (Windows). With FSR Ray Regeneration, when the game sends no sharpness (Resident
  Evil Requiem sends 0), AMDNR now sharpens by 0.25 after RR. A game that sends its own value keeps it; the
  path-traced profile keeps 0. To turn it off: Image > Sharpness, tick Override, slider 0
  ([Sharpness] OverrideSharpness=true, Sharpness=0).
• Control Resonant: no more false "Upscaler failed to run!" popup (FSR Ray Regeneration skipped one frame to
  re-create itself at the game's render size; that is now a log warning).
• Smaller: the skin classifier's "Guide only" mode carries a "can blur non-skin" warning; the log names the FSR
  upscaler in use and its version (also the one after Ray Regeneration); an FSR FG log line no longer says "XeFG";
  Streamline's [RR_TAG_DIAG] lines are no longer logged every frame.

Not in this hotfix: the Assetto Corsa report (the mouse in the menu) is still being looked into.

Linux / Proton

• The menu works. It now attaches to the game window. Confirmed by a player on Steam Proton (RX 9070 XT,
  vkd3d-proton, Resident Evil Requiem): the game boots, the menu opens and takes the mouse, and Save report works,
  also with frame generation switched on. If the window is still lost, one "Menu window lost" warning with the fix
  replaces a log line every frame.
• Known issue: Ray Regeneration can show pink / magenta patches on Proton; the default sharpening is off there now, but if you still see them use plain FSR (FSR 4 on RX 9000) and send a Save report.
  On Proton, AMDNR adds no sharpening after Ray Regeneration when the game sends none (Windows: 0.25); Image >
  Sharpness > Override still sets your own value.
• Frame generation now turns on without crashing, but fps counters count the generated frames too: under a
  60 fps cap or 60 Hz V-Sync that is 30 real frames, which looks like 30. Leave it off on Proton for now
  ([FrameGen] FGOutput=nofg), or use it only when the game reaches about 60 fps without it, on a screen faster
  than 60 Hz.
• Save report now shows whether the game ran on vkd3d-proton or DXVK.
• Neural Rendering does not run on Linux (Windows only). The AMDNR Launcher is a Windows program that can run
  under Proton (experimental, not tested by us yet, see below). The README has a new Linux / Proton section (the
  launcher under Proton, install by hand, HDR, reporting).
  Linux (experimental, untested by us): run the same AMDNR-Launcher.exe through Proton.
  Steam > Add a Non-Steam Game > pick the exe > Properties > Compatibility > force Proton Experimental.
  Start it, pick your game, INSTALL, close it. Then on the game: Properties > Launch Options:
  WINEDLLOVERRIDES="dxgi=n,b" %command% (use the DLL name the launcher shows if it is not dxgi)
  Start the game from Steam. FSR works; Ray Regeneration can show pink patches (use plain FSR); keep frame generation off; neural rendering needs Windows. If the launcher does not open, install the zip by hand and send a report.

You asked, we built it: AMDNR Launcher 0.3.4.1

Built from the first Discord feedback. AMDNR-Launcher.exe on this release (0.3.4.1, signed):
https://github.com/3zwr1/AMD-NR---OptiScaler/releases/download/Alpha0.3.4.1/AMDNR-Launcher.exe

• Nine languages: English, Arabic, Chinese (Simplified), French, Spanish, Portuguese, Italian, Russian and Polish.
  It follows your Windows language (else English); pick another under LANGUAGE or in SETTINGS, also offered on
  first start. Doctor findings, install messages and the COLLECT LOGS report stay in English so support can read
  them.
• Search the library.
• Favourites: starred games come first.
• Hide games (HIDDEN shows them again).
• Rename a game.
• CHOOSE GAME .EXE: pick the game's exe yourself when the launcher picked the wrong one or none.
• Cyberpunk 2077 and The Witcher 3 (REDengine): found in the right folder now.
• PLAY and OPEN FOLDER on every game's page.
• STORES: switch whole stores on and off; folders you add by hand stay listed until you remove them.
• UNINSTALL asks first and removes everything the mod placed (logs, caches, crash dumps, unfinished reports); it
  keeps finished Save report zips and a DLL that is not an OptiScaler (the game's own).
• COLLECT LOGS on any game, installed or not, with a scan report.
• Handheld / APU recognition (ROG Ally Z1 Extreme and other Ryzen APUs), with a note that AMDNR on them is still in
  testing.
• Linux (experimental, not tested by us yet): the same Windows exe can run under Proton, shows a notice with what
  to do there and also looks for games in your Linux Steam library (start the game from Steam: PLAY is off under
  Wine / Proton). Steps in the README ("Linux / Proton"); tell us on
  Discord if it works.

Update

• AMDNR Launcher users: it updates for you. The launcher offers RESTART TO UPDATE for itself; then press
  REPAIR / UPDATE on a game that shows "Update available".
• By hand, coming from 0.3.4: replace OptiScaler.dll only (the file you renamed, e.g. dxgi.dll, with the
  new one renamed the same way). LmxxfNrRuntime.dll, LmxxfNrRuntime.pak and your danielblnc runtime files are
  unchanged: keep them. Your OptiScaler.ini works as it is.
• On this release: AMDNR-v0.3.4.1.zip, SHA256SUMS.txt, AMDNR-Launcher.exe (0.3.4.1, signed), and the
  danielblnc runtime zips v0.4.1-Runtime.zip and v0.4.0-Runtime.zip (unchanged: the same files as on
  Alpha0.3.4). v0.3.3-Runtime.zip and Runtime.zip stay on the
  Alpha0.3.4 release.
• Which danielblnc runtime zip: on RX 9000, v0.4.1-Runtime.zip (or v0.4.0-Runtime.zip) from this release;
  on RX 7000, v0.3.3-Runtime.zip from the
  Alpha0.3.4 release. The lmxxf runtime
  needs no runtime zip on RX 7000 and RX 9000; handheld APUs use lmxxf only. The AMDNR Launcher picks the zip for
  your GPU.
• New install, or older than 0.3.4: the AMDNR Launcher, or by hand AMDNR-v0.3.4.1.zip (plus the runtime zip
  for your GPU for the danielblnc runtime, see above) and the README's installation guide.

Full notes: CHANGELOG.md.

Credits

Daniel Blanco (danielblnc) (the danielblnc runtime, used with his permission;
https://github.com/danielblnc/DLSS-NR-on-AMD), lmxxf (Kien, MIT;
https://github.com/lmxxf/dlss5-on-amd-9070xt-porting), AMDNR's RDNA 3 backend by 3zwr1, TheAutomatic,
Matheus, Dagherbou, wilsjo2, burak113 (the FSR Ray Regeneration preprocessor), RenoDX
(clshortfuse), Coldwood1026, Nukem9 and OptiScaler (Overclockers and contributors). The full list with
links is in the README and in the 0.3.4 release notes. AMDNR is Copyright (c) 2026 3zwr1 (AMDNR), GPL-3.0 with the
additional terms in Licenses/AMDNR_NOTICE.txt.

