GeForce RTX 4070 SUPER DLSS 4.5 vs DLSS 5: baseline and outlook

GeForce RTX 4070 SUPER DLSS 4.5 vs DLSS 5: baseline and outlook

DLSS 4.5 is the production baseline for a GeForce RTX 4070 SUPER. The card supports Super Resolution, Ray Reconstruction, and Frame Generation today. NVIDIA has not announced DLSS 5 launch support for any RTX 40 GPU, so future support for the 4070 SUPER remains unknown. A studio shipping on a fixed date should keep any DLSS 5 work out of certification, marketing, and milestone performance targets until NVIDIA publishes a supported-hardware list.

That still leaves plenty to plan around. The desktop card has a useful 1440p performance envelope, a 12 GB frame buffer, and a mature Ada Lovelace feature set. Measured native results establish the baseline; DLSS 4.5 then gives the team several independent ways to trade image quality, render cost, smoothness, and latency.

The hardware budget behind the preset

The desktop GeForce RTX 4070 SUPER is a mid-range Ada Lovelace card aimed at 1440p gaming with hardware ray tracing and AI upscaling. Its 12 GB of GDDR6X is enough for ultra texture settings at 1440p in most current games. Headroom tightens when path tracing, dense volumetric fog, and high-resolution shadow maps run together, especially if a game tries to keep 4K texture sets partly resident at 1440p.

Super Resolution helps because it lowers the resolution of the expensive base render before reconstructing the final 1440p image. It does not solve every memory problem, but it can reduce the pixel-scaled work that competes with those assets.

The card also exposes the same NVENC encoder, CUDA compute support, and DirectX 12 Ultimate feature set as other RTX 40 desktop models. DirectStorage can change asset-streaming cost, while mesh shaders change geometry batching. Neither belongs to DLSS, but both affect the budget available to the upscaler and ray-tracing passes.

DLSS 4.5 is three separate tools

A game does not simply switch “DLSS” on. The engine controls three features independently, and QA should test them that way:

• DLSS Super Resolution renders at a lower internal resolution and reconstructs the output with spatial and temporal data. It reduces fragment-shader cost and other work that scales with pixel count.
• DLSS Ray Reconstruction replaces hand-tuned spatio-temporal denoisers for ray-traced reflections, global illumination, ambient occlusion, and shadows. One model can handle the full chain.
• DLSS Frame Generation uses optical flow to insert one interpolated frame between two rendered frames. It raises the displayed frame count without reducing the cost of the rendered frames.

Those combinations behave differently. Super Resolution at a lower base resolution is a real rendering-budget change. Frame Generation at native 1440p is primarily a presentation and latency trade. A producer should record the enabled features, preset, and internal resolution rather than treating DLSS as one binary setting.

DLSS 5 remains an unsupported planning assumption

No official DLSS 5 support has been announced for the GeForce RTX 4070 SUPER, and NVIDIA has not confirmed whether or when RTX 40 cards will receive it. The supported statement is narrower: this card runs DLSS 4.5 Super Resolution and Ray Reconstruction, plus Frame Generation.

That distinction protects the schedule. If a vertical slice depends on an unannounced feature, it can fail on the first QA machine that lacks the expected path. The reverse assumption is risky too: a planned fallback cannot depend on DLSS 5 arriving later. The GeForce 40 series family page identifies the 4070 SUPER as an Ada Lovelace part and positions the family above the RTX 30 generation for AI workloads, but it is not a DLSS 5 support promise.

Measured native performance at 1440p

These are direct average-FPS measurements for the desktop GeForce RTX 4070 SUPER at 2560×1440. They come from the Notebookcheck page for this exact GPU. None is interpolated from an RTX 4070 or RTX 4070 Ti, and none is a DLSS-enabled result.

The two Cyberpunk 2077 readings cover different game states around the 1.6 path-tracing update: 91.3 FPS is the post-patch reading and 83.8 FPS is the pre-patch reading. Alan Wake 2 at 73.8 FPS is the heavier result in the set and works as a conservative floor for a current 1440p path-traced build. CPU, driver, patch, and test-scene differences still matter, so these figures are anchors rather than universal guarantees.

For a vertical slice, 91.3 FPS is a realistic upper reference for this measured workload and 73.8 FPS is the safer target. A materially higher 1440p goal needs a lower internal resolution, a lighter preset, or a faster minimum-spec GPU.

How the three features change that baseline

Frame Generation can make a HUD counter report roughly twice the rendered rate by inserting an interpolated frame. The rendered workload remains 91.3 FPS in the Cyberpunk example, and the synthesis stage can add latency or show artifacts during fast motion. I would not enable it by default in a competitive game without an end-to-end latency test.

Super Resolution changes the base workload. Rendering at 1280×720 and reconstructing 2560×1440 cuts the fragment count by roughly a factor of four. Preset labels can vary with the selected configuration, so verify the current SDK mapping rather than copy a label blindly. The pixel-count saving remains the same. Fine hair, skin pores, and other stable detail usually show the trade most clearly, while racing and sports games often tolerate a more aggressive setting.

Ray Reconstruction is more useful to the renderer than it is obvious in a menu. It can replace separate reflection, GI, AO, and shadow denoisers, reduce custom shader maintenance, and improve temporal stability at low sample counts. A team with a mature denoiser stack has more migration work than a studio building its first ray-traced title.

Build around 4.5 and leave a clean hook

The release should depend on DLSS 4.5 only. A feature flag keeps the renderer ready for a later SDK without turning an unannounced feature into a delivery risk.

