For most players, DLSS Quality is the best place to start at 1080p or 1440p, while DLSS Balanced or Performance makes more sense at 4K when a demanding game needs extra GPU headroom. Ultra Performance uses such a low input resolution that it should usually be reserved for 4K or 8K—and even then, only when the other modes cannot reach your performance target.
The preset name does not tell the whole story. DLSS Performance at 4K starts with roughly 1920 × 1080 pixels, but the same preset at 1440p starts at approximately 1280 × 720. That is why a mode that looks impressively clean at one output resolution can appear soft or unstable at another.
Quick recommendation
| Output resolution | Start with | Move lower when… |
|---|---|---|
| 1920 × 1080 | Quality | Balanced only if the image remains clean and you need more FPS |
| 2560 × 1440 | Quality | Balanced if ray tracing or a heavy GPU load prevents your target frame rate |
| 3840 × 2160 | Quality or Balanced | Performance for demanding ray tracing, path tracing or high-refresh gaming |
| 7680 × 4320 | Performance | Ultra Performance if the game remains GPU-limited |
These are starting points, not universal rules. The game, DLSS model, GPU, monitor size, viewing distance and image content can all change the best choice. Use the recommendations to choose your first setting, then test it during real movement rather than judging a static screenshot.
DLSS Quality vs Balanced vs Performance vs Ultra Performance
DLSS Super Resolution constructs a higher-resolution output from a lower-resolution input using current-frame information, motion vectors and data from previous frames. NVIDIA describes it as sampling multiple lower-resolution images and using temporal feedback to reconstruct detail. It runs on the Tensor Cores in GeForce RTX graphics cards. This is separate from Frame Generation, which creates additional displayed frames.
| Mode | Typical linear scale | Pixels rendered | Best use | Main compromise |
|---|---|---|---|---|
| DLAA | 100% | 100% | Maximum image quality when performance is already sufficient | No upscaling performance benefit |
| Quality | About 66.7% | About 44.4% | 1080p and 1440p; image-quality priority at 4K | Smallest performance gain of the upscaling modes |
| Balanced | About 58% | About 33.6% | 1440p when necessary; a strong middle ground at 4K | More reconstruction artifacts than Quality |
| Performance | 50% | 25% | 4K ray tracing, path tracing and high-refresh targets | Fine detail and motion stability can deteriorate |
| Ultra Performance | About 33.3% | About 11.1% | 8K, or selected 4K uses with newer DLSS models | Very little source information remains |
The percentages are measured across each dimension, not by total pixel count. Performance mode uses 50% of the width and 50% of the height, which means it renders only 25% as many source pixels. Ultra Performance uses about one-third of each dimension and therefore renders only around one-ninth of the output pixel count.
Those reductions do not translate directly into fixed FPS gains. DLSS has its own processing cost, and parts of a game’s workload do not scale with resolution. A CPU-limited game may gain almost nothing when you move from Quality to Performance, while a heavily GPU-limited path-traced game may gain substantially. Any article promising one universal percentage for each mode is oversimplifying the result.
DLSS render resolution calculator
Select your output resolution and DLSS mode to estimate the resolution rendered before reconstruction. Games may round dimensions differently or use dynamic resolution, so treat the result as a close reference.
Quality is normally the safest DLSS starting point at 1440p.
Common DLSS internal resolutions
| Output | Quality | Balanced | Performance | Ultra Performance |
|---|---|---|---|---|
| 1920 × 1080 | 1280 × 720 | 1114 × 626 | 960 × 540 | 640 × 360 |
| 2560 × 1440 | 1707 × 960 | 1485 × 835 | 1280 × 720 | 853 × 480 |
| 3440 × 1440 | 2293 × 960 | 1995 × 835 | 1720 × 720 | 1147 × 480 |
| 3840 × 2160 | 2560 × 1440 | 2227 × 1253 | 1920 × 1080 | 1280 × 720 |
| 7680 × 4320 | 5120 × 2880 | 4454 × 2506 | 3840 × 2160 | 2560 × 1440 |
Balanced mode is commonly described as approximately 58%, so its dimensions may differ slightly between games. Newer NVIDIA app overrides can also expose custom scaling percentages instead of limiting you to the traditional presets.

Which DLSS mode should you use?
Best DLSS mode at 1080p
Start with DLSS Quality. Its approximate 1280 × 720 input is already modest, so Balanced and Performance have much less source information from which to reconstruct fine detail. Hair, foliage, wires, particle edges and objects newly revealed from behind another surface can become soft, noisy or unstable in motion.
