GeForce RTX 5090 D DLSS 5 benchmark

GeForce RTX 5090 D DLSS 5 benchmark: what verified launch data actually shows

Only three verified average-FPS results are available for this exact desktop model, and all three come from IT Home’s native 4K testing without DLSS. The GeForce RTX 5090 D enters its public review window on September 3, 2026, alongside the scheduled launch of DLSS 5 for the RTX 50 Series. Until a primary source publishes more data, every other game, resolution, and upscaling mode remains unverified for this card.

That distinction matters on launch day. The confirmed numbers establish a native 4K baseline; they do not predict DLSS 5 performance or settle how the card will behave after driver, game, and plug-in updates.

RTX 5090 D launch context

The GeForce RTX 50 Series is NVIDIA’s Blackwell-generation lineup. The desktop RTX 5090 D is the China-region version of the RTX 5090, with adjusted power and compute profiles for local certification rules. The RTX 50 Series Wikipedia page describes the family, including shared memory, display, and encoding features, but it does not provide benchmark figures for this card.

Direct measurement remains limited while DLSS 5 waits for its scheduled RTX 50 Series release. The first independent results, including IT Home’s test of the Galax GeForce RTX 5090 D General, concentrate on native 4K rendering. The IT Home review of the Galax GeForce RTX 5090 D is the primary source for the numbers below. It also describes the DLSS 4 Multi Frame Generation behavior already available on RTX 50 Series hardware. Secondary reports should be checked against that source.

Expect the first few weeks to bring driver revisions, game patches, and DLSS 5 plug-in updates. A development team should record the cited driver and game build, use one primary review for its initial baseline, and schedule a later update instead of treating launch figures as settled.

Verified native 4K averages

IT Home provides all three verified averages currently available for the desktop GeForce RTX 5090 D. The table preserves the source settings and resolution without interpolation, rounding, or community results.

Game	Resolution	Settings	Average FPS	Source
Counter-Strike 2	3840×2160	4K, High quality, FPS Benchmark workshop map	426	IT Home direct measurement
Naraka: Bladepoint	3840×2160	4K, Highest quality, native/default AA path	242	IT Home direct measurement
Shadow of the Tomb Raider	3840×2160	4K, Highest quality preset	244	IT Home direct measurement

Don’t average these rows. Counter-Strike 2 is a fast competitive game with a relatively simple render path, Naraka: Bladepoint combines crowd effects with large arenas, and Shadow of the Tomb Raider uses a single-player action-adventure workload with high-quality lighting and ambient occlusion. All three are native 4K results, not DLSS 5 tests, so each one is a labeled baseline rather than a general claim about the card.

A hypothetical DLSS 5 multiplier such as “roughly X times” or “around Y percent” would be unsupported. The result depends on the selected mode, game integration, frame-generation cap, and DLSS plug-in version. Leave that multiplier unverified until a primary source publishes a paired native and DLSS 5 comparison on this exact card.

How DLSS 5 changes the render path

DLSS 5 is the next version of NVIDIA’s real-time upscaling and frame-generation pipeline. In a typical integration, the engine renders at a lower internal resolution, an upscaling model reconstructs the output target, and frame generation inserts frames between reconstructed outputs. The player sees a higher presented frame rate while the GPU works on a smaller base frame. Ray Reconstruction and Deep Learning Anti-Aliasing are also part of the DLSS framework. Available features depend on the game, engine integration, and DLSS SDK version.

Launch-day questions remain title-specific: which games support DLSS 5, which engines include the matching plug-in, and whether the chosen path is Quality, Balanced, Performance, or Ultra Performance. The RTX 50 Series Wikipedia page places DLSS 5 within the hardware family, while IT Home documents the card’s existing DLSS 4 Multi Frame Generation behavior. Use the native IT Home figures as a floor and run a paired comparison after the target game’s DLSS 5 plug-in is available.

Support also requires the game, engine hooks, and driver to align. Until they do, a DLSS 5 test for that title is a plan rather than a measurement. Check each item before recording a result:

Confirm that the game has an official DLSS 5 plug-in build for the engine version in use.
Confirm that the DLSS SDK shipped in the engine matches the version that the card’s driver supports.
Confirm that DLSS 5 frame generation is exposed at the desired path, including quality, balanced, performance, or ultra performance.
Record the driver build, the game build, and the engine version in any benchmark log so that the result is reproducible.
Compare the native and DLSS 5 numbers side by side at the same render scale, rather than treating one as a substitute for the other.

