GeForce RTX 4070 Ti SUPER DLSS 4.5 vs DLSS 5
The desktop GeForce RTX 4070 Ti SUPER has a measurable DLSS 4.5 baseline. Its DLSS 5 position is less complicated than the rumor cycle suggests: NVIDIA hasn’t announced support for the card. Developers sizing a target SKU and owners deciding whether to upgrade should keep those two facts separate.
For the NVIDIA release window covered here, the shipping set is DLSS 4.5 Super Resolution, Ray Reconstruction, and Frame Generation. No official DLSS 5 launch support or later commitment has been announced for the GeForce RTX 4070 Ti SUPER. The benchmark discussion below concerns software that runs today; everything about a later generation remains conditional.
GeForce RTX 4070 Ti SUPER DLSS 4.5 vs DLSS 5: the comparison in one paragraph
DLSS 4.5 works on the desktop GeForce RTX 4070 Ti SUPER today. DLSS 5 does not have official support on this card. That makes the useful comparison a measured 4.5 baseline against the requirements a future release would need to meet, not a same-card performance contest that cannot yet be run.
The current side can be checked with average-FPS measurements at 2560×1440 and with the SDK behavior developers already configure. The future side consists of open questions about internal resolution, motion stability, Ray Reconstruction, Frame Generation, and engine integration.
Measured QHD performance on the exact desktop SKU
The Notebookcheck data for this GPU records three average frame rates at 2560×1440. The values below are reproduced as reported, without smoothing, interpolation, or conversion to another resolution.
The two Cyberpunk 2077 rows come from different game versions on the same GPU and resolution, which explains the different averages. Alan Wake 2 is the heavier path-traced workload, and its 82.5 FPS result is a practical lower reference for this class of QHD card with the title’s requested ray-tracing load.
| Game | Resolution | Average FPS on a desktop GeForce RTX 4070 Ti SUPER | Source |
|---|---|---|---|
| Cyberpunk 2077 1.6 | 2560×1440 (QHD) | 103.6 | Notebookcheck measurement of this exact GPU SKU |
| Cyberpunk 2077 | 2560×1440 (QHD) | 94.9 | Notebookcheck measurement of this exact GPU SKU |
| Alan Wake 2 | 2560×1440 (QHD) | 82.5 | Notebookcheck measurement of this exact GPU SKU |
These results fix the variables that matter: exact desktop GPU, game version, resolution, and measured average. Any future upscaler still has to run within the card’s fixed shader count, memory bandwidth, and optical-flow budget. A broad “RTX 40 series” claim isn’t a substitute for a repeatable test on this SKU.
Why present measurements and future support stay separate
DLSS combines a temporal upscaler with auxiliary passes. Super Resolution handles reconstruction. Ray Reconstruction replaces hand-tuned denoisers with a learned model. Frame Generation uses optical flow to insert an intermediate frame. A new release can change one or several of those paths, along with the driver interface that exposes them.
The SDK, engine plugin, and driver all need to agree before a feature appears on a shipping card. Until NVIDIA publishes DLSS 5 specifications and a supported GPU list, there is no responsible way to attach a DLSS 5 frame rate or quality result to the GeForce RTX 4070 Ti SUPER.
How developers configure DLSS 4.5
In Unreal Engine 5, Unity, and custom renderers, the integration follows the same broad pattern. A plugin or SDK call receives a lower-resolution color buffer plus motion vectors, depth, and optional exposure data, then returns a higher-resolution color buffer for final composition. The main controls are the four Super Resolution presets, Ray Reconstruction, and Frame Generation.
Super Resolution presets
Each preset changes the input render scale and therefore the cost paid before reconstruction. The underlying algorithm doesn’t change in a way most players would spot without a side-by-side comparison:
- Ultra Performance: aggressive internal scale, used when the engine has a heavy fragment or ray-tracing cost and a small fraction of native resolution is enough to feed the network.
- Performance: a balanced default for many 1440p titles, and the most common preset for GeForce RTX 4070 Ti SUPER owners who want stable frame rates with ray tracing on.
- Balanced: a midpoint between Performance and Quality, useful when the developer wants the look of Quality with the cost closer to Performance.
- Quality: a conservative internal scale, used when the developer or the player is willing to spend more GPU time on the base render to preserve detail, especially in fine geometry and text.
These are cost and quality choices, not a simple ranking. A QHD project seeking a stated frame-rate lock with ray tracing may settle on Performance or Balanced. At 1080p, Quality may better preserve edge stability. The target frame time and the actual image decide.
