GeForce RTX 5090 D v2 DLSS 5 benchmark: what game developers actually know at launch
Only two average-FPS figures have been measured and reproduced for this exact desktop model: 418 FPS at 1440p and 289 FPS at 4K in Cyberpunk 2077 using DLSS 4. DLSS 5 is scheduled for the wider RTX 50 Series on 2026-09-03 at 21:00 PT, and figures for the new stack remain launch-gated until then.
Developers should keep those categories separate. The published DLSS 4 results are usable; any DLSS 5 frame rate for the GeForce RTX 5090 D v2 remains unverified until the release window closes and a primary source measures it.
Why the RTX 5090 D v2 needs separate testing
The GeForce RTX 5090 D v2 is a market-specific model with a different memory subsystem from the standard RTX 5090 D. Its narrower bus and lower capacity can change memory-bound scenes, high-resolution texture streaming, and large framebuffers even when compute performance looks similar. Use card-specific traces rather than borrowing results from the wider RTX 5090 family.
A PC target matrix should give the RTX 5090 D v2 its own row for four reasons:
- It is a desktop GPU, not a laptop, Mobile, Max-Q, SUPER, or Ti variant. Any performance claim must be tied to this exact model and form factor, and any number borrowed from a neighboring variant should be discarded.
- It ships into a regional channel, so a meaningful slice of end users will pair it with CPU and driver combinations that differ from a North American review rig. That pairing shapes the frame-time curve, not just the headline average.
- It enters the market in the same window as DLSS 5, so the launch decision has to answer two questions at once: how the card performs under today’s upscaler, and how it will behave under the next-generation stack once that stack becomes officially available.
- It carries a narrower memory bus than the rest of the 5090 family, so any game that depends on bandwidth rather than raw compute will sit on a different point of the cost curve.
Its place in the RTX 50 Series
NVIDIA positions the RTX 50 Series around a new generation of ray tracing, neural rendering, and frame generation. The GeForce RTX 5090 D v2 is a regional desktop part within that family, not a flagship replacement or a refresh of an existing SKU. It shares drivers, the DLSS runtime, and the shader compiler with the rest of the lineup.
Place it in the QA matrix as a regional desktop model and note that its memory bus differs from the rest of the RTX 5090 family. The wider product lineup establishes canonical model names, approximate positioning, and release order.
Most PC games won’t use the RTX 5090 D v2 as their main optimization target. It is better suited to upper-bound testing in vertical slices and benchmark reels, just as the original RTX 5090 D has been used.
What is official for DLSS 5 at launch
DLSS 5 is scheduled to launch with the RTX 50 Series refresh on 2026-09-03 at 21:00 PT. The GeForce RTX 5090 D v2 is officially supported as an RTX 50 Series product, but its DLSS 5 performance data is gated until that release time. Do not quote a model-specific DLSS 5 frame rate yet.
Keep these restrictions in internal documentation and player-facing copy until launch:
- Do not state that the GeForce RTX 5090 D v2 already runs DLSS 5 in a shipping build. The card is supported, the launch data is not, and those are not the same sentence.
- Do not publish DLSS 5 frame-rate numbers for the 5090 D v2 from pre-release drivers, partner builds, or leaked tables. Any such number will not survive the driver branch that lands with the public release.
- Do not carry DLSS 4 figures forward as DLSS 5 figures, even when the workload, scene, and resolution look identical. The two are different stacks, and the cost-quality trade moves between them.
- Do not extend a number measured on a different RTX 50 Series SKU onto the 5090 D v2. The memory subsystem difference makes that substitution unsafe in memory-heavy scenes.
The verified Cyberpunk 2077 results
Tom’s Hardware reproduced two Yesky measurements from a Cyberpunk 2077 DLSS 4 benchmark on this exact model. They describe the current upscaler stack, not DLSS 5.
Cyberpunk 2077 averaged 418 FPS at 2560×1440 and 289 FPS at 3840×2160. Those are the only numeric FPS claims currently suitable for public copy or a QA report. Extending them to DLSS 5, another resolution, a ray-tracing preset, or a different scene would turn a measurement into speculation.
| Game | Resolution | Settings | Metric | Result |
|---|---|---|---|---|
| Cyberpunk 2077 | 2560×1440 | DLSS 4 benchmark at 1440p | Average FPS | 418 |
| Cyberpunk 2077 | 3840×2160 | DLSS 4 benchmark at 4K | Average FPS | 289 |
Using the two numbers in production planning
Two averages cannot support a complete optimization plan, but they can answer a few early engineering questions. The table pairs each result with a likely use and the validation still required.