If your GPU already delivers the frame rate you want, compare Quality with native rendering or DLAA. DLAA uses the DLSS reconstruction system at native input resolution, prioritising anti-aliasing and image stability instead of performance. Native rendering is not automatically superior: a good reconstruction model can sometimes handle fine edges or temporal stability better than a game’s native TAA.
Performance and Ultra Performance are generally poor choices at 1080p. Ultra Performance would reconstruct the whole image from approximately 640 × 360. A newer model can do impressive work with that input, but it cannot recover reliable information that never existed.
Best DLSS mode at 1440p
Quality is the safest default at 2560 × 1440, reconstructing from approximately 1707 × 960. Balanced can be worthwhile if you are GPU-limited and need a little more performance, particularly on a smaller monitor or at a greater viewing distance.
Performance mode drops the input to 1280 × 720. It may still be acceptable in a strong implementation, but it is not a universal 1440p sweet spot. Check foliage, fences, distant detail, transparencies and moving characters before keeping it enabled. Ultra Performance at 1440p begins around 853 × 480 and is difficult to recommend.
Your monitor’s size and viewing distance influence how visible these compromises become. Our monitor size and resolution calculator can help you compare pixel density and viewing distance.
Best DLSS mode at 4K
At 4K, all modes begin with more source pixels. Quality reconstructs from 1440p, Balanced from roughly 1253p and Performance from 1080p. That makes Balanced and Performance substantially more viable than they are at lower output resolutions.
Start with Quality if your frame rate is already close to its target. Move to Balanced when the game is clearly GPU-limited. Performance becomes a sensible option for demanding ray tracing, path tracing or high-refresh 4K gaming, especially with the newer transformer models. Independent image-quality testing by TechSpot found Balanced or Performance to be the general DLSS 4 sweet spot at 4K, while Quality or Balanced remained preferable at 1440p.
DLSS 4.5 has made Ultra Performance more credible at 4K than it was with older models, but it still reconstructs from only 1280 × 720. Treat it as an additional option when Performance cannot reach your target—not as the automatic best setting.
What changes visually between DLSS modes?
A lower input resolution does not simply make the whole image uniformly blurrier. Reconstruction errors are more likely to appear in particular types of detail:
- Thin geometry: wires, fences, branches and railings may shimmer, break apart or lose definition.
- Disocclusion: newly revealed pixels behind a moving character or object have little temporal history and may show trails or instability.
- Fine repeating patterns: grilles, mesh, patterned clothing and distant architecture can alias or produce moiré-like artifacts.
- Hair and foliage: fine strands and leaves may appear grainy or unstable during motion.
- Transparent effects: holograms, particles and effects without ideal motion-vector data can be difficult to reconstruct.
- Text and interface elements: a poor integration can affect UI clarity, although modern DLSS integrations generally separate interface rendering from reconstruction.
Static screenshots can hide these weaknesses because temporal upscalers accumulate information across frames. Test while moving the camera, driving quickly, passing behind foreground objects and viewing particle-heavy scenes. The best mode is the lowest one that remains visually stable during the way you actually play.
How DLSS 4.5 changes the comparison
DLSS 4.5 Super Resolution introduced NVIDIA’s second-generation transformer model. NVIDIA says it uses five times the compute of the first transformer model and improves temporal stability, anti-aliasing, motion clarity, lighting detail and ghosting. The largest changes are intended for Performance and Ultra Performance, where the model has fewer source pixels available.
Through the NVIDIA app, supported games can use recommended model overrides:
- Model K: the recommended model for DLAA, Quality and Balanced.
- Model M: optimised for Performance mode.
- Model L: optimised for 4K Ultra Performance mode.
All GeForce RTX generations can use DLSS 4.5 Super Resolution in supported titles, but hardware differences matter. RTX 40- and 50-series GPUs can accelerate the new models using FP8 precision. RTX 20- and 30-series GPUs lack native FP8 support, so NVIDIA warns that Models M and L carry a heavier performance cost on those cards. Owners of older RTX hardware may find that Model K provides a better overall balance.
The NVIDIA app also allows preset or custom Super Resolution overrides in compatible games. This makes it possible to choose a scaling level between the traditional modes when, for example, Quality is slightly too slow but Balanced gives away more image quality than necessary.
DLSS Super Resolution is not Frame Generation
DLSS has become an umbrella for several distinct technologies. Keeping them separate prevents misleading FPS comparisons:
- Super Resolution reconstructs a high-resolution frame from a lower-resolution render. Quality, Balanced, Performance and Ultra Performance are Super Resolution modes.
