Before enabling AMDLSS5 on an RX 9070 XT, establish what you gain visually and what you lose in performance. This guide sets out a repeatable comparison against native rendering and unmodified FSR, with a separate check for the mod installed but its neural network disabled.
What AMDLSS5 can tell RX 9070 XT owners
AMDLSS5 is an unofficial neural-rendering mod for AMD cards, with the RX 9070 XT listed as supported. It works on top of a game's upscaler and is unaffiliated with AMD or NVIDIA. As checked on 13 November 2026, version 2.0.0 is the latest listed release. This guide proposes a comparison method; HD OPTI has not benchmarked it or established whether its visual changes justify the cost.
Neural rendering, upscaling and Frame Generation answer different questions. Neural rendering changes lighting and material appearance; upscaling reconstructs an image from a lower rendering resolution; Frame Generation adds intermediate frames. Keep these effects separate when judging the mod. An unofficial implementation should not be assumed to match an official game's integration.
DLSS 5 Explained: Nvidia’s Neural Rendering Could Change Game Lighting Forever
Check the requirements before testing
The current AMDLSS5 documentation lists Windows 10 or 11 in 64-bit form, a recent AMD Adrenalin driver, an AVX2-capable CPU and a compatible DirectX 12 game. Resident Evil Requiem is listed among the supported games; its documented path uses FSR 3.1 or newer. Check the exact game instructions before installing because several other titles use different upscaler settings.
Keep this experiment in a supported offline game. Support for an RX 9070 XT does not establish compatibility with every game or permission to use the mod with anti-cheat. The current documentation also warns about GPU hangs during loading with Adrenalin metrics, recording or instant replay enabled. Follow its preflight guidance: disable those features, set Adrenalin performance tuning to Default and reboot before testing. Record these changes so every comparison uses the same driver configuration.
If you decide to proceed, use the project's release page and current instructions. The documented package is AMDLSS5-2.0.0.zip; extract the complete folder before opening AMDLSS5-Instalador.exe. The installer provides Install, Update and Remove actions. In game, F10 switches the neural network on or off and End opens the mod menu. These instructions describe the project's workflow; this article is not an executable safety audit.
AMDLSS5 releases and installation instructions
Compare four conditions at 1080p and 1440p
Use Resident Evil Requiem as one documented RX 9070 XT testing option. First establish a stable unmodified configuration. Save the game build, driver version, CPU, RAM, graphics preset, ray-tracing settings, resolution, FSR mode and display settings alongside the results. Choose a repeatable route from the same save, including a detailed lighting scene and a controlled camera turn.
Best Resident Evil Requiem Settings for FPS and Stability
| Condition | Configuration | Question it answers |
|---|---|---|
| A: Native reference | Mod removed; native rendering; Frame Generation off | What does the original image and performance look like? |
| B: FSR baseline | Mod removed; fixed FSR mode; Frame Generation off | How does unmodified upscaling compare? |
| C: Installed, network off | Same FSR setup; mod installed; F10 network off; extra filters off | Does the installed path differ from unmodified FSR? |
| D: Neural rendering on | Same setup as C; F10 network on; extra filters off | What changes when the neural network is enabled? |
C is a diagnostic control, not a promise that switching the network off removes all mod overhead. Compare D against C to isolate the toggle as closely as possible, and D against B to assess the complete practical cost of using the mod. Compare all images against A to judge whether the changes suit the game's original appearance.
- Run the full comparison at 1920×1080, then repeat it at 2560×1440. Keep the FSR mode fixed within each resolution and record the actual mode or override in use.
- Disable in-game and driver Frame Generation throughout these first passes. Keep optional mod image filters off; version 2.0 provides separate colour, contrast and lighting filters that would otherwise complicate the comparison.
- Warm up the route before recording. Capture three 60-second passes per condition using the same route, starting point and camera movement. These durations and repetitions are a suggested method.
- Alternate the installed network-off and network-on runs, then repeat the unmodified FSR baseline at the end. Investigate large run-to-run differences before deciding the mod caused a change.
- Keep the frame cap, V-Sync, VRR and recording configuration consistent. For a separate practical check, use your normal frame cap to see whether each condition can sustain it.
Record frame times as well as FPS
Use a frame-capture tool such as PresentMon to save per-frame data. Record average FPS, the same tool's 1% low calculation, a frame-time graph and any repeatable spikes. Keep the raw captures and tool version. Record VRAM usage if your monitoring setup exposes it reliably; do not turn a memory-usage increase alone into a claim that memory caused a hitch.
PresentMon distinguishes frame duration, display timing and input-related latency metrics. Frame-type tracking requires application or driver instrumentation, so do not assume it recognises every generated frame from an unofficial mod. A higher FPS reading does not independently establish lower input latency. If a valid latency measurement is unavailable, record control feel as a subjective observation and leave latency unmeasured.
PresentMon capture tool and documentation
Add Frame Generation only after the network-off and network-on comparison is complete, and label that as a separate test. AMDLSS5's release notes say its counter includes generated frames for the mod's x2 Frame Generation in Red Dead Redemption 2 and Shadow of the Tomb Raider. Do not compare that total directly with a rendered-only baseline or assume it represents twice the responsiveness.
Judge the image during movement
Use matching stills to inspect faces, contact shadows and material detail, then repeat the same camera movement at normal playback speed. Look for unstable edges, brightness changes after menus, detail that trails during turns, and distracting changes after camera cuts. These are inspection targets, not defects HD OPTI has observed in version 2.0.
Make the judgement in two stages. First, compare B and D with optional filters disabled. Then, if you prefer D, enable one extra filter at a time and repeat the quality and performance check. A more dramatic image is only useful if you prefer it in motion and can afford its processing cost.
| Your observation | Next check | Practical decision |
|---|---|---|
| D looks better and sustains your target | Repeat in another demanding scene | Consider keeping it for this game and configuration |
| D looks better but misses the target | Compare D with B at the same cap and inspect frame times | Choose between the image change and the performance you need |
| C is already slower or less stable than B | Recheck the installed configuration and repeat B after removal | Investigate the installed path before blaming the neural network |
| D changes the image in a distracting way | Confirm extra filters are off and repeat the camera turn | Return to B if you prefer the original presentation |
| Only the generated-frame counter improves | Recheck the Frame Generation-off passes | Keep rendered performance and displayed output in separate records |
What would justify an efficiency verdict?
A useful RX 9070 XT result needs the unmodified FSR baseline, repeatable frame-time captures, exact settings and matching motion comparisons. Without those, this guide cannot recommend AMDLSS5 as an FPS upgrade or declare its visual enhancements worth the cost. Treat a successful test as evidence for that game, version, scene and configuration, then repeat it where you actually play.
FAQ
No. It is an unofficial project unaffiliated with AMD or NVIDIA. The project lists the RX 9070 XT as supported, but that does not establish equivalence with an official DLSS 5 game integration.
F10 toggles the neural network. Keeping the mod installed with the network off provides an extra control, but it does not prove the installed path has no overhead. Comparing network-on against both controls helps distinguish the toggle's effect from the full cost of using the mod.
It cannot establish input latency on its own. The mod's counter includes generated frames in its documented x2 Frame Generation modes for Red Dead Redemption 2 and Shadow of the Tomb Raider. Start with Frame Generation off and report latency separately only when it can be measured.
Keep the game build, driver, graphics preset, resolution, FSR mode, ray tracing, display settings and capture method fixed within a comparison. Leave optional mod filters and Frame Generation off while testing the neural network, then evaluate those features separately.