Discord: https://discord.gg/AMDNR — bug reports with the Save report zip in #bug-report.

AMDNR-v0.3.4

GitHub release — AMDNR v0.3.4

Title: AMDNR v0.3.4 — the new menu, faster lmxxf on RX 7000, handheld APUs (experimental), one set of controls for both runtimes, AMDNR Screen GI (preview), Save report

Tag: v0.3.4

---

DLSS 5 Neural Rendering on AMD, built into OptiScaler.

Join the Discord (support, bug reports, test builds): https://discord.gg/AMDNR
Ko-fi: https://ko-fi.com/3zinr

The danielblnc runtime is Daniel Blanco's work. The AMD neural runtime in the *Runtime.zip files
(dlssnr_amd_pass1..3.dll) is DLSS-NR on AMD by Daniel Blanco (danielblnc) -
https://github.com/danielblnc/DLSS-NR-on-AMD. Copyright (c) 2026 Daniel Blanco, all rights reserved.
AMDNR ships it unmodified, with his permission; it is not AMDNR's work. Please support his project.
The lmxxf runtime is built on lmxxf by Kien (MIT). Full credits for everyone else are at the end of this page.

Highlights

• The new menu. One layout for both runtimes, danielblnc and lmxxf, and every tab redone in the same look.
  The Neural tab starts with Enable, the runtime and one status line, then Performance, Quality, Image look and,
  while a game runs it, Ray Regeneration. Rarely used tools sit in a closed row (Diagnostics / Runtime options /
  Experimental). Help opens when you hover a label. Same ini keys: your OptiScaler.ini works.
• lmxxf is faster on RX 7000 (RDNA 3): the NR size now snaps to lmxxf's cheaper network size tier by default on
  RX 7000, Radeon 8060S / 8050S and the handheld APUs. On an RX 7800 XT, 1440p with FSR Quality (a 1706x960
  render) now runs the 900 tier instead of the 1080 tier: 73.3 -> 52.2 ms per network run. A 1080p render at NR
  resolution 70% now runs the 720 tier instead of the 900 tier: 52.2 -> 34.4 ms (network time in lmxxf's probe on a tester's card, outside a game). RX 9000 is unchanged.
• Handheld APUs (experimental): Z1 Extreme / Z2 / Radeon 780M and Z2 Extreme / Radeon 890M / 880M run lmxxf
  through AMDNR's RDNA 3 backend, with two new, smaller network sizes (360p and 576p) and the model every 4th
  frame by default. Slow: first results on a ROG Ally (Z1 Extreme): 54.7 ms per network run at 360p in lmxxf's probe outside a game;
  in a game, one tester's run of Shadow of the Tomb Raider (1280x720, XeSS, Handheld preset): 62 ms on average at 360p with the model every 4th
  frame, about 29 fps with NR on. Reports wanted.
• lmxxf is faster on RX 9070 / RX 9070 XT: network time 15.35 -> 14.08 ms at 1080p (10.79 -> 9.96 ms at 900p, 6.97 -> 6.63 ms at 720p) on an RX 9070 XT,
  measured outside a game, the same network output bit for bit,
  with lmxxf's new kernels from lmxxf 0.31 (Kien, MIT) on top of AMDNR's c32w.
• Both runtimes, one set of controls: lmxxf gains Network output, Encoding, Residual edge fade and Game exposure; danielblnc gains its runtime settings (Network style, Tone curve, Black lift, Game exposure) and
  Network history with Model interleave off.
  • lmxxf now also follows Native character mask, Structure intensity and Character structure.
• Fast mode (opt-in): lmxxf runs its network one size tier lower: about 29% less network time at 1080p (RX 9070
  XT, measured outside a game), fine detail a little softer. danielblnc builds that have their own Fast mode get the
  same row.
• AMDNR Screen GI (preview, off by default): AMDNR's own screen-space bounce light and ambient occlusion, before
  NR and the upscaler, with NR on or off; about 1 ms at High for a 1080p render on an RX 9070 XT (GPU lab, not a
  game). Neural > Experimental. The inherited Screen-space GI is retired from the menu.
• Save report: one button (Neural > Diagnostics, or Advanced > Logging) zips your logs, your ini and a report
  for a bug report, with your Windows user name and PC name masked.
• Menu input fixes: with the menu open, a game that hooks the keyboard at a low level (Assetto Corsa) no longer
  blocks Alt+Tab and the Windows key for the whole desktop, and one key press toggles the menu or NR once.
• Game fixes: Shadow of the Tomb Raider no longer crashes when the upscaler starts (seen on a ROG Ally); Marvel's
  Midnight Suns no longer crashes on danielblnc; lmxxf keeps running in games that submit their frames late (The Last
  of Us Part II); no false "No clean exit" after quitting Forza Horizon 6.
• lmxxf no longer switches its colours by itself in long sessions (its stats counters were not cleared, which
  could trigger a false self-heal).

Which files to replace

All three changed. Coming from 0.3.3.x, replace:

1. OptiScaler.dll (rename it as before, e.g. dxgi.dll or winmm.dll)
2. LmxxfNrRuntime.dll (0.3.4, Properties > Details)
3. LmxxfNrRuntime.pak (the 0.3.3.2 pak plus new entries: the handheld builds for gfx1103 and gfx1150, and lmxxf 0.31's kernels (Kien, MIT);
   the c32w kernels are unchanged)

Keep the three together. Mixed with older files:

• new OptiScaler.dll + 0.3.3.2 LmxxfNrRuntime.dll: NR runs, but lmxxf ignores the character mask and structure
  controls (one log line), has no small network sizes, and a handheld gets only the slow 720 tier;
• new LmxxfNrRuntime.dll + 0.3.3.2 pak: desktop GPUs run as before; a handheld gets "no lmxxf modules for this
  GPU";
  • RX 9070 / 9070 XT keep 0.3.3.2's speed without the new pak;
• a handheld with an older OptiScaler.dll: the new runtime refuses it ("this handheld needs OptiScaler.dll 0.3.4 or
  newer").

Your danielblnc files stay. Keep a backup of the old three. Your own OptiScaler.ini works; new keys use their
defaults, but see Settings you saved before below. If the status line says c32w=off:nofile with the new pak,
an old DLSS5-AMD\native-game-tiled-assets folder next to the game's exe is used instead of the pak: remove or
rename that folder.

New install: extract AMDNR-v0.3.4.zip into the game folder, rename OptiScaler.dll to dxgi.dll (or
winmm.dll / version.dll / OptiScaler.asi with an ASI loader), then INSERT for the menu. For danielblnc,
also extract one runtime zip from the table below into the game folder.