Ray Reconstruction
Ray Reconstruction is the learned replacement for hand-tuned spatiotemporal denoisers used by path-traced lighting. It can stabilize indirect light, reduce ghosting from moving lights, and retain fine geometry at low sample counts. Its cost sits inside the wider ray-tracing budget, which is why Alan Wake 2’s 82.5 FPS QHD result is a useful lower reference for this card under the heaviest path-traced load.
Frame Generation
Frame Generation runs after upscaling. Optical flow between the previous and current rendered frames supplies an intermediate frame that is added to the output. It isn’t free, and the visible frame rate includes both rendered and synthesized frames. An on-screen counter should make that distinction clear.
The feature also interacts with frame pacing. Variable-refresh displays, fixed-rate cinematics, and physics loops tied to a simulation rate all need to distinguish a real frame from a generated one. A higher average can still feel worse if the integration ignores cadence.
Benchmark the workload, not a portable percentage
The 103.6, 94.9, and 82.5 FPS results belong to their specific games, versions, resolution, GPU, and driver. A patch, setting change, ray-tracing budget, or driver update can move them. Optical-flow changes, shader scheduling, and the ray-tracing compiler all feed into the final average.
Production decisions should compare two configurations on the same machine under the same workload. I wouldn’t set a final recommended spec from one launch run. The result has to survive the QA bench, certification workflow, and a player’s machine after release.
Do not transfer these results to another platform
Every figure here belongs to the desktop GeForce RTX 4070 Ti SUPER at 2560×1440. It doesn’t transfer to a laptop GPU, a Max-Q design, another SUPER or Ti card, or a neighboring RTX 40 model.
The same rule applies to DLSS 5. Any release will have its own supported GPU list, driver requirement, and engine integration. Until that list exists, no official support has been announced for this card.
Five questions a future DLSS release must answer
If a DLSS 5 release follows earlier major revisions, it may change one or more of these areas. None has been specified for this GPU, so each remains a question.
1. Does the internal render scale change?
DLSS 4.5 renders below the output resolution and reconstructs the image. A new generation could alter the input-to-output ratio, sharpness, motion-vector weighting, or the way depth locks the result to scene geometry. Any change would affect the base workload, but its size cannot be predicted from the current release.
2. Is fast motion more stable?
Hair, foliage, transparent particles, and thin geometry can ghost or smear when temporal data loses the subject. A new model could handle those cases differently. Whether this card would receive the change depends on the model’s fit for Ada-class tensor cores, which NVIDIA hasn’t published.
3. How does Ray Reconstruction change?
A new denoiser could weight the radiance cache differently or improve glossy reflections. Its cost would sit on top of an already heavy path-tracing budget. Without a specification, there is no honest number to attach to it.
4. What happens to Frame Generation?
A later generation could synthesize two frames per real frame, change optical-flow cost, or add latency compensation for input-bound games. The impact on the GeForce RTX 4070 Ti SUPER depends on how its Ada optical-flow hardware is used, and that detail hasn’t been published.
5. Does the engine interface move?
A new release could require different buffers, shader bindings, plugin builds, or preset handling. The amount of rework depends on how well the engine already isolates the DLSS integration.
Until NVIDIA publishes a specification and names this card, forward-looking FPS, latency, and image-quality claims have no measurement behind them.
Prepare the project without promising the feature
- Keep the DLSS integration behind a feature flag, not hard-coded, so a new SDK can be evaluated without a full engine rebuild.
- Track DLSS quality presets, Ray Reconstruction, and Frame Generation as separate toggles, so a future change to one path can be tested without disturbing the other two.
- Hold a measured DLSS 4.5 baseline in the QA database for the desktop GeForce RTX 4070 Ti SUPER SKU, so the same workload can be re-run when a new driver or a new SDK is available and the difference can be attributed to the change, not to the test setup.
This gives the team its own answer when a supported successor arrives. Third-party figures remain useful because they pin a result to a specific GPU, resolution, and game version, but they don’t replace an internal regression scene.
What the numbers mean for an owner
At QHD with ray tracing, the 82.5 FPS Alan Wake 2 result is a practical lower reference for a heavy path-traced workload. Cyberpunk 2077 1.6 at 103.6 FPS provides an upper reference among the listed results. The 94.9 FPS Cyberpunk 2077 run sits between them. That 82.5 to 103.6 FPS range describes the current operating window recorded by the source, not a promise for every game.
If that current performance meets the workload, there is no need to delay a project or replacement decision for an unsupported feature. If it doesn’t, compare available cards on shipping capabilities rather than assume DLSS 5 will change this one.
