A leaked Nvidia DLSS 5 file has spent the first days of September 2026 tearing through PC gaming forums, not because it looks good, but because of how badly it runs. The library surfaced inside an early-access build of NBA 2K27, and within hours modders had pulled it out and forced it into other games entirely. What they found was a neural rendering pass that, in early and clearly unfinished form, can cut frame rates by roughly half. This is not the polished DLSS 5 launch Nvidia has been teasing. It is a raw, unoptimized build that escaped into the wild before Nvidia was ready to talk about it.
What Just Leaked: Nvidia DLSS 5 Surfaces Inside NBA 2K27
The discovery traces back to the PC early-access build of NBA 2K27. Buried inside the game’s files, users found a library that Windows and file-inspection tools identify as belonging to Nvidia’s next-generation upscaling stack, not the DLSS 4 runtime that ships in current RTX 50-series titles. Reports describe the leak as an unreleased or unfinished version of Nvidia’s AI-based upscaling and neural rendering technology, the kind of thing that normally stays locked inside a driver branch until Nvidia is ready to announce it publicly.
That is the core tension in this story. DLSS 5 was not supposed to be public yet. A sports game shipped it by accident, and the PC modding community did what it always does when it finds an unlocked door: it walked through. Within a day, the file had been extracted, renamed into other games’ folders, and tested on hardware Nvidia had not sanctioned for this kind of experiment.
The File at the Center of the Leak: nvngx_dlssnr.dll
The specific file drawing attention is named nvngx_dlssnr.dll. The “NR” suffix points to Neural Rendering, the feature set Nvidia has attached to DLSS 5 in its public roadmap comments. According to reports examining the file, it carries a version stamp of 310.8.0.0 and weighs in at roughly 158 MB, notably heavier than the DLSS libraries currently shipping in RTX 50-series games.
File: nvngx_dlssnr.dll
Identified version: 310.8.0.0
Approximate size: 158 MB
Found in: NBA 2K27 (PC early-access build)
Function: DLSS 5 Neural Rendering runtimeNothing about the leak suggests NBA 2K27 was ever meant to expose DLSS 5 to players. There is no visible neural rendering toggle in the game’s graphics menu, and 2K has not commented on why the file shipped with the build at all. The most likely explanation making the rounds is mundane: a build pipeline pulled in a newer Nvidia SDK package than intended, and the DLL rode along for the ride. It is the kind of packaging mistake that happens more often than publishers like to admit, but it rarely hands the internet a preview of unreleased silicon-vendor technology.
What DLSS 5 Neural Rendering Actually Changes
DLSS has evolved from a simple upscaler into something closer to a full post-processing pipeline. Neural Rendering, as far as can be verified from the current leak, is aimed at reconstructing finer detail directly through a neural network pass rather than leaning purely on traditional temporal upscaling and frame generation. That is a meaningfully different job than what DLSS 4 does today, and it explains why the runtime is so much larger and so much heavier on the GPU.
The important caveat, repeated across every outlet that has tested the leak, is that this build is not close to finished. It is an early, unoptimized runtime that was never meant to run outside of Nvidia’s internal test environment, let alone across a mismatched set of games it was never validated against. Judging DLSS 5’s final performance by this leak is a bit like judging a shipped car by a mule chassis running on a test track. The chassis is real. The final tuning is not there yet.
How Modders Extracted and Deployed the Leaked Runtime
Once nvngx_dlssnr.dll was identified, the extraction process was straightforward by modding standards. Nvidia’s DLSS libraries are designed to be swapped in and out of a game’s install folder, which is exactly how the community has spent years forcing newer DLSS versions into older titles, or forcing DLSS into games that only shipped with a rival upscaler. Pulling the file from NBA 2K27 and dropping it into another game’s directory took modders very little time once the discovery went public.
That accessibility is part of why this leak spread so fast. DLSS DLL swapping is one of the most common tricks in the PC modding toolkit, and a community already fluent in doing it for DLSS 3 and DLSS 4 didn’t need to learn anything new to start poking at DLSS 5. The barrier to entry was close to zero, which is exactly why Nvidia is unlikely to be pleased about how this went public.