Downloads

File	What	Who
AMDNR-v0.3.4.zip	AMDNR: OptiScaler with AMD Neural Rendering and the complete lmxxf runtime (LmxxfNrRuntime.dll + LmxxfNrRuntime.pak)	everyone
v0.4.0-Runtime.zip	danielblnc's runtime 0.4.0 (by Daniel Blanco, unmodified, with his permission) + weights. Newest public build; the new danielblnc runtime settings work with it	danielblnc users who want the newest runtime
v0.3.3-Runtime.zip	danielblnc's runtime 0.3.3 (by Daniel Blanco, unmodified, with his permission) + weights; the new danielblnc runtime settings work with it	danielblnc users who prefer 0.3.3
Runtime.zip	danielblnc's runtime 0.3.1 (by Daniel Blanco, unmodified, with his permission) + weights. The new runtime settings are greyed with it	danielblnc users, tested

Use all three pass DLLs from one zip.

Settings you saved before

Some values you saved yourself act differently in 0.3.4, and a few defaults changed:

• Preset (Quality / Balanced / Performance) now sets only NR resolution (100 / 85 / 70%) and turns Dynamic NR off;
  it no longer changes Temporal stability, Sharpening, Detail or Colour strength, so your NR style stays.
• Settings saved while on danielblnc now also act on lmxxf: AmdEncoding (sRGB / Gamma 2.2), AmdResidualFade,
  AmdNetworkOutput=true and AmdRtgiEnabled=true (Screen-space GI, which costs GPU time and VRAM).
• [DlssNr] AutoMask=false now turns danielblnc's native character mask off (0.3.3.2 kept it on).
  • On lmxxf, AutoMask=false, a Structure intensity other than 1, your own Character structure and the NR
    styles that set structure (Cinematic, Crisp, Natural, Vivid) now change the picture too, on every GPU; each change
    rebuilds lmxxf's network (about a one-second hitch).
• AmdInterleaveModelHistory=true now also keeps danielblnc's network history with Model interleave off.
• A saved AmdInterleavePacing between 0 and 1 now also paces lmxxf with Model interleave on, which lowers the frame
  rate; the default (-1, automatic) stays off on lmxxf.
• lmxxf's "Classic carry" interleave preset is now saved as AmdInterleavePreset=1 (0.3.3.2 wrote 6); an old 6 still
  reads as Classic carry on lmxxf.
• Reset appearance filter resets only the appearance filter ([AmdLook]), no longer NR resolution, passes or
  structure.
• On RX 7000, Radeon 8060S / 8050S and the handhelds the tier snap is on unless your ini says
  [DlssNr] AmdLmxxfTierSnap=false.

What changed

The menu

• One look on every tab, drawn from the approved mock: new theme, text tabs with a red underline, a one-line
  footer (Menu Scale, Save Settings, Close), help on hovering a label (no more (?) markers), Menu Scale scales
  everything. The window title reads AMDNR v0.3.4 - <exe> - <game>; the startup splash says "AMDNR - Insert for
  menu".
• Header: AMDNR (hover for the version and build), Discord, GitHub (it opens AMDNR's README, which replaces the
  old Guide button) and the copyright, then the credits row ("Neural runtimes: DLSS-NR on AMD by Daniel Blanco
  (danielblnc) - lmxxf by Kien (MIT)"; click Daniel's name for his GitHub, hover for the full credits).
• Neural tab, top: Enable with its key, the Neural runtime combo with one state word (running / restart the game to
  switch / not installed / not for this GPU), the running runtime's credit as a link, and one status line (state,
  size, NR and model rate, and the NR cost in ms). At most one orange line below it, with a fix button
  where there is one (Retry lmxxf, Switch to danielblnc after an lmxxf error, Open Upscaling for a D3D11 / Vulkan game
  on the wrong upscaler). Preset is three buttons; NR style re...

AMDNR-v0.3.3.2

GitHub release — AMDNR v0.3.3.2

Title: AMDNR v0.3.3.2 — faster lmxxf on RX 9070 (XT), lmxxf on iGPU / two-GPU PCs, Edit accumulation interleave, RE Requiem skin classifier

Tag: v0.3.3.2

---

DLSS 5 Neural Rendering on AMD, built into OptiScaler.

Join the Discord (support, bug reports, test builds): https://discord.gg/AMDNR
Ko-fi: https://ko-fi.com/3zinr

The danielblnc runtime is Daniel Blanco's work. The AMD neural runtime in the *Runtime.zip files
(dlssnr_amd_pass1..3.dll) is DLSS-NR on AMD by Daniel Blanco (danielblnc) -
https://github.com/danielblnc/DLSS-NR-on-AMD. Copyright (c) 2026 Daniel Blanco, all rights reserved.
AMDNR ships it unmodified, with his permission; it is not AMDNR's work. Please support his project.
Full credits for everyone else are at the end of this page.

Highlights

• lmxxf is faster on RX 9070 / RX 9070 XT (RDNA 4): network time 17.8 -> 15.3 ms at 1080p on an RX 9070 XT
  (720p: 8.0 -> 6.9 ms), the same image bit for bit. New "c32w" one-wave kernels for the network's C32
  layers, AMDNR's own work, ship in LmxxfNrRuntime.pak.
• lmxxf on PCs where the game's GPU is not HIP device 0: a Ryzen desktop with its integrated graphics
  on, a laptop with an AMD APU and a Radeon, or two AMD GPUs. It used to fail on its first frame
  (hipErrorInvalidHandle (400)) and stay off for the whole session; it now selects the right GPU on every call
  (not yet tested on a PC with two AMD GPUs).
• lmxxf no longer keeps about 100 MB of VRAM (and as much RAM) on every NR resolution, DLSS mode or
  resolution change. A smaller rest of about 10-25 MB per change remains (not yet tested in a game).
• Model interleave: new default preset "Edit accumulation", against the one-frame ghost and the 30 Hz
  grain, at no extra cost over Guided fill v2 (not yet tested in a game).
• danielblnc: no more style jump at Residual strength 0.99, or when NR resolution leaves 100% (not yet
  tested in a game).
• Resident Evil Requiem, FSR Ray Regeneration: a new skin classifier that leaves streets, walls, coats and
  umbrellas alone; RR bias mask strength 0 by default; a warning when REFramework is missing.
• Ray Reconstruction is offered on RX 9000 (RDNA 4) only by default, so RX 7000 keeps the game's own
  denoiser (not yet tested in a game on RX 7000).

Which files to replace

All three changed. Coming from 0.3.3.x, replace:

1. OptiScaler.dll (rename it as before, e.g. dxgi.dll or winmm.dll)
2. LmxxfNrRuntime.dll
3. LmxxfNrRuntime.pak (the new pak carries the c32w kernels)

Your danielblnc files stay. Keep a backup of the old three. Your own OptiScaler.ini works; the new keys use
their defaults. LmxxfNrRuntime.dll 0.3.3.2 also works with the old pak, just without the speed-up
(c32w=off:nofile).