Start with the known measurement, state the limited assumption it supports, and list the conditions that would invalidate it. The figures are workload-specific references, not universal promises.
| Benchmark result | Developer question | Likely use case | Validation gap |
|---|---|---|---|
| 418 average FPS at 1440p (DLSS 4) | Is our high-refresh esports or simulation build over-budget or under-budget on this card? | Setting a realistic high-Hz cap for menus, replays, and lightweight gameplay loops. | Different scene, ray-tracing preset, or DLSS mode can move the number significantly. |
| 289 average FPS at 4K (DLSS 4) | Can we target 4K with our heaviest effects on a flagship card? | Calibrating ultra preset, ray-traced reflections, and screen-space effects for 4K shipping modes. | Specific RT-heavy path tracing presets have not been measured on this card. |
| DLSS 4 stack (both rows) | How does our current DLSS integration look before DLSS 5 arrives? | Regression baseline for upscaler behavior across the RTX 50 Series. | DLSS 5 changes the cost-quality curve, so this is a starting point, not an end state. |
What the two results leave unanswered
The 418 and 289 FPS averages cover one game, one DLSS path, and two resolutions on the desktop RTX 5090 D v2. A developer comparing the RTX 50 Series still needs to account for these gaps:
- No native rendering FPS is published. The published results already apply DLSS 4, so the 5090 D v2 cannot be compared to a card measured at native resolution at the same level of confidence.
- No ray-tracing preset is named in the dataset. Cyberpunk 2077 has several RT paths, and the average FPS shifts depending on which one is enabled, so any RT claim is unsupported for this card.
- No frame-time variance, 1% lows, or 99th percentile numbers are provided. Average FPS smooths over stutter that matters in shipped gameplay, and a vertical slice built on average FPS alone will miss the real frame pacing issues.
- No CPU bottleneck is documented. The 5090 D v2 will behave differently when paired with a flagship CPU versus a mid-range one, and the published figures do not state which side of that pairing is being measured.
- No DLSS 5 figures exist. Any sentence that places a DLSS 5 number on this card, at this resolution, in this scene, is unsupported by the current dataset and should be removed from any public-facing document.
The 384-bit memory-bus limit
The GeForce RTX 5090 D v2 uses a 384-bit memory interface, narrower than the rest of the RTX 5090 family. Early tests place it within roughly 2.2 percent of the original RTX 5090 D in some cases, but that finding is workload-specific.
A Cyberpunk 2077 DLSS 4 result at one resolution cannot guarantee the same gap in another game, resolution, or ray-tracing-heavy scene. Memory pressure increases with larger framebuffers, peak texture streaming, and large simultaneous working sets. A title that stays within available bandwidth may show a small difference; one that exceeds it may open a wider gap.
Tom’s Hardware reproduces the per-resolution figures and describes the 2.2 percent result in that limited context. The 418 and 289 FPS figures show that the card performs well in this already-upscaled workload, not every shipping game.
Compare the card within the RTX 50 Series
Use the wider RTX 50 Series lineup to compare canonical model names and approximate positioning. Tom’s Hardware supplies the per-resolution measurements and early context for the memory-bus question.
In a hardware matrix, list the RTX 5090 D v2 as a regional desktop flagship and record the upscaler and preset used for every FPS figure. Then repeat the same scenes on an internal reference build and capture frame-time variance before calling the target validated.
Validate before locking the target specification
These profiling checks apply to any RTX 50 Series desktop card, including the GeForce RTX 5090 D v2:
- Pick two representative scenes, one CPU-light and one CPU-heavy, to expose any pairing bottleneck before the GPU itself becomes the focus of the report.
- Capture 1% low and 99th percentile frame times, not only average FPS, because stutter and hitching survive the average but ruin shipped builds.
- Record VRAM peak working set, not just the total size, to see how much headroom remains for streaming and large texture pools on the 5090 D v2’s narrower bus.
- Lock the driver version used in the test and include it in the report, because RTX 50 Series performance moves between driver branches, and a number without a driver string is not reproducible.
- Re-run the same scenes under the new upscaler when DLSS 5 is officially available, instead of assuming the DLSS 4 path will carry the same numbers forward into the next release.
Use it differently for vertical slices and shipping builds
The RTX 5090 D v2 is unlikely to be the main target for most PC releases. It works better as an upper-bound card for a vertical slice or benchmark reel. The published 418 FPS at 1440p and 289 FPS at 4K show that a Cyberpunk 2077-class DLSS 4 workload leaves room for engine overhead and profiling.
A shipping budget needs more safety margin and usually targets lower-tier hardware. Use the RTX 5090 D v2 to confirm that the engine’s most expensive path works at the high end, not to justify settings the average player will never see.
Test the regional configuration in QA
The RTX 5090 D v2’s market will include driver branches, OEM cards, and CPU pairings uncommon in other regions. A certification pass should include at least one locally sourced card, a driver matching the regional channel, and a scene that stresses compute and memory together.
Publishers should test the upper end of the card’s range, the lower end of its driver branch, and common local CPU pairings. Two published figures cannot replace that pass, though they can set expectations and provide a reference after a driver update.