RTX 40 series context
The GeForce RTX 4070 Ti SUPER belongs to the Ada Lovelace RTX 40 series. The generation defines its driver cadence, tensor hardware, optical-flow block, and present DLSS 4.5 support. Public NVIDIA 4.5 documentation includes Ada as a supported tier, while the series reference records the release period, SKU lineup, and architecture.
That family context doesn’t confirm DLSS 5. A new generation still needs its own supported GPU list.
What is settled and what remains open
DLSS 4.5 support and the three QHD measurements are settled. DLSS 5 support isn’t. A two-column status record is more honest than a performance comparison because no DLSS 5 measurement exists for this card.
| Topic | Status for the GeForce RTX 4070 Ti SUPER | Source |
|---|---|---|
| DLSS 4.5 Super Resolution | Shipping, supported | Public NVIDIA DLSS 4.5 documentation |
| DLSS 4.5 Ray Reconstruction | Shipping, supported | Public NVIDIA DLSS 4.5 documentation |
| Frame Generation | Shipping, supported | Public NVIDIA DLSS 4.5 documentation |
| DLSS 5 support | No official support has been announced | NVIDIA has not announced DLSS 5 support for this GPU |
| Average FPS in Cyberpunk 2077 1.6 at 2560×1440 | 103.6 | Notebookcheck measurement of this exact GPU SKU |
| Average FPS in Cyberpunk 2077 at 2560×1440 | 94.9 | Notebookcheck measurement of this exact GPU SKU |
| Average FPS in Alan Wake 2 at 2560×1440 | 82.5 | Notebookcheck measurement of this exact GPU SKU |
Other production references on the site
Readers making target-SKU and engine decisions may also be comparing player-facing walkthroughs with production case studies. The closest internal references are a modern JRPG project retrospective and the Subnautica 2 map reference, both of which connect a measured workload to a specific development choice.
Work to schedule for the next quarter
- Re-run the same three benchmark entries on the same desktop GeForce RTX 4070 Ti SUPER SKU on the latest driver and record the new averages. If the numbers move, attribute the change to the driver, not to the DLSS version.
- Track the public NVIDIA DLSS release notes and the SDK changelog for any new feature exposed to the engine integration. A change there is the leading indicator of a future feature flip.
- Hold a single test scene that exercises the four DLSS quality presets, Ray Reconstruction on and off, and Frame Generation on and off, and run it on the same SKU under the same driver whenever the SDK changes. The scene becomes the team’s internal comparison record.
A repeatable scene is the best preparation. If DLSS 5 becomes supported, the team can run the same workload and measure the difference its players will actually see.
Frequently asked questions
Does the GeForce RTX 4070 Ti SUPER officially support DLSS 5?
No. NVIDIA hasn’t announced DLSS 5 support for this card. The shipping set is DLSS 4.5 Super Resolution, Ray Reconstruction, and Frame Generation.
Which DLSS features ship today?
DLSS 4.5 Super Resolution, Ray Reconstruction, and Frame Generation are available through NVIDIA’s public SDK and standard Unreal Engine, Unity, and custom-engine integrations.
What average FPS has been measured at 2560×1440?
Notebookcheck records 103.6 FPS in Cyberpunk 2077 1.6, 94.9 FPS in Cyberpunk 2077, and 82.5 FPS in Alan Wake 2 on the desktop GeForce RTX 4070 Ti SUPER at QHD.
Can anyone quote a DLSS 5 uplift for this card?
No. A card-specific uplift requires official support and a measurement. Neither exists for DLSS 5 on this GPU.
Do these figures apply to a laptop?
No. They are desktop results. Laptop and Max-Q variants have different power, thermal, and clock behavior and need separate tests.
What would DLSS 5 need to change?
It could affect internal render scale, motion stability, Ray Reconstruction, Frame Generation cost or latency, and the engine interface. NVIDIA hasn’t specified which changes will apply to this card.
Should a developer schedule DLSS 5 now?
Schedule support for the version that ships. Keep the current integration modular, store a measured 4.5 baseline, and test any new SDK only after it officially exposes the path on this GPU.
How should an owner use the baseline?
The listed QHD results span 82.5 to 103.6 FPS. Use that measured range as a reference for present workloads, then compare it with your own games and settings. Don’t attach an unannounced DLSS 5 gain to the keep-or-replace decision.
What would change the status?
An NVIDIA DLSS 5 specification and supported GPU list that names the desktop GeForce RTX 4070 Ti SUPER would change it. The team could then re-run the same games and replace the conditional discussion with measurements tied to the new SDK.
Leave a Reply