New install: extract AMDNR-v0.3.3.2.zip into the game folder, rename OptiScaler.dll to dxgi.dll (or
winmm.dll / version.dll / OptiScaler.asi with an ASI loader), then INSERT for the menu. For danielblnc,
also extract one runtime zip from the table below into the game folder.

Downloads

File	What	Who
AMDNR-v0.3.3.2.zip	AMDNR: OptiScaler with AMD Neural Rendering and the complete lmxxf runtime (LmxxfNrRuntime.dll + LmxxfNrRuntime.pak)	everyone
v0.4.0-Runtime.zip	danielblnc's runtime 0.4.0 (by Daniel Blanco, unmodified, with his permission) + weights. Newest public build	danielblnc users who want the newest runtime
v0.3.3-Runtime.zip	danielblnc's runtime 0.3.3 (by Daniel Blanco, unmodified, with his permission) + weights. Not yet tested in a game	danielblnc users who prefer 0.3.3
Runtime.zip	danielblnc's runtime 0.3.1 (by Daniel Blanco, unmodified, with his permission) + weights	danielblnc users, tested

Use all three pass DLLs from one zip.

What changed

lmxxf

• c32w kernels (RX 9070 / RX 9070 XT): 17.8 -> 15.3 ms network time at 1080p, 8.0 -> 6.9 ms at 720p, on an
  RX 9070 XT, the same image bit for bit. LmxxfNrRuntime.dll checks the kernels' SHA-256 before it loads them
  and runs the old kernels if they do not match. lmxxf's status line and lmxxf_backend.log show c32w=on when
  they run (c32w=off:<reason> otherwise). RX 9060 XT keeps the old kernels for now: the environment variable
  AMDNR_C32W=1 turns them on there (untested); AMDNR_C32W=0 turns them off everywhere. RX 7000 is not
  affected.
• PCs with more than one AMD GPU: LmxxfNrRuntime.dll now selects the network's GPU on every call and gives
  the thread its old GPU back afterwards (reported on an RX 7900 XTX). On our single-GPU PC the output is
  identical to 0.3.3.1's, also when frames are sent from a second thread. Not yet tested on a PC with two AMD
  GPUs.
• VRAM leak: an AMD HIP driver never gives back a D3D12 buffer that was imported and mapped, so lmxxf now
  makes these buffers once per network size and reuses them (lmxxf_backend.log: hip buffers ... imports N, reuses M). Leak probe, 40 changes: about +0.4 GB instead of +1.8 GB; not yet tested in a game.
• lmxxf no longer stays off for a whole session when the game never submits the list of the first neural frame
  (it re-arms after 8 frames and matches by COM identity after 16). Found by reading the code; not seen in a
  game.
• lmxxf stops at once when the GPU is removed (driver reset or crash), as danielblnc already did.
• The Neural tab says "lmxxf stopped after an error: ..."; amd_bridge.log lists every HIP device and marks the
  game's GPU; lmxxf's status line names the HIP device it uses.
• LmxxfNrRuntime.dll is 0.3.3.2 r2 (Properties > Details); the runtime combo reads "lmxxf (0.3.3.2)".

danielblnc

• Residual strength and the style jump: at 100% NR resolution, 0.99 now gives 99% of 1.00 and the NR styles
  look the way their strength says (no hidden Residual limit over the whole look). Below and above 100% (also
  Dynamic NR steps and the Balanced / Performance presets), Residual strength and limit act only on the model's
  own edit, before the look, temporal stability and sharpening, so changing NR resolution no longer clips the
  look. Default at 100% is unchanged, bit for bit. [DlssNr] AmdEditShaper=false switches this off away from
  100%. Not yet tested in a game.
• Less VRAM for new NR sizes: away from 100% NR resolution, working sizes above about 1 MP are rounded to
  64-pixel steps, so dynamic resolution, DLSS modes and Dynamic NR reuse a few sizes instead of keeping
  75-350 MB of VRAM per pass for each new one ([DlssNr] AmdNrSizeStep, 0 = exact sizes as before). 100% is
  unchanged.
• Clearer message when your danielblnc files are not a build this AMDNR drives: the Neural tab and
  amd_presr.log name the file and its version, say whether it is newer, older, a re-published copy or missing,
  and what to do.
• Warning when danielblnc's standalone DLSS-NR on AMD is installed next to AMDNR (his version.dll,
  winhttp.dll or dxgi.dll, or dlssnr_on_amd_setup.exe). AMDNR already runs his runtime; remove the
  standalone copy.
• Cost hints in the menu: danielblnc on RX 7000, and danielblnc with an NR input of 1440p or more.

Both runtimes

• Model interleave: "Edit accumulation" is the new default ([DlssNr] AmdInterleavePreset=10; lmxxf runs it
  as 11). No picture is carried over any more: every frame is this frame's raw picture plus the model's carried
  correction, checked raw against raw on every frame. This removes the held picture's one-frame ghost and its
  30 Hz grain and costs no more than Guided fill v2. On danielblnc the model frames show the fitted correction
  too; for the model's exact picture pick "Guided fill v2" (Neural > Performance > Interleave preset, or
  AmdInterleavePreset=6); on lmxxf "Classic carry" is the old fill. Also fixed for every preset: the
  "Model-frame ghost" meter and the Self-tuning totals now show values, a refused model call is no longer
  counted as a model frame, and a wrong barrier state in interleave's motion accumulation is fixed. Not yet
  tested in a game.
• Dynamic NR resolution no longer bounces between two steps: it steps down after 1.5 s over the target, back
  up only after 5 s of headroom, and waits a minute after a step up that could not hold. Its frame timer is
  precise now, so at high targets you may see a higher NR resolution and a lower frame rate than before. Not yet
  tested in a game.
• Still-surface steadiness (Neural > Quality, [DlssNr] AmdStabilityStaticRelax): steadies shadows and flat
  areas that pulse while nothing moves. Off by default; try 0.5 if shadows flicker on still walls.

Resident Evil Requiem / FSR Ray Regeneration

• Robust skin classifier (new default): the old SSS guide marked 25-39% of a night street as skin (walls,
  coats, umbrellas). The Robust classifier also checks how much the game's SSS pass moved the pixel, whether the
  surface around it agrees, and whether the albedo is skin-toned: 0.5-2.3% of the frame in the same scenes, with
  faces and hands still covered. Neural > Quality > Ray Regeneration > Skin classifier; "Guide only (0.3.3)" is
  the old mask.
• Skin smoothing is on by default in Resident Evil Requiem (strength 1.0, radius 16, guide thresholds
  0.0265 / 0.0414) against face flicker. A value you set yourself wins. Other games keep it off.
• RR bias mask strength 0 by default there (a tester saw less blotching and grain at 0). A value you set in
  [FSR-RR] FfxDenoiserBiasMaskStrength wins.
• Texture route for the path-traced profile ([FSR-RR] FfxDenoiserProfileTextureRoute, default 1.0): with
  the profile on, textured and very dark surfaces go back to the normal routing, against the etched look the
  profile gave textures in 0.3.3. The profil...