The Games Modders Have Already Tested It On
Reports confirm the leaked runtime has been forced into a handful of existing PC titles as testers try to get a feel for what DLSS 5 Neural Rendering looks and performs like outside its native NBA 2K27 environment. The games named across coverage include Control, Final Fantasy VII Rebirth, The Elder Scrolls V: Skyrim, Cyberpunk 2077, and Grand Theft Auto V. Control has become the de facto benchmark title for this leak, largely because its existing ray-traced lighting workload makes any added GPU overhead easy to spot.
Results across these titles are described as inconsistent. Neural Rendering visibly changes the look of surfaces and character models in some scenes, occasionally in ways testers describe as overcooked or unnatural, which lines up with the “unfinished” framing that has followed the leak from day one. This is a preview build behaving like a preview build.
The Numbers: FPS Drops on RTX 5070 Ti and the Wider RTX 50-Series
The performance data is where this leak gets genuinely alarming, and it is also the part generating the most searches. Testing on an Nvidia RTX 5070 Ti in Control at 4K, with ray tracing set to high and DLSS running in Quality mode, produced roughly 71 to 72 frames per second with Neural Rendering switched off. Turning it on dropped that figure to roughly 35 FPS, a reduction of around 51 percent. That is not a minor image-quality tax. That is losing half your frame rate for a feature still labeled experimental.
A separate benchmark circulating in the same coverage describes a midrange RTX 50-series card in Control dropping from 86 FPS down into the low 30s, in the 32 to 34 FPS range, once Neural Rendering was enabled, a loss in the 40 to 50 percent range depending on the exact scene tested. The pattern holds across both data points: whatever DLSS 5 Neural Rendering is doing under the hood, it is currently GPU-hungry enough to erase most of the performance advantage DLSS is supposed to deliver in the first place.
| Test Configuration | GPU | Neural Rendering Off | Neural Rendering On | Approx. FPS Drop |
|---|---|---|---|---|
| Control, 4K, RT High, DLSS Quality | RTX 5070 Ti | ~71-72 FPS | ~35 FPS | ~51% |
| Control, midrange RTX 50-series test | RTX 50-series (midrange) | ~86 FPS | ~32-34 FPS | ~40-50% |
Two things stand out from that table. First, the drop is consistent across two different testers and two different cards, which suggests the overhead is coming from the neural network pass itself rather than a one-off configuration problem. Second, even the higher-end RTX 5070 Ti isn’t spared. If this were purely a low-end hardware issue, Nvidia’s flagship-adjacent tier would look better insulated than it does.
Why an “Unfinished” Build Performs This Badly
Neural rendering passes are, by design, computationally heavier than the temporal upscaling techniques DLSS has relied on since its first version. Running an additional inference pass across a scene, especially one still full of debug paths and unpruned network layers, is exactly the kind of workload that tanks frame rates before an engineering team gets a chance to optimize it. Shipping software almost always looks worse in this raw state than it will once a vendor spends months trimming the model, quantizing weights, and tuning driver-level scheduling around it.
That is the charitable read, and it is also the most likely one given Nvidia’s track record. DLSS 3’s frame generation launched rough in a handful of early titles before driver updates smoothed it out. DLSS 4’s multi-frame generation went through a similar early-adoption bumpiness. A neural rendering pass leaking this early, this unpolished, fits the same pattern rather than breaking it.
Nvidia’s Silence and What It Means
As of this writing, Nvidia has not confirmed NBA 2K27 as a DLSS 5 title, has not announced a general DLSS 5 release, and has not addressed the leak directly. That silence is not unusual for a company managing an unplanned disclosure, but it leaves reviewers, modders, and buyers working from incomplete information. Every performance figure attached to this leak needs an asterisk: this is not what DLSS 5 will look like at launch, it is what an accidentally-shipped preview build looks like today.
The lack of an official statement also means there is no confirmed timeline correction. Nvidia has previously discussed DLSS 5 in the context of its ongoing roadmap without committing to a hard release date, and nothing about this leak changes that publicly stated position. Readers should treat any specific launch date currently circulating online as unverified until Nvidia says otherwise.
Historical Context: How DLSS Got Here
DLSS has gone through a fairly public evolution since its debut. The first version, tied to the RTX 20-series, used a per-game trained network and produced mixed results. DLSS 2 moved to a generalized model that worked across titles without per-game training, which is the point most engineers consider the technology’s real inflection. DLSS 3 introduced frame generation alongside the RTX 40-series, synthesizing entirely new frames rather than just upscaling existing ones. DLSS 4 pushed that further with multi-frame generation for the RTX 50-series, generating multiple interpolated frames per rendered frame.