Map each result to its workload

Each verified 4K result can inform a matching development workload, provided the team also defines its target hardware and acceptable frame time.

Counter-Strike 2 averaged 426 FPS at native 4K High on the FPS Benchmark workshop map. That sits above the 360 Hz ceiling of most current esports displays and well beyond the 240 Hz displays common on development benches. In this workload, input latency and display refresh are more likely limits than GPU throughput. The result says nothing about the 1080p path commonly used for professional Counter-Strike 2 testing.

Naraka: Bladepoint averaged 242 FPS at native 4K Highest, indicating enough performance for a 240 Hz display in the tested dense crowd scene. Against a 4K 60 FPS target, that is roughly four times the required frame rate. A different crowd density, particle budget, or post-processing chain will change the result, so 242 FPS remains specific to IT Home’s conditions.

Shadow of the Tomb Raider averaged 244 FPS at native 4K Highest. Its baked and screen-space lighting workload runs well above 60 FPS and provides a reference for engines with a similar lighting model. It does not predict performance in a modern path-traced or virtual-geometry-heavy project.

Assign each result to a workload class, record the test conditions, and compare only with project scenes that genuinely resemble it. Anything else needs a separate test.

Choose native 4K, DLSS 5, or a hybrid path

A PC target matrix should pair each resolution and refresh rate with a rendering path. A single-player game targeting 4K at 60 Hz can usually remain native for most scenes and reserve DLSS 5 for the heaviest work. At 4K and 120 Hz or higher, plan for the engine to switch between native rendering and upscaling according to per-frame cost.

The matrix uses verified native 4K results as a floor and leaves unknown DLSS 5 multipliers unresolved.

Target frame rate at 4K	Workload class	Native path viability on the GeForce RTX 5090 D	DLSS 5 path viability on the GeForce RTX 5090 D
60 Hz, single-player action-adventure	Shadow of the Tomb Raider class	Comfortable headroom at 244 average native	Reserved for the heaviest scenes, with paired comparison required
120 Hz, mixed-mode arena	Naraka: Bladepoint class	Workable at 242 average native in a dense crowd	Likely required for scenes above the 242 average, with mode choice unresolved
240 Hz, competitive PC	Counter-Strike 2 class	Headroom well above 240 in the FPS Benchmark workshop map	Reserved for input-latency experiments and not the default path
360 Hz and above, competitive PC	Counter-Strike 2 class on smaller resolution paths	Not represented in the verified data set	Not represented in the verified data set

The 360 Hz row falls outside the verified set. The 60 Hz and 120 Hz DLSS 5 entries also require paired testing. If a project needs a 360 Hz competitive path, test it separately instead of inferring a multiplier.

Choose the path per project and per scene, not per card. The same RTX 5090 D may run a single-player game at native 4K and 60 Hz with DLSS 5 disabled, then use a different 1080p path for a competitive game. A demanding cutscene may also switch one project from native rendering to DLSS 5 for a single frame. Expose that decision in the profiler instead of hiding it behind a global flag.

Benchmark checklist for production work

A code review or producer’s report needs more than one FPS average. Record enough information for another developer or QA lead to repeat the test:

Record the GPU SKU, the card vendor, the cooler variant, the power supply, and the case airflow so the test is reproducible on another bench.
Record the driver build, the NVIDIA app or control panel settings, and any vendor tuning software used in the test.
Record the game build, the engine version, the DLSS SDK version, and the DLSS 5 mode used in the test.
Record the resolution, the render scale, the dynamic resolution settings, and the frame-generation cap.
Record the scene or benchmark path, the camera path, the time-of-day setting, and the loading state of the scene at the start of the capture.
Record the average FPS, the 1 percent low, the 0.1 percent low, and the frame time variance rather than the average alone.
Record the system-level metrics: CPU model, memory speed, storage drive, and any background process that can perturb the result.
Capture a paired native and DLSS 5 run in the same scene with the same camera path so the multiplier is honest.