AMDNR-v0.3.3.1

AMDNR v0.3.3.1 — hotfix: danielblnc 0.4.0 + 0.3.3 support, RE Requiem Ray Regeneration fix

The danielblnc runtime is Daniel Blanco's work. The AMD neural runtime in the *Runtime.zip files
(dlssnr_amd_pass1..3.dll) is DLSS-NR on AMD by Daniel Blanco (danielblnc) -
https://github.com/danielblnc/DLSS-NR-on-AMD. Copyright (c) 2026 Daniel Blanco, all rights reserved.
AMDNR ships it unmodified, with his permission; it is not AMDNR's work. Please support his project.

What's new

• danielblnc 0.3.3 is supported (new, not yet tested in a game). danielblnc 0.3.1 and 0.3.2 keep working.
• danielblnc 0.4.0 is supported (Daniel's notes: 42% faster than 0.3.3; the new fast kernels run on RX 9000
  only). Get it in v0.4.0-Runtime.zip. 0.4.1 is supported too, ahead of its release. Not yet tested in a game.
• Resident Evil Requiem / PRAGMATA: FSR Ray Regeneration's path-traced profile is opt-in again. In 0.3.3
  it turned on by itself there and corrupted the picture. Tick it in Neural > Quality if you want to try it.
• Only OptiScaler.dll (and the ini text) changed. LmxxfNrRuntime.dll and LmxxfNrRuntime.pak are the same as in 0.3.3.
• Everything else in 0.3.3 is unchanged: see the 0.3.3 release and CHANGELOG.md.

Downloads

File	What	Who
AMDNR-v0.3.3.1.zip	AMDNR (OptiScaler + the lmxxf runtime)	everyone
v0.4.0-Runtime.zip	danielblnc's runtime 0.4.0 (by Daniel Blanco, unmodified, with his permission) + weights. Newest; not yet tested in a game	danielblnc users on RX 9000 who want the newest runtime
v0.3.3-Runtime.zip	danielblnc's runtime 0.3.3 (by Daniel Blanco, unmodified, with his permission) + weights. New, not yet tested in a game	danielblnc users who prefer 0.3.3
Runtime.zip	danielblnc's runtime 0.3.1 (by Daniel Blanco, unmodified, with his permission) + weights	danielblnc users, tested

Coming from 0.3.3? Replace OptiScaler.dll. That is enough for the lmxxf runtime. To use danielblnc 0.4.0 (or
0.3.3), also extract v0.4.0-Runtime.zip (or v0.3.3-Runtime.zip) into the game folder.

Bug reports with logs: https://discord.gg/AMDNR (#bug-report).

AMDNR-v0.3.3

GitHub release — AMDNR v0.3.3

Title: AMDNR v0.3.3 — lmxxf on RX 7000 (RDNA 3), RenoDX colour, lmxxf Full network + RAM fix, XeFG up to 10X, Vulkan fixes

Tag: v0.3.3

---

DLSS 5 Neural Rendering on AMD, built into OptiScaler.

Join the Discord — support, bug reports, test builds: https://discord.gg/AMDNR
(new server — the old invite still works)
Ko-fi: https://ko-fi.com/3zinr

Downloads

File	What	Who
AMDNR-v0.3.3.zip	OptiScaler with AMD Neural Rendering, the complete lmxxf runtime (new kernels, now also for RX 7000), and Nukem9's dlssg-to-fsr3	everyone
Runtime.zip	danielblnc's runtime and weights (v0.3.1, unchanged; this build also drives danielblnc 0.3.2)	RX 7000 / RX 9000 users who pick the danielblnc runtime

Install: extract the zip into the game folder, rename OptiScaler.dll to dxgi.dll (or winmm.dll /
version.dll / OptiScaler.asi with an ASI loader), INSERT for the menu. Replace LmxxfNrRuntime.dll
and LmxxfNrRuntime.pak too - they ship together with this OptiScaler.