Neural Rendering, the feature tied to this DLSS 5 leak, would represent another step change: moving parts of the rendering pipeline itself, not just the upscaling or frame-interpolation stage, through a neural network. That is a bigger architectural shift than any single previous DLSS revision, which helps explain why an early build is hitting frame rates this hard. It is doing more work per frame than any prior version attempted.
| DLSS Generation | Paired GPU Generation | Core Technique |
|---|---|---|
| DLSS 1 | RTX 20-series | Per-game trained upscaling |
| DLSS 2 | RTX 20/30-series | Generalized temporal upscaling |
| DLSS 3 | RTX 40-series | Frame generation added |
| DLSS 4 | RTX 50-series | Multi-frame generation |
| DLSS 5 (leaked build) | RTX 50-series (tested) | Neural Rendering pass (unconfirmed final form) |
That table is a useful reminder of how quickly Nvidia has been iterating. Four major DLSS revisions across three GPU generations is an aggressive cadence, and it puts real pressure on Nvidia to keep each new version feeling meaningfully better than the last rather than just heavier. For more on how the officially-announced version of DLSS 5 is positioned to launch, see our earlier coverage of Nvidia’s planned DLSS 5 debut and the separate report on modders running DLSS 5 on older RTX 4090 hardware.
Market Impact: What This Means for Nvidia, AMD, and Game Studios
For Nvidia, an uncontrolled leak is mostly a communications headache rather than a financial one. DLSS remains a software differentiator layered on top of hardware Nvidia already sells at a premium, and a rough early build doesn’t change the underlying silicon roadmap discussed around Nvidia’s most recent record quarter. But it does complicate the marketing narrative. Nvidia would almost certainly have preferred to introduce Neural Rendering with curated demo scenes and controlled benchmarks, not a 51 percent frame-rate hit measured by modders in a game it never intended to ship it in.
For AMD and Intel, the leak is a minor opening. Every headline about DLSS cutting frame rates in half, even with the unfinished-build caveat attached, is a headline that makes FSR and XeSS look comparatively stable by contrast. Neither AMD nor Intel has to do anything to benefit from this news cycle; they just have to let it run. For game studios integrating Nvidia’s SDKs, the leak is also a quiet warning about build hygiene: a sports title with no announced neural rendering feature still managed to ship the DLL that started this whole story, which is the kind of packaging slip most publishers dread seeing dissected in public.
Competitive Landscape: DLSS 5 vs FSR and XeSS
DLSS has generally held a perception edge over AMD’s FSR in blind image-quality comparisons, an advantage covered in our DLSS vs FSR comparison. Neural Rendering, if it ships close to what’s leaked, would push that gap further on the image-quality side while temporarily giving up ground on raw performance, at least until Nvidia optimizes the runtime. That is a real tradeoff for competitive and high-refresh-rate gamers who care more about frame rate than incremental detail reconstruction.
Intel’s XeSS occupies a different lane entirely, aimed at broader hardware compatibility rather than chasing Nvidia’s top-end image quality. None of that changes because of this leak, but the timing is notable: Nvidia is trying to widen its technical lead at the exact moment its own hardware ecosystem, including products discussed in our coverage of the RTX Spark rollout across OEM partners, is scaling into more devices than ever. A heavier, more demanding DLSS 5 has to run well across that whole spread, not just on flagship cards.
Why Leaked Runtimes Like This Are a Recurring Risk
This is not the first time an unreleased Nvidia component has escaped through a third-party game build, and it will not be the last. Modern GPU vendors ship SDK components through shared driver and middleware channels that touch dozens of studios simultaneously, and any one of those studios can accidentally bundle a file that was never meant to leave internal testing. The DLSS DLL-swapping culture that PC gamers built up over years of legitimate performance tweaking is the same mechanism that makes leaks like this spread within hours instead of staying contained.