- DLAA uses the reconstruction technology at native input resolution for anti-aliasing and image quality.
- Frame Generation inserts an AI-generated frame between conventionally rendered frames on supported RTX 40- and 50-series GPUs.
- Multi Frame Generation can insert multiple generated frames on RTX 50-series GPUs.
- Ray Reconstruction replaces multiple hand-tuned denoisers in ray-traced rendering.
DLSS 4.5 can generate up to five additional frames for each conventionally rendered frame in its 6X Multi Frame Generation mode on RTX 50-series hardware. That does not mean Ultra Performance Super Resolution alone multiplies FPS sixfold. Generated frames improve displayed smoothness, but they do not update game simulation or player input like conventionally rendered frames. Our guide to frame generation examines that distinction in detail.
How to choose the right mode on your own PC
- Set a useful target. Choose the frame rate you actually need for your monitor and game rather than chasing the highest number available.
- Begin with Quality. Test at your normal graphics settings in a demanding part of the game.
- Confirm that you are GPU-limited. If GPU utilisation is not close to its limit, a lower DLSS mode may produce little benefit because the CPU or another component is holding back performance.
- Move down one mode at a time. Try Balanced before Performance. Stop as soon as your target becomes stable.
- Inspect motion, not just FPS. Look for shimmer, trails, unstable foliage, soft textures and UI problems while moving.
- Check latency separately. Super Resolution can reduce latency by raising the conventionally rendered frame rate, while Frame Generation has a different latency trade-off. See our guide to whether DLSS increases input lag.
- Retest after major updates. A game patch, driver update or model override can change both performance and image quality.
Competitive players should normally protect responsiveness and motion clarity rather than enabling every visual feature. Our guide to the best DLSS settings for competitive games focuses on that use case.
Does a lower DLSS mode reduce power consumption?
It can, but only if the GPU does not use the freed capacity to render more frames. With an uncapped frame rate, moving to Performance may allow the GPU to produce additional FPS while remaining near its power limit. Combining DLSS with a sensible frame-rate cap is more likely to reduce power, heat and fan noise.
You can compare the result using measured gaming power and frame rate in our GPU efficiency calculator. Test each mode in the same scene and keep ray tracing, Frame Generation and the frame cap consistent.
Frequently asked questions
Is DLSS Quality better than Performance?
DLSS Quality normally produces a cleaner and more stable image because it begins with more source pixels. Performance provides more GPU headroom. Quality is generally preferable at 1080p and 1440p, while Performance becomes much more useful at 4K.
Is DLSS Balanced worth using?
Yes. Balanced is useful when Quality falls just short of your target frame rate. It is particularly effective at 4K, where its approximate 2227 × 1253 input retains substantially more information than Performance mode’s 1920 × 1080.
Should I use DLSS Performance at 1440p?
Only if Quality and Balanced cannot deliver the performance you need and the image remains acceptable in that particular game. Performance at 1440p reconstructs from approximately 1280 × 720, so artifacts are more visible than they are with Performance at 4K.
Is DLSS Ultra Performance only for 8K?
It was originally positioned primarily for 8K, but DLSS 4.5 makes Ultra Performance more usable at 4K. It is still an aggressive mode that uses only about one-ninth of the output pixel count. Avoid it at 1080p and generally at 1440p.
Does DLSS Performance add input lag?
Super Resolution itself often reduces rendering latency when it raises the game’s real frame rate and relieves a GPU bottleneck. Frame Generation is a separate feature with a different latency cost. Do not judge latency from the displayed FPS number when generated frames are enabled.
Does DLSS use less VRAM?
Rendering at a lower internal resolution can reduce the size of some render targets, but it does not proportionally reduce all VRAM use. Textures, geometry, the frame buffer, ray-tracing data, reconstruction history and the game’s asset management still consume memory. Treat any VRAM reduction as game-dependent.
The bottom line
DLSS Quality, Balanced, Performance and Ultra Performance are not quality levels in isolation. They are different amounts of source information supplied to the reconstruction model. The same preset can therefore produce a very different result at 1080p, 1440p and 4K.
Start with Quality, confirm that the GPU is the limiting factor, and move down only until you reach a stable target frame rate. At 1080p, that will usually mean staying with Quality. At 1440p, Quality or Balanced is the sensible range. At 4K, Balanced and Performance become strong options, while DLSS 4.5 makes Ultra Performance worth testing in especially demanding circumstances.
If you are comparing NVIDIA’s reconstruction with AMD’s alternative rather than choosing between DLSS modes, continue with our guide to FSR versus native resolution.