What's new

• New Discord server: https://discord.gg/AMDNR — the menu's Discord button opens it; the old invite still works.
• danielblnc 0.3.2 is supported (new, not yet tested in a game): AMDNR now drives danielblnc's 0.3.2
  runtime as well as 0.3.1, so the 0.3.1 Runtime.zip keeps working unchanged. In 0.3.2 only the GPU
  kernels changed; the host side is the same apart from one entry point, and the existing weights file works.
• lmxxf now runs on RX 7000 (RDNA 3) and Strix Halo — our own RDNA 3 backend, by 3zwr1. Slower than on
  RX 9000: start with NR resolution at 67%.
• Colour composition: RenoDX (experimental) on both runtimes, under Neural > Image look — RenoDX's
  colour handling after the model, with a Highlight guard and skin / environment controls. Classic (the
  default) keeps your picture exactly as it was. On lmxxf the Highlight colour guard (now on by default)
  runs before the composition, so it no longer cancels Composition colour on bright pixels (new in this
  build, not yet tested in a game). Composition colour multiplies the saturation that is there, so in
  near-grey scenes such as Silent Hill 2's fog it stays subtle - try 3-4; Composition detail only scales
  the model's own change in brightness. The log now records every slider change.
• lmxxf Full network (Neural > Performance, off by default): runs all 71 of the network's blocks
  instead of skipping three - slightly more faithful, about 0.5 ms slower at 1080p (16.6 -> 17.1 ms on
  an RX 9070 XT).
• lmxxf's RAM leak is fixed, GPU-bound games included: its RAM use climbed for as long as NR ran
  (about 45 GB an hour at 60 NR fps; 14.8 GB in one Silent Hill 2 session). The runtime now also frees
  that memory while the GPU is still busy with the next frame, never waiting on it. On a pipelined probe
  (RX 9070 XT, 1080p, frames in flight like a GPU-bound game) it stays flat over 18000 frames, same
  picture, median frame time about 0.1 ms higher. lmxxf also no longer keeps about 106 MB of VRAM on
  every network rebuild at the 1080 tier. danielblnc is not affected (its runtime already frees it).
• lmxxf colours that suddenly turn grey: fixed: Highlight colour guard and Auto-exposure highlight
  cap (Neural tab) are now on by default - tested in Silent Hill 2. Turn both off to get the 0.3.2
  picture back. Also on: an exposure self-heal for a blown-out feed and a lower auto-exposure floor (not
  yet tested in a game). lmxxf only; danielblnc's equivalent is planned for 0.3.4.
• XeFG frame generation up to 10X, opt-in and paced (6X stays the default; D3D12 games only - XeFG has
  no Vulkan output). Set XeFG ceiling (restart) under FG Output (4X, 6X, 8X or 10X), restart, then pick
  the multiplier in the MFG combo. Above 6X it needs OptiScaler's own XeFG provider with Extra pacing on;
  a game's own XeSS 3 copy stays at 6X. 10X needs a 360 Hz+ display and a frame cap at refresh / 10;
  latency is high, and the provider reserves about 128 MiB more VRAM at 4K. 7X-10X is not yet confirmed
  in a game: testers, please send OptiScaler.log.
• XeFG: new "Allow dilated motion vectors" option.
• Multi-frame generation stutter fix, 3X-10X (not yet tested in a game): the ~1 fps lock after XeFG
  turns on, after FG off/on or a multiplier change, is now bounded, and the overlay no longer resets
  command allocators the GPU still uses. Testers at 4X-10X: please send OptiScaler.log.
• lmxxf on Vulkan games no longer freezes (Indiana Jones; the cause was a lazy weight upload inside
  OptiScaler's Vulkan bridge - expect a one-time ~1 s hitch when NR starts). If lmxxf is held back on a
  Vulkan title, the Neural tab's Retry lmxxf button clears it.
• danielblnc on Vulkan games (Indiana Jones): Neural passes 2-3 now run as 1 pass (each extra pass
  used to stall about 5 s, then NR stopped), the 80 ms post-submit wait is gone, and a note under the
  runtime choice explains its log lines. New opt-in [DlssNr] Amd (experimental, off by
  default, not yet tested in a game): expected to remove the ~5 s pause on the first NR frame - testers,
  set it to true and send OptiScaler.log, amd_presr.log and dlssnr_on_amd.log.
• Streamline start-up hardened, with diagnostics: OptiScaler's Streamline plugin hooks can no longer
  turn a plugin that refuses to load into slInit error 0x18, and several null / stale-pointer cases in
  slInit are guarded. Seen with NBA 2K27 on AMD, but not confirmed as its cause: OptiScaler.log now logs
  slInit returned ... and [SLINIT] lines - send it with any Streamline start-up error.
• lmxxf is faster: lmxxf 0.29's kernels and options, bit-exact, about 16.7 ms instead of 20 ms per
  network run at 1080p on an RX 9070 XT.
• FSR Ray Regeneration path-traced profile, automatic for Resident Evil Requiem and PRAGMATA: less
  grain on faces under path tracing, plus an experimental skin smoothing and A/B controls.
  Menu: Neural > Quality > Ray Regeneration.
• Highlight proxy (danielblnc, experimental, off by default): highlights are decoded from the original
  frame, so they no longer clip at 200 and the model's dimming is no longer magnified.
• RR controls now in the AMD menu, shown only while the game runs Ray Reconstruction.
• UE5 robustness fixes for lmxxf.
• Logs: the previous session's log is kept (OptiScaler.previous.<exe>.log) and a crash is flagged
  ("no clean exit recorded"); both runtimes' logs now carry VRAM, budget and process memory figures.
• Fixes: danielblnc after Ray Regeneration no longer applies the camera jitter to the finished image;
  a failed XeFG swapchain init no longer keeps its context (about 0.8-1 GB of VRAM at 4K); menu text fixes.

Everything else: CHANGELOG.md in the zip.

Known issues

• lmxxf: each NR resolution or DLSS mode change keeps about 97 MB of VRAM and as much RAM at the 1080
  tier until the game restarts (an AMD HIP driver leak; the fix is planned for 0.3.4).
• danielblnc: each new NR size above about 1 MP keeps about 75-350 MB of VRAM per Neural pass until the
  game restarts (read from the runtime's code, not yet measured in a game). The NR resolution slider now
  moves in 5% steps to keep the number of sizes down; restart the game after many changes.

Copyright

AMDNR is Copyright (c) 2026 3zwr1 (AMDNR), a GPL-3.0 fork of OptiScaler. Any copy, fork or derivative
that uses AMDNR's work must keep its notices and credit AMDNR by 3zwr1 (GPL-3.0 section 7(b); see
Licenses/AMDNR_NOTICE.txt in the zip). The upstream work below stays with its authors, under their own
licences. The source code will be published here with AMDNR 0.5.0.

Credits

• TheAutomatic — DLSS 5 AMD project — https://github.com/TheAutomatic/dlss-5-amd-project
• Dagherbou, wilsjo2 and Matheus — the DLSS-NR and pre-SR code this tree descends from (GPL-3.0)

---

• danielblnc — DLSS-NR on AMD — https://github.com/danielblnc/DLSS-NR-on-AMD (Runtime.zip, unmodified)

The danielblnc runtime is Daniel Blanco's work. The AMD neural runtime in Runtime.zip
(dlssnr_amd_pass1..3.dll) is DLSS-NR on AMD by Daniel Blanco (danielblnc) -
https://github.com/danielblnc/DLSS-NR-on-AMD. Copyright (c) 2026 Daniel Blanco, all rights reserved.
AMDNR ships it unmodified, with his permission; it is not AMDNR's work. Please support his project.

---

• lmxxf — https://github.com/lmxxf/dlss5-on-amd-9070xt-porting (the HIP runtime and kernels, MIT)
• Nukem9 — dlssg-to-fsr3 — https://github.com/Nukem9/dlssg-to-fsr3 (GPLv3, unmodified)
• RenoDX — clshortfuse — https://github.com/clshortfuse/renodx (colour composition maths, MIT)
• Coldwood1026 — XeFGUnlock (GPL-3.0), the base of the built-in XeFG multi-frame unlock and its pacing
• burak113 — the FSR Ray Regeneration preprocessor (OptiScaler branch ffx-denoise-experimental, GPL-3.0)
• OptiScaler — Overclockers — https://github.com/Overclockers/OptiScaler-Releases

AMDNR-v0.3.2

GitHub release — v0.3.2

Title: AMDNR v0.3.2 — Vulkan games work, lmxxf colours fixed, frame generation on Vulkan

Tag: v0.3.2

---

DLSS 5 Neural Rendering on AMD, built into OptiScaler. 0.3.2 fixes what you reported on 0.3.1 and
adds frame generation for Vulkan games.

Join the Discord — support, bug reports, test builds: https://discord.gg/AMDNR
Ko-fi: https://ko-fi.com/3zinr

Downloads

File	What	Who
AMDNR-v0.3.2.zip	OptiScaler with AMD Neural Rendering, the complete lmxxf runtime, and Nukem9's dlssg-to-fsr3	everyone
Runtime.zip	danielblnc's runtime and weights (v0.3.1, unchanged)	RX 7000 / RX 9000 users who pick the danielblnc runtime

Install: extract the zip into the game folder, rename OptiScaler.dll to dxgi.dll (or winmm.dll /
version.dll / OptiScaler.asi with an ASI loader), pick a runtime on the first launch, INSERT for
the menu. README in English, 中文, Português, Español.