A feature flag and a small interface stub cost little compared with a delayed certification build. Do not add a call to an entry point that does not yet exist; keep the abstraction ready and wait for a documented SDK contract.

If DLSS 5 reaches the card

The Notebookcheck benchmarks and specs remain the correct baseline for this SKU. The possibilities below are conditional, not announced features:

• Base frame rate could rise at the same render resolution if a new spatial-temporal reconstruction model lowers per-pixel cost or remains stable with fewer resolve passes.
• A revised frame-generation mode could reduce latency or improve motion handling through updated optical flow or transformer-based synthesis. That would not raise the rendered frame rate by itself.
• Ray Reconstruction could extend to effects such as caustics, sub-surface scattering, or participating media. The renderer would still pay for the underlying samples.

None of those possibilities belongs in a roadmap until NVIDIA confirms the feature, the card, the driver, and the SDK. Even then, the studio must repeat its tests on the exact minimum-spec hardware.

Confirmed features versus the open question

Filter public claims before they reach production

Give public information different weight according to its source:

• Press coverage that paraphrases community expectations can show what people are discussing, but it rarely provides a supported-hardware list or a publisher-backed date.
• Leaked benchmark screenshots do not establish the GPU, driver, scene, or test method. Keep them out of performance targets.
• NVIDIA blog posts, SDK notes, and driver release notes can justify a production change because they name supported hardware and software versions.

One team member should track the NVIDIA developer blog and GeForce driver notes, logging each SDK version and its hardware list. Until that list changes, the build, marketing copy, and specification should stay on DLSS 4.5.

Choose the internal resolution from the worst scene

The 4070 SUPER is primarily a 1440p card, with some 4K headroom when DLSS is active. Internal resolution drives fragment cost, so set it from the heaviest scene rather than tuning around an easy benchmark.

• 1280×720 internal, upscaled to 2560×1440 with Super Resolution quality preset. Lowest fragment cost among the comfortable options, comfortable headroom for path tracing, mild image-quality penalty in fine detail such as hair and skin pores.
• 960×540 internal, upscaled to 2560×1440 with Super Resolution performance preset. Half the fragment cost again, larger image-quality penalty, useful for VR or high-refresh targets where the priority is motion smoothness over still detail.
• Native 2560×1440 with no upscaler. The reference image-quality path, the most expensive, and the one that matches the 91.3 FPS Cyberpunk 2077 1.6 measurement above. Useful as a benchmark, rarely useful as a shipping default.
• Native 3840×2160 with no upscaler. Outside the card’s comfortable 1440p envelope, only viable in light workloads, and not a sensible production target for a path-traced title at the 4070 SUPER’s price tier.

Pick the base resolution that meets the frame-rate target in the worst scene, then use the Super Resolution preset to balance the rest of the game. Layer Frame Generation on afterward; it changes the displayed rate, not the base render budget.

Sources and measurement limits

The GeForce 40 series reference supplies the family and Ada Lovelace context. The Notebookcheck page supplies the three 1440p average-FPS values and its test-platform details. Those are the only external sources used here. Compare internal results with the same CPU, driver, patch, and preset context whenever possible.

Frequently asked questions

Does the GeForce RTX 4070 SUPER support DLSS 5?

No official support has been announced, and no public NVIDIA hardware list includes the card. It supports DLSS 4.5 Super Resolution, Ray Reconstruction, and Frame Generation today.

What is the difference between DLSS 4.5 and DLSS 5?

DLSS 4.5 is a shipped bundle on this GPU. DLSS 5 is a successor name without a published RTX 4070 SUPER feature list, driver list, SDK list, or release commitment.

Is DLSS 4.5 the same as DLSS 4?

DLSS 4.5 is NVIDIA’s current shipping version of the DLSS bundle on RTX 40 hardware. Driver and SDK releases can alter the available capabilities and presets, so check the notes for the version included in the build.

What base render resolution should I target?

For 1440p output, the supplied production profile uses 1280×720 internal at the Quality preset, a fourfold reduction in fragment count from native 1440p. A 1080p output can stay closer to native. A 4K output needs a much lower internal resolution and a more aggressive preset.

Does Frame Generation add latency?

It inserts an interpolated frame and improves apparent smoothness, but it does not lower the input lag of the rendered frames. Measure end-to-end latency before enabling it by default, especially in competitive multiplayer.

Should I plan a future game around DLSS 5 on this card?

No. Ship against DLSS 4.5, leave a clean feature flag, and revisit the plan only if NVIDIA publishes support for the RTX 4070 SUPER.

Where do the FPS numbers come from?

Notebookcheck measured 91.3 FPS in Cyberpunk 2077 1.6, 83.8 FPS in Cyberpunk 2077, and 73.8 FPS in Alan Wake 2 on the desktop card at 2560×1440. These are native baseline measurements, not DLSS-enabled results.

Can a 1080p monitor benefit from DLSS 4.5?

Yes, particularly with path tracing or other expensive effects. The card is usually comfortable at native 1080p, so the right preset depends on the game rather than an engine default.

What should a producer ask before locking the specification?

Ask which of the three DLSS features are integrated, what preset and base resolution the performance profile uses, and which scenes, motion patterns, minimum rendered rate, and latency limits make up the Frame Generation QA matrix.

What is the safest assumption for a near-term PC release?

Assume DLSS 4.5 support and no DLSS 5 support. If a later driver adds the card, treat it as a post-launch feature with a new QA pass and specification line.