NBA 2K27’s official DLSS 5 implementation can make players look strikingly more realistic. Independent testing also exposes a heavy neural-rendering cost before Multi Frame Generation turns the final counter into a much bigger number.
DLSS 5 is no longer a leaked experiment being forced into games that were never designed for it.
Nvidia released the technology officially in NBA 2K27 for GeForce RTX 50-series PCs and laptops, as well as GeForce NOW. The first independent testing of that supported implementation answers one of the questions left open by the leak: can developers make it look intentional?
In this first game, the answer is often yes. The next answer is less comfortable. DLSS 5’s neural-rendering pass can consume more than half of the underlying frame rate before Multi Frame Generation starts building the much larger number shown on screen.
That makes the launch a genuine trade-off, not another automatic DLSS upgrade. The image can improve. The rendering bill can be brutal.
The finished implementation looks directed
Quest Novum’s earlier report on the leaked experiments argued that neural rendering would need an art director. Modders had exposed controls and pushed the pre-release component into unsupported games, producing striking but sometimes uncanny changes to faces, materials and lighting. Those experiments showed what could go wrong. They could not show what a finished integration would look like.
NBA 2K27 is the first proper answer. Nvidia says DLSS 5 uses the game engine’s rendered frame as its foundation, then lets developers choose models and tune Structure Intensity, Tone Intensity and masking. Artists can decide how strongly the model changes high-frequency material detail, how it adjusts wider lighting and colour, and which characters or objects it should leave alone.
Players do not receive that unrestricted control. In NBA 2K27, the public setting is an on-or-off toggle. The granular choices sit with the developer, although a studio could choose to expose selected controls in its own menu.
That distinction appears to matter. PC Gamer’s launch-day testing found the official treatment much more tasteful than the leaked demonstrations. Close-ups showed more convincing skin, hair, facial shadowing and fabric detail. Some faces still looked over-contrasted, and the improvement was easiest to see in photo mode, replays and other close shots rather than during the usual zoomed-out game.
This is not proof that DLSS 5 will suit every art style. It is evidence that a restrained, developer-tuned implementation can avoid the worst excesses of an unsupported global filter.
The big FPS numbers are not the whole story
Nvidia’s headline result says an RTX 5090 can reach up to 370 frames per second at 4K, Ultra settings and ray tracing with DLSS 5. That figure uses the full DLSS stack, including Performance-mode Super Resolution and 6X Multi Frame Generation.
It describes the final stream of displayed frames with every acceleration tool enabled. It does not isolate how quickly NBA 2K27 renders after the DLSS 5 neural pass has been added.
TechSpot removed Multi Frame Generation from the comparison and tested the official implementation across the RTX 50-series range with Nvidia’s supported driver. At 4K, it measured the RTX 5090 falling from 192fps without DLSS 5 to 76fps with it, a 60% reduction. Its RTX 5080 result fell from 185fps to 53fps, a 71% reduction.
PC Gamer’s separate RTX 5080 test produced an even harsher result in one specific configuration. At native 4K with the Ultra preset and ray tracing, average performance dropped from 183fps to 49fps when DLSS 5 was enabled. That is the widely reported 73% reduction.
It is not a universal DLSS 5 figure. PC Gamer recorded live quick-match play without a repeatable built-in benchmark, and it described the result as representative of the average hit in its run rather than definitive game performance. At 1440p on the same system, the reduction was smaller but still severe: 230fps became 93fps.
These are TechSpot and PC Gamer measurements, not Quest Novum benchmarks. They cover one launch game, current drivers and the first public model. They do establish that the cost can be substantial even when the official integration is doing exactly what its developer intended.
Upscaling barely touches the expensive part
The awkward part is where DLSS 5 sits in the pipeline. TechSpot found that the official neural-rendering pass works after Super Resolution and is tied mainly to the final output resolution. Lowering the internal render resolution makes the rest of the game cheaper, but the DLSS 5 pass still has to process the same 1440p or 4K output.
On an RTX 5060 Ti 16GB at 1440p, TechSpot measured 129fps with DLAA and no DLSS 5. Enabling neural rendering cut that to 46fps. Switching from DLAA to Quality upscaling only raised the DLSS 5 result to 50fps.
Its estimated processing times explain the ceiling. At 4K, the neural pass alone added about 7.9 milliseconds on an RTX 5090, 13.7ms on an RTX 5080 and 17ms on an RTX 5070 Ti. Once the effect itself takes longer than the 16.7ms budget for a 60fps frame, ordinary quality settings cannot solve the problem on their own.
Frame Generation can still make motion look smoother and raise the displayed counter. PC Gamer reached 92fps at 4K with Quality upscaling, 2X Frame Generation and DLSS 5 on its RTX 5080. The same settings without DLSS 5 produced 255fps.
Both numbers are useful. They answer different questions. One describes the frames sent to the display. The other comparison exposes the rendering capacity surrendered to generate the improved image.
RTX 40 support is planned, not live
DLSS 5 is officially available today on RTX 50-series hardware and GeForce NOW. Nvidia has also said that it plans to work on RTX 40-series support after RTX 50 performance is more fully tuned, as reported to The Verge.
There is no RTX 40 release date, and official support is not live on those cards today. Nvidia expects model updates later in the autumn to improve the current RTX 50 baseline first. RTX 30-series and RTX 20-series support has not been announced.
That leaves the launch version in an unusual position. It is a visual-quality feature carrying the DLSS name, but it behaves more like an expensive new rendering effect than a performance boost. Owners with abundant headroom can trade frames for better-looking close-ups, materials and lighting. Everyone else has a perfectly sensible option: leave it off.
The image is better. The bill is real.
The leaked experiments did not tell the whole story. They made DLSS 5 look uncontrolled because they removed the people who were supposed to control it. NBA 2K27 shows that artists can use the same idea with far more restraint.
The official release also closes the performance question those experiments could not answer. Even in a properly integrated game, the current neural model can be punishing before generated frames are added on top.
DLSS 5 at launch is not free performance. It is an optional image-quality effect whose cost must be paid in the underlying render, then partly offset in the displayed output with upscaling and generated frames.
The better image is real. So are the frames surrendered to produce it.