What's new

• Vulkan games (Indiana Jones and the Great Circle): the game starts, and the lmxxf runtime no
  longer freezes it.
• Frame generation on Vulkan: Nukem9's dlssg-to-fsr3 is in the zip as
  OptiScaler\amdnr_dlssg_fsr3.dll. Set FGInput=DLSSG, FGOutput=DLSSG,
  FGNvngxReplacement=Nukems, Dxgi=true, and the game's DLSS Frame Generation option runs FSR 3 FG.
• lmxxf colours match danielblnc's: auto-exposure when the game publishes no exposure texture
  (AmdLmxxfAutoExposure, on by default). The "Edit colour 0" workaround is not needed any more.
• Neural runtime combo: switching is saved at once; a half-installed lmxxf is named.
• Discord and GitHub buttons at the top of the menu.
• Ray Reconstruction: a game FSR-RR cannot serve (Satisfactory) says why in the Neural tab; RR
  tuning (AMD's defaults / disocclusion threshold) is live in the menu — RE Requiem skin ghosting:
  try "AMD's default tuning".

Everything else: CHANGELOG.md in the zip.

Credits

• TheAutomatic — DLSS 5 AMD project — https://github.com/TheAutomatic/dlss-5-amd-project
• danielblnc — DLSS-NR on AMD — https://github.com/danielblnc/DLSS-NR-on-AMD (Runtime.zip, unmodified)
• lmxxf — https://github.com/lmxxf/dlss5-on-amd-9070xt-porting (the HIP runtime, MIT)
• Nukem9 — dlssg-to-fsr3 — https://github.com/Nukem9/dlssg-to-fsr3 (GPLv3, unmodified)
• OptiScaler — Overclockers — https://github.com/Overclockers/OptiScaler-Releases

AMDNR-v0.3.1

GitHub release — AMDNR v0.3.1

Title: AMDNRv0.3.1 — fixes from the first 0.3.0 reports + NR style presets

Tag: v0.3.1

---

DLSS 5 Neural Rendering on AMD GPUs, built into OptiScaler. This is a fix release for 0.3.0: if 0.3.0
did nothing for you, start here. Many thanks to TheAutomatic (DLSS 5 AMD project), whose releases
and HIP toolchain the lmxxf runtime integration builds on.

We need your support — join the Discord: https://discord.gg/AMDNR · Ko-fi: https://ko-fi.com/3zinr

README in English | 中文 | Português | Español

Downloads

File	What it is	Who needs it
AMDNR-v0.3.1.zip	OptiScaler with AMD Neural Rendering, both runtimes' host code, and the complete lmxxf runtime (LmxxfNrRuntime.dll + LmxxfNrRuntime.pak)	everyone
Runtime.zip	danielblnc's neural runtime and weights (v0.3.1, unmodified)	RX 7000 and RX 9000 users who pick the danielblnc runtime

Install as in 0.3.0: extract the zip into the game folder, rename OptiScaler.dll to dxgi.dll (or
winmm.dll / version.dll / OptiScaler.asi with an ASI loader - need one of those),
pick a runtime on the first launch, INSERT for the menu, HOME toggles Neural Rendering.

Fixed

• lmxxf installed on its own never ran - in any game. The 0.3.0 zip ships lmxxf alone, and the
  neural pass only opened for danielblnc's files, so an lmxxf-only install fell through to the
  NVI path (n.dll is missing in the log, menu stuck at "waiting for a DirectX 12
  SR frame"). Reported in Neverness to Everness; it was every game.
• Where Winds Meet, danielblnc runtime: NR silent. Our shaders were being
  compiled with the game's own old d3dcompiler_47.dll, which fails on the temporal pass's shader
  since 0.2.x and that stopped the whole backend. All our shaders now use the system compiler;
  the temporal pass is optional if anything else fails.
• Crash on a DLSS quality change with lmxxf (GTA V Enhanced, RDR1 reports): the temporal pass
  reallocated textures the GPU could still be reading. The queue is drained first.
• Satisfactory with DLSS Ray Reconstruction on: NR did nothing. The game asks for RR but
  publishes none of its inputs; the handle falls back to FSR, and the neural pass now runs on it
  as on any SR title instead of waiting behind an RR that is not there.
• Private AMD runtime hash mismatch: pass 1 / pass1? in the menu (Skyrim SE report): the pass DLLs
  were a 0.2.16 set this host does not drive. The status line now says which build it found and what
  to install. Runtime.zip here is danielblnc's 0.3.1: dlssnr_amd_pass1.dll = 7,304,192 bytes,
  SHA256 b108d640....

Added

• NR style presets (Neural tab, under Preset): Default, Cinematic, Crisp, Natural, Vivid, and
  three custom style slots saved in the ini. A style sets the look controls together; the combo
  reads Custom when the sliders match no style.

Known

• Final image mode (games with FSR 1 or no upscaler at all) is still off; the work in progress for
  it is in the code for 0.4.0.
• RX 7000, RX9000 (RDNA 3, RDNA 4) runs the danielblnc runtime only; handheld APUs and RX 6000 have no runtime yet.
• Where Winds Meet: spoofing as DLSS does not work there (the game's native upscaler is FSR 2); use the
  XeSS input, as the reporter does.

Credits

• TheAutomatic — DLSS 5 AMD project — https://github.com/TheAutomatic/dlss-5-amd-project —
  many thanks: the releases, the HIP toolchain and the layout this build's lmxxf integration
  follows.
• DLSS-NR on AMD — danielblnc — https://github.com/danielblnc/DLSS-NR-on-AMD — the runtime and
  weights in Runtime.zip, unmodified.
• lmxxf / dlss5-on-amd-9070xt-porting — https://github.com/lmxxf/dlss5-on-amd-9070xt-porting —
  the HIP neural runtime (MIT) built in-tree as LmxxfNrRuntime.dll.
• Matheus / dlss-5-amd — https://github.com/MatheusGViana/dlss-5-amd-project
• OptiScaler — Overclockers — https://github.com/Overclockers/OptiScaler-Releases

AMDNR-v0.3.0

DLSS 5 Neural Rendering on AMD GPUs, built into OptiScaler: two neural runtimes (danielblnc, lmxxf), model interleave, residual composition, XeSS frame generation up to 6X, FSR Ray Regeneration. We need your support — join the Discord: https://discord.gg/AMDNR · Ko-fi: https://ko-fi.com/3zinr

Release title

AMDNRv0.3.0 — DLSS 5 Neural Rendering on AMD, now with two runtimes

Release description

DLSS 5 Neural Rendering running on AMD GPUs, built into OptiScaler so it works in any Direct3D 12
game OptiScaler already hooks (Direct3D 11 and Vulkan titles through its D3D12 bridge).

Many thanks to TheAutomatic — the DLSS 5 AMD project releases, the HIP toolchain work
(rtc_compile, the module recipes) and the asset layout are what made the second runtime in this
build possible. This release stands on his work as much as on ours.

