Official press-kit image © Team17 / Expression GamesStart here
- Record the GPU, CPU, memory, display resolution, current game build and the frame-rate target you actually need.
- Run one repeatable scene and write down the baseline before changing anything.
- Compare Master Quality first, then image reconstruction. A current full-release comparison places those two controls ahead of the remaining quality options for performance impact.
- If the GPU supports both paths, compare TSR and DLSS at the same output resolution. Do not compare two different resolutions and call the result a settings win.
- Test FX, reflections and Hardware Lumen only after the first comparison. Keep texture quality separate because available VRAM changes that decision.
- Repeat the same scene. Keep one change only when the result is clearly better for the stated target.
What the full-release tests can—and cannot—tell us
One independent full-release comparison reports Master Quality as its largest gain outside image reconstruction on an RTX 3080 Ti. Because the accessible article is missing CPU and resolution context, its 91-to-138 FPS observation is useful for ordering the first test only; it is not a portable FPS promise.
A separate full-release benchmark used MSI Afterburner, maximum graphics quality and native rendering at three resolutions. Its published 60-FPS average thresholds show how quickly native maximum quality becomes demanding:
| Native test resolution | Reported GPU class for at least a 60-FPS average | What to do with this result |
|---|---|---|
| 1920×1080 | Radeon RX 7800 XT or GeForce RTX 5060 Ti class | Treat maximum native quality as a demanding comparison point, not the required starting preset. |
| 2560×1440 | Radeon RX 9070 XT or GeForce RTX 4070 Ti Super class | Test reconstruction before assuming individual low settings will solve a resolution-bound result. |
| 3840×2160 | GeForce RTX 5090 class | Native maximum quality is not a sensible universal target; record reconstruction and Frame Generation separately. |
These are third-party full-release observations, not Field Signal testing. Different scenes, patches, drivers and CPUs can change the result.
Choose features by constraint
- TSR and DLSS: the official performance page presents both as image-reconstruction options. The correct choice depends on supported hardware, output resolution and the result you measure.
- Frame Generation: officially supported on compatible NVIDIA GeForce RTX 40-series cards and newer. The same source warns that a low base frame rate can increase input latency.
- Hardware Lumen: described by the official performance page as an optional, high-end lighting workload that needs GPU headroom. The full-release comparison found a smaller hit than during beta testing, but that single test does not make it a default recommendation.
Measure one variable
- 1Baseline
Same sceneSame routeSame duration
- 2Change one
Change one settingKeep everything else exactly the same
- 3Repeat
Same sceneSame routeSame duration
- 4Keep / undo
Keep if betterUndo if worse
Equal in importance
| Build | Resolution | Setting changed | Observation | Result |
|---|---|---|---|---|
If a setting does not persist after relaunch
The current official launch thread has acknowledged settings-persistence problems, but it does not authorise deleting configuration files, editing .ini files, changing permissions or disabling cloud saves. Keep the recovery check reversible:
- Record the platform, current build and the original setting.
- Change one setting and apply it through the visible game interface.
- Relaunch once and compare the displayed value with the record.
- If it reverted, restore the original value and stop stacking local changes.
- Recheck the current official thread, then report the platform, build, original value, tested value and relaunch result.
When to stop tuning
Keep a change only when the repeated run is meaningfully better for your stated goal and does not introduce a worse tradeoff. Undo unclear results. After a game, driver or operating-system update, start a new row rather than averaging unlike builds.