Careful of fake copies. The only download for this project is this releases page. Fake websites and lookalike GitHub repositories exist (#88, #115); this project has no executable installer. The standalone
CAREFUL_FAKE_MALICIOUS_FEEDER.txtasset explains how to check what you downloaded. SHA-256 ofDLSS5-Feeder-1.17.0.zip:11A96B36AE89EF75B3CFF0E03849DB35591143FE4DF171986502905809469E12. The genuine zips are scanned on VirusTotal, for instance the 1.16.0-beta.2 zip; look this release's hash up there yourself.
There is a one-command installer. Download
Install-DLSS5Feeder.ps1, put it beside the game's executable, and runpowershell.exe -ExecutionPolicy Bypass -File .\Install-DLSS5Feeder.ps1. The standaloneAUTOMATIC_INSTALLATION_AVAILABLE.txtasset explains every option.
Does not work with your game? I can't test every game reported in the issues: I don't own most of them, and I don't have the time to try them all. Read this README section. It gives a prompt for an AI coding agent that reads your logs against the source and opens a pull request if it finds a legitimate fix.
1.17.0 gathers the three 1.17.0 betas into one release. Every binary embeds release commit 03710dd, and the add-ons identify themselves as 1.17.0; check line 1 of dlss5-feed.log before reporting a retest. Since 1.17.0-beta.3 only the version number and one log message changed.
The neural model runs on RTX 50 only. NVIDIA's signed
nvngx_dlssnr.dll310.8.0, and the modified 310.8.Lecram build the installer downloads, create the neural feature on RTX 50-series GPUs only. On an RTX 20/30/40 DLSS still runs but the neural pass does not:ReShade.logsaysfeature 18 create failed with 0xbad00001and you only see DLSS's anti-aliasing (#131). Get the modifiednvngx_dlssnr.dllmade for your GPU from the RenoDX Discord.Verify-DLSS5Feeder.ps1says so at the end.
32-bit games: install
dlss5-feed.addon32andhost64\dlss5-feed-host64.exefrom this zip. The IPC protocol is v11, as in beta.2 and beta.3; a 1.16.x or 1.17.0-beta.1 half refuses to talk to this one, and the log says which side is old.
What changed since 1.16.0-beta.7
Details for each item are in the beta notes: beta.1, beta.2, beta.3.
- wilsjo2's OptiScaler-DLSSNR-PreSR-Multipass is a supported neural consumer (#126), beside Dagherbou's OptiScaler_DLSSNR. The feeder warns about three OptiScaler settings that silently stop the pass (
FinishedPicture=true, aTargetProcessNamenaming another exe, a leftover forwarder DLL). For model resolutions below 100%, use the fixed build v0.8.92 on this project's fork. - RenoDX DLSS 5 v6.1 and newer. v6.1.0, v7.0.0-rc8 and v8.0.1 pass the helper's self-test 300/300 on driver 617.14, where v4.7 still fails (#54); the helper now reads their version and stops warning about them.
- Installer: asks which OptiScaler fork (wilsjo2 by default); downloads the newest RenoDX add-on and neural model from the RHI repository on every run instead of the pinned 4.70; keeps its download cache per release.
- Output stabiliser (experimental, both add-ons, every transport). One compute pass after the evaluate keeps the shown picture where the game's frame did not change and follows the model where it did. DLSS 5 page, Output stabiliser: Hold strength (0 = off, the default) and Change tolerance; README rows
hold_strength/hold_tolerance. - #130 — the same-device D3D12 session no longer opens a second DLSS contract in a game that loaded its own DLSS;
native_dlss_ok=1overrides. - #129 — on D3D9 the effect compiles to a pass-through stub that says what to do, instead of
X3531. - #121 — 32-bit Vulkan games under Wine/Proton stop cleanly instead of crashing on the D3D12 fence import (
wine_fence_import=1tries anyway). The driverOutOfDatehint now mentions that Linux drivers are numbered separately (615.x). - #97 — every D3D12 object the feeder creates is named, so DRED output shows whether a freed resource is the feeder's.
- Diagnostics:
settle_evalsandhold_input; the 64-bit add-on prints VRAM usage against the OS budget with itsSTALLlines.
Verified, and not
All three binaries build; the effect and the Vulkan layers are unchanged since beta.3. The betas' self-tests and game runs are in their notes. Not run in a game: the 64-bit stabiliser. The Lecram model has only been run on an RTX 50.