Always include a frame-time graph. Average FPS hides the spikes players notice, so the production report should show the average, low percentiles, and frame times against the target refresh rate.

Common launch-review mistakes

Early coverage varies with test benches, builds, and reporting methods. Watch for these mistakes before using a review in a target matrix:

Treating a single reviewer’s average FPS as a promise. A single review is one data point, and the result depends on the test bench, the driver build, and the game build.
Mixing native and DLSS results in the same chart. The two paths have different cost profiles, and combining them hides the cost of the DLSS path itself.
Ignoring the 1 percent low and the 0.1 percent low. The average can look comfortable while the worst frames are poor, and a player feels the worst frames.
Reading a 1080p result into a 4K target matrix. The two paths are not interchangeable, and the GPU cost scales with render target size.
Treating a synthetic benchmark as a game benchmark. Synthetic numbers do not capture the engine cost of a specific title, and they should be labelled as synthetic in any production report.
Quoting a DLSS 5 multiplier before DLSS 5 is officially available. The multiplier depends on the DLSS 5 mode, the game integration, and the version of the plug-in, and it is not a fixed number across the platform.

Keep the first review as a dated snapshot. Replace it with a paired comparison once the target title has a stable DLSS 5 build.

Turn the verified results into a project plan

A producer or technical lead needs the confirmed figures, the project decision, and the outstanding tests in one place. This format separates verified data from pending work without inventing a DLSS 5 multiplier.

Section	Verified content	Pending content
Native 4K average FPS for the GeForce RTX 5090 D	Counter-Strike 2 at 426, Naraka: Bladepoint at 242, Shadow of the Tomb Raider at 244 (all from IT Home)	Other titles, other settings, other resolutions
DLSS 5 paired comparison on the GeForce RTX 5090 D	None published in the verified set	Paired native and DLSS 5 run for each target title
Driver and game build	As cited by IT Home in the original review	Driver and game build at the team’s own test bench
Workload mapping	Workload class for each of the three verified games	Mapping of project scenes to the verified workload classes

Give each pending row an owner, an exit criterion, and a test plan. The exit criterion is a published paired comparison for the target title, using the benchmark checklist above. With three verified results and a defined route to more data, the team doesn’t need to chase every secondary launch review.

When to revisit the baseline

Don’t lock the baseline during the launch window. Revisit it after the driver, game build, and DLSS plug-in begin to stabilize:

Two weeks after the public DLSS 5 release time, revisit the driver notes and the game patch notes for the target title.
One month after the release time, capture a paired native and DLSS 5 run in the same scene with the same camera path.
Three months after the release time, record a new average, new low percentiles, and a new frame-time chart, and replace the launch snapshot with the updated set.
After any major engine upgrade, treat the new engine build as a separate baseline and run a fresh comparison rather than editing the previous one.

A major patch may justify earlier testing, while a stable single-player game can wait longer. The schedule should keep the baseline current without forcing a retest for every driver revision.

What remains unverified

Plan a separate test for every value missing from the three-result dataset:

No DLSS 5 multiplier for the GeForce RTX 5090 D in any of the three verified games.
No 1080p or 1440p native FPS for the GeForce RTX 5090 D in the verified set.
No ray-traced FPS for the GeForce RTX 5090 D in the verified set, including path-traced or hybrid-traced workloads.
No power, thermal, or acoustic measurement for the GeForce RTX 5090 D in the verified set.
No frame-time chart, 1 percent low, or 0.1 percent low for the GeForce RTX 5090 D in the verified set.
No comparison with neighbouring RTX 50 Series cards, neighbouring generations, or competing platforms in the verified set.

Missing data defines the next test. A team that needs a ray-traced baseline should run paired native and ray-traced passes, record the relevant DLSS 5 mode, and store the result as a new data point rather than extending the launch set.

Map the benchmark to a real project

A small project can use the verified set as a conservative floor. Shadow of the Tomb Raider can inform native 4K at 60 Hz for a single-player action-adventure game, Naraka: Bladepoint can inform native 4K at 120 Hz for arena multiplayer, and Counter-Strike 2 can inform native 4K above 240 Hz for competitive PC work. DLSS 5 adds potential headroom; it doesn’t replace those native baselines.