We need your support

Join the Discord server: https://discord.gg/AMDNR — help, #bug-report, test builds,
and the place where the next build gets decided. Two runtimes, dozens of titles and a handful of
testers: every report with a log, every screenshot, every "works here" makes the next release
better for everyone. available there too if you need it (not in
these archives; the AMD path does not need it).

Read me : English | 中文 | Português

Support the project: https://ko-fi.com/3zinr

Downloads

File	What it is	Who needs it
AMDNR-v0.3.0.zip	OptiScaler with AMD Neural Rendering, both runtimes' host code, and the complete lmxxf runtime (LmxxfNrRuntime.dll + LmxxfNrRuntime.pak, 382 MB of weights and modules)	everyone
Runtime.zip	danielblnc's neural runtime and weights (v0.3.1, unmodified)	RX 7000 and RX 9000 users who pick the danielblnc runtime

Install

1. Download AMDNR-v0.3.0.zip and extract it. Copy everything from it into the game folder
   (next to the game's .exe).
2. Pick a runtime (both is fine; the menu lets you switch):
   • lmxxf (RX 9000 only): nothing more to do - LmxxfNrRuntime.dll and LmxxfNrRuntime.pak
     came with step 1.
   • danielblnc (RX 7000 and RX 9000): extract Runtime.zip and copy its files into the game
     folder too.
3. Rename OptiScaler.dll to dxgi.dll. If the game does not launch or the mod does not load,
   try d3d12.dll, winmm.dll, version.dll or dbghelp.dll instead — one name at a time, never
   several copies.
4. Launch the game and press INSERT for the menu. On the first launch that finds both runtimes
   it asks which one to use; the Neural tab changes it later (takes effect on the next start).

Still not working? Report it in #bug-report with the .log files from the game folder
(lmxxf_backend.log when lmxxf is selected — its lmxxf inputs: and lmxxf stats @N: lines
usually say why).

What's new in 0.3.0

• A second neural runtime: lmxxf (HIP, RDNA 4). Selectable beside danielblnc's; the menu
  says what your GPU can run. Its edit is applied one frame late and carried by the motion
  vectors, so the frame never waits for the network (about 32 ms a job at 1080p, 15 ms at the
  720 tier, on an RX 9070 XT).
• LmxxfNrRuntime.pak: the 463-file assets folder is one encrypted, authenticated file
  beside the runtime DLL; nothing is unpacked to disk. The folder layout still works.
• The network's own temporal history (DLSS 5's history path), working under Model
  interleave through a two-frame vector chain, with upstream's history guard and output-side
  smoothing.
• Real Neural passes (2 and 3 launches per frame); the history loop stays a one-pass loop,
  so passes no longer compound into ghosting.
• Edit shaper under Image look: Edit detail, Edit colour, Edge guard, Output smoothing; the
  Appearance filter runs under lmxxf too.
• After Ray Reconstruction: lmxxf now writes its result back into the RR output (new, not yet
  confirmed in a shipping title — try it with RR on and report).
• Toggle hotkey: HOME switches Neural Rendering on and off in-game (both runtimes), with an
  On / Off notice; rebind it beside the Enable checkbox or under Interface > Keybinds.
• Diagnostics for any title: lmxxf_backend.log gets an lmxxf inputs: line (formats,
  sizes, motion scale, depth, mask, exposure) and a stats line every 600 frames (exposure, fed
  brightness, the model's edit, the carried edit, keep, reactive mean, vector health).
• Self-healing: unusable motion vectors, a depth test that refuses everything, or an
  exposure that blacks out the feed are each detected from those stats and worked around for the
  session, with one line in the log saying so.
• Fixed for both runtimes: a title's one-channel reactive mask read as "everything is
  reactive" (GTA V Enhanced showed no effect at all; fixed), the NR-resolution crash (the job in
  flight is abandoned instead of consumed), the NR % slider snapping back, a Reset flag held up
  by the game, the menu's hip_ms readout, the product name (AMD-NR v0.3.0 / OptiScaler v11.0.0), the GPU support readout, Neural passes under NR resolution, Debug view 5 (the
  edit applied directly, an A/B for "it does not work").

Full notes: RELEASE-NOTES.md. Compatibility table for lmxxf: README.md.

Known

• RX 7000, RX9000 (RDNA 3, RDNA 4) runs the danielblnc runtime only; handheld APUs and RX 6000 are not
  supported by either runtime yet.
• The after-RR placement of lmxxf, Vulkan titles with lmxxf, and the TLOU / Where Winds Meet /
  RE Requiem RR reports are still waiting for logs.

Credits

• TheAutomatic — DLSS 5 AMD project — https://github.com/TheAutomatic/dlss-5-amd-project —
  many thanks: the releases, the HIP toolchain and the layout this build's lmxxf integration
  follows.
• DLSS-NR on AMD — danielblnc — https://github.com/danielblnc/DLSS-NR-on-AMD — the runtime and
  weights in Runtime.zip, unmodified.
• lmxxf / dlss5-on-amd-9070xt-porting — https://github.com/lmxxf/dlss5-on-amd-9070xt-porting —
  the HIP neural runtime (MIT) built in-tree as LmxxfNrRuntime.dll.
• Matheus / dlss-5-amd — https://github.com/MatheusGViana/dlss-5-amd-project
• OptiScaler — Overclockers — https://github.com/Overclockers/OptiScaler-Releases

Licence: GPL-3.0 (LICENSE); third-party licences in Licenses\. Not affiliated with NVIDIA, AMD
or any publisher. Use at your own risk.

AMDNR-v0.2.1

AMDNR v0.2.1 — DLSS 5 Neural Rendering on AMD

Release description

DLSS 5 Neural Rendering running on AMD GPUs, built into OptiScaler so it works in any Direct3D 12
game OptiScaler already hooks.

Discord: https://discord.gg/AMDNR — support, #bug-report, test builds, and NV
`` if you need it (not in these archives; the AMD path does not need it).

What's new in 0.2.1

Small fixes
Added new stuff you can try it before 0.3.0

Support the project

Kofi: https://ko-fi.com/3zinr

Credits

• DLSS-NR on AMD — danielblnc — https://github.com/danielblnc/DLSS-NR-on-AMD
• DLSS 5 AMD project — TheAutomatic — https://github.com/TheAutomatic/dlss-5-amd-project
• Matheus / dlss-5-amd — https://github.com/MatheusGViana/dlss-5-amd-project
• lmxxf / dlss5-on-amd-9070xt-porting — https://github.com/lmxxf/dlss5-on-amd-9070xt-porting / 0.3.0
• OptiScaler — Overclockers — https://github.com/Overclockers/OptiScaler-Releases

Runtime.zip is danielblnc's v0.3.1 runtime and weights, unmodified. Licence: GPL-3.0 (LICENSE);
third-party licences in Licenses\. Not affiliated with NVIDIA, AMD or any publisher. Use at your
own risk.