Track driver branches with every result
RTX 50 Series performance changes between driver branches. Pre-release, public launch, and first-hotfix drivers may all produce different results. The RTX 5090 D v2’s narrower memory bus can amplify that movement because memory-management or texture-scheduling changes may shift its compute and bandwidth balance faster than on the rest of the RTX 5090 family.
Record the driver branch, date, and exact build hash, then repeat the scene on at least one newer branch before publishing a figure. If driver A produces 418 FPS and driver B produces 402 FPS, QA needs both records, while public copy must quote the result tied to the current driver.
Retest after DLSS 5 becomes available
After DLSS 5 releases on 2026-09-03 at 21:00 PT, the supported RTX 5090 D v2 will use the same family integration subject to the minimum driver branch. The upscaler’s cost and quality balance will change, so 418 and 289 FPS under DLSS 4 cannot support a forward-looking DLSS 5 claim. Rerun the same Cyberpunk 2077 scene at both resolutions with the new stack before publishing a DLSS 5 figure.
Until then, keep the DLSS 4 results as the active reference and mark DLSS 5 measurements as pending verification.
The usable launch baseline
The RTX 5090 D v2 is a desktop RTX 50 Series card scheduled to receive DLSS 5 at the family launch on 2026-09-03 at 21:00 PT. Its verified record contains 418 FPS at 1440p and 289 FPS at 4K in a Cyberpunk 2077 DLSS 4 benchmark, plus an early finding that it sits within roughly 2.2 percent of the original RTX 5090 D in some reproduced tests. DLSS 5 frame rates, native performance, ray-tracing preset behavior, and 1% lows remain unverified.
As new drivers and DLSS 5 measurements arrive, keep every result tied to its game, resolution, upscaler path, and this card’s distinct memory profile.
Frequently asked questions
Is the GeForce RTX 5090 D v2 officially supported by DLSS 5?
The GeForce RTX 5090 D v2 is an RTX 50 Series desktop product, and DLSS 5 is scheduled to launch as part of the wider RTX 50 Series refresh on 2026-09-03 21:00 PT. Until that launch window passes, public DLSS 5 frame-rate numbers on this exact card should not be quoted in any player-facing or press-facing document.
What FPS values can be quoted for the 5090 D v2 today?
Two Cyberpunk 2077 numbers are verified and reproducible: 418 average FPS at 2560×1440 under the DLSS 4 benchmark, and 289 average FPS at 3840×2160 under the same benchmark. Both come from a Yesky measurement reproduced in the Tom’s Hardware benchmark table and are tied to this exact desktop SKU.
Why are only DLSS 4 numbers available and not DLSS 5 numbers?
DLSS 5 is scheduled to launch on 2026-09-03 21:00 PT, and the launch data for the GeForce RTX 5090 D v2 is gated until that release time. Specific DLSS 5 frame-rate results on this card are not part of the verified dataset, and any quote of a DLSS 5 number on the 5090 D v2 at this point would be unsupported.
Does the 5090 D v2 perform the same as the original 5090 D?
In the Cyberpunk 2077 DLSS 4 benchmark reproduced in the Tom’s Hardware piece, the 5090 D v2 sits within roughly 2.2 percent of the original 5090 D in some of the listed tests. That delta is workload-specific and does not extend to native rendering, ray-tracing-heavy scenes, or memory-bound workloads outside the tested titles.
Is the 5090 D v2 a desktop or a laptop GPU?
The 5090 D v2 is a desktop card. FPS figures from a laptop, Mobile, Max-Q, SUPER, Ti, or neighboring variant should not be transferred onto it, and the same rule applies in reverse, because the memory subsystem, power budget, and driver branch are not the same across those form factors.
What should a developer do with these two FPS values in practice?
Treat them as anchors for a Cyberpunk 2077-class scene under DLSS 4, not as universal performance promises. Use them to set upper-bound expectations for a vertical slice, and re-validate against the same scenes once DLSS 5 is officially available on 2026-09-03 21:00 PT, with a matching driver branch.
Does the analysis give a recommendation for buyers?
No. This benchmark context is for developers, not buyers. Pricing, regional availability, and a purchase recommendation fall outside the verified dataset and need different sources.
Where can the underlying benchmark table be read in full?
The full per-resolution table appears in the Tom’s Hardware source linked above, with the original measurements credited to Yesky. The Wikipedia RTX 50 Series page covers the wider lineup and approximate position of each model.
Will the 418 and 289 numbers change once DLSS 5 ships?
Most likely yes, because the upscaler stack changes and the cost-quality curve moves. The 418 and 289 average FPS values are tied to the DLSS 4 benchmark path, and they should be re-measured under DLSS 5 before being used as a forward-looking reference for the GeForce RTX 5090 D v2.
What driver branch should be paired with these numbers in a QA report?
Use the exact driver branch and driver date used in the original test, and record both in the report. RTX 50 Series performance moves between driver branches, and a benchmark number without a driver string is not reproducible on the GeForce RTX 5090 D v2 or on any other card in the family.