Larger projects should map their scenes to the closest verified workload. A Tomb Raider-like scene may use the 244 FPS average for early planning, while a Naraka-like scene may use 242 FPS. Anything unlike those workloads belongs in a separate test. Make the mapping explicit so the matrix doesn’t rely on one average that describes none of the project’s scenes.

The source data covers a desktop part only. Supporting a laptop, mobile, Max-Q, D, SUPER, Ti, or neighboring model requires a separate test because the desktop figures don’t transfer to smaller or differently configured SKUs.

Frequently asked questions

What verified average FPS does the GeForce RTX 5090 D deliver at 4K without DLSS?

The verified set, drawn from a direct measurement published by IT Home, records 426 average FPS in Counter-Strike 2 at 4K High quality on the FPS Benchmark workshop map, 242 average FPS in Naraka: Bladepoint at 4K Highest quality on the native or default AA path, and 244 average FPS in Shadow of the Tomb Raider at 4K Highest quality. All three rows are native 4K numbers and do not include any DLSS mode.

Does the GeForce RTX 5090 D support DLSS 5 at launch?

DLSS 5 is scheduled to launch as an RTX 50 Series feature, and the desktop GeForce RTX 5090 D is part of the RTX 50 Series. Game-level support depends on each title shipping a DLSS 5 plug-in, the engine version exposing the right hooks, and the driver build recognising the feature. A definitive per-title answer requires checking the game’s official DLSS support list rather than assuming a default.

Where can the verified GeForce RTX 5090 D DLSS 5 benchmark numbers be checked?

The primary source for the three verified numbers is the direct measurement published by IT Home for the Galax GeForce RTX 5090 D General card. The review is the reference for the test conditions, driver build, and game versions used. Trace any secondary restatement back to that source before using it in a project plan.

Why is there no DLSS 5 multiplier in the verified data set?

The verified data set records only native 4K average FPS. A DLSS 5 multiplier depends on the upscaling mode, the frame-generation cap, the game integration, and the DLSS SDK version, and the multiplier is not a fixed number across the platform. The conservative reading is to leave the multiplier unverified until a primary source publishes a paired native-versus-DLSS 5 comparison for this exact card in a specific title.

Can the 4K native numbers be used to plan a 1080p or 1440p target matrix?

No. The verified numbers are 4K results, and GPU cost scales with render target size. A 1080p or 1440p result will be higher in a given scene, but the size of the gain depends on the engine, the scene, and the resolution scale. A project that targets 1080p or 1440p should plan a separate test rather than read the 4K result into a smaller resolution.

Is the GeForce RTX 5090 D the same card as the standard RTX 5090?

No. The GeForce RTX 5090 D is the China-region variant of the RTX 5090 with adjusted power and compute profiles suitable for local certification rules. The two SKUs share the RTX 50 Series platform and the broader feature set, but the exact power, compute, and certification details differ, and a benchmark for one SKU is not automatically a benchmark for the other.

How should a development team treat the launch-day numbers?

Treat launch-day numbers as a snapshot rather than a promise. Record the test conditions, label the workload class, and revisit the result after the driver, game build, and DLSS plug-in stabilize. Any project plan using the launch snapshot should mark pending rows and define when they will be replaced.

What other data points should a team capture alongside the GeForce RTX 5090 D DLSS 5 benchmark?

A complete capture includes the driver build, the game build, the engine version, the DLSS SDK version, the resolution, the render scale, the scene or camera path, the average FPS, the 1 percent low, the 0.1 percent low, the frame-time variance, and the system-level metrics such as CPU, memory, and storage. A paired native and DLSS 5 run in the same scene with the same camera path is the most useful capture for a project plan.

Are the verified numbers representative of every game?

No. The verified numbers cover three specific games, each with a specific render path and a specific test scene. A project with a different engine, a different lighting model, a different crowd density, or a different post-processing chain will not produce the same average, and the verified rows are workload labels rather than general claims about the card.

What is the next test a team should plan after reading the verified data?

The next test is a paired native and DLSS 5 run in the project’s own target title, with the project’s own driver build, game build, and DLSS SDK version. The test should record the average, the low percentiles, and a frame-time chart, and the result should replace the launch snapshot as the new baseline for that project.

GeForce RTX 5090 D DLSS 5 benchmark