There’s also a quieter risk worth naming: running an unverified, unofficial binary pulled from a leaked game build inside other software is not without its own trust implications. Nvidia’s official DLSS libraries are signed and distributed through vetted channels; a DLL extracted from an early-access build and manually dropped into unrelated games sits outside that chain of custody. Readers experimenting with the leak should understand they are running software Nvidia never intended, or verified, for that use.
Predictions: Where DLSS 5 Goes From Here
- Nvidia will eventually confirm DLSS 5 and Neural Rendering formally, likely with curated benchmarks that look far better than the leaked numbers, once the runtime is optimized closer to launch.
- The final shipping version will claw back most of the performance loss. A 51 percent hit is not something Nvidia can ship at scale; expect driver and model optimization to bring that closer to single-digit or low-double-digit overhead by general release.
- 2K and other studios will tighten SDK packaging checks after this incident, since an accidental leak through a sports title’s early-access build is an embarrassing way for a hardware partner’s unreleased tech to go public.
- Expect a narrower, curated wave of “official” DLSS 5 previews from Nvidia in the months following this leak, an attempt to reset the public narrative away from the rough modded footage currently circulating.
- AMD and Intel will lean into performance-per-watt messaging in their next FSR and XeSS updates, using this leak’s frame-rate numbers as an implicit contrast point even without naming DLSS 5 directly.
What Buyers and Modders Should Actually Do Right Now
For most PC gamers, the practical advice is simple: don’t chase this leak for daily use. The performance cost is severe, the visual results are inconsistent, and nothing about the current build represents what Nvidia will eventually ship. Curiosity-driven testing is one thing; relying on an unofficial, unsigned DLL pulled from an early-access sports game for regular play is a different risk calculation entirely, especially for anyone running competitive titles where frame-time consistency matters more than a marginal detail bump.
Buyers evaluating RTX 50-series purchases should also resist reading too much into the leaked benchmarks. The RTX 5070 Ti’s roughly 51 percent frame-rate hit says more about an unfinished neural network than it does about the card’s underlying capability. Judging hardware by software that isn’t done cooking is how misleading narratives get started, and this leak has already generated plenty of those in gaming forums over the past several days.
Frequently Asked Questions
What is nvngx_dlssnr.dll?
It is a library file found inside NBA 2K27’s PC early-access build, identified as version 310.8.0.0 and roughly 158 MB in size. Reports connect it to Nvidia’s DLSS 5 Neural Rendering feature.
How did DLSS 5 leak through NBA 2K27?
The exact cause hasn’t been confirmed publicly. The most widely discussed explanation is that the game’s build pipeline pulled in a newer Nvidia SDK package than intended, bundling an unreleased file into the shipped early-access build.
Which games have modders tested the leaked DLSS 5 build in?
Reports name Control, Final Fantasy VII Rebirth, The Elder Scrolls V: Skyrim, Cyberpunk 2077, and Grand Theft Auto V as titles where modders forced the leaked runtime to run.
How much does DLSS 5 Neural Rendering hurt frame rates?
Testing on an RTX 5070 Ti in Control at 4K with ray tracing on high and DLSS Quality showed a drop from roughly 71-72 FPS to about 35 FPS with Neural Rendering enabled, close to a 51 percent reduction. A separate test on a midrange RTX 50-series card showed a comparable 40-50 percent drop.
Has Nvidia officially confirmed DLSS 5?
No. As of this report, Nvidia has not confirmed NBA 2K27 as a DLSS 5 title and has not announced a general release date or official specifications for DLSS 5.
Is it safe to run the leaked DLSS 5 file in other games?
It carries real risk. The file is unsigned in the sense that it was never distributed through an official, sanctioned channel for use outside NBA 2K27, and it’s an early, unoptimized build. Most players should treat it as a curiosity rather than something to use for regular gaming.
Will the final version of DLSS 5 perform this badly?
Unlikely. Early, unoptimized builds of new rendering technology typically run far heavier than shipping versions. Nvidia has a track record of tuning frame generation and upscaling features significantly between early testing and public launch.
Does this leak affect current DLSS 4 performance?
No. DLSS 4, currently shipping in RTX 50-series titles, is unaffected. This leak concerns a separate, unreleased runtime and does not change the performance of games already using DLSS 4.
Sources and further reading: Nvidia’s official DLSS technology page, Nvidia GeForce news, Tom’s Hardware GPU coverage, VGTimes’ report on the leak, and Wccftech.
