Why the DLSS 5 Leak Rewrites NVIDIA's Deal With Gamers
A leaked build of DLSS 5 reveals Neural Rendering—a feature that demands nearly double the GPU cost instead of saving frames. The implications reach far beyond a single gaming trick.
A File Leaked From a Basketball Game. What It Uncovered Is Bigger Than That.
A single graphics DLL walked out of an early access build of NBA 2K27 and before the afternoon was over, it was running in games it was never meant to touch. That is the surface of the DLSS 5 leak. What lies beneath the clips circulating right now is something more consequential than a meme: a free, unsanctioned preview of what NVIDIA wants players to sacrifice in exchange for its most aggressively marketed AI feature yet.
The leak itself was almost banal. Renan Maniero, a hardware observer, noticed the library inside the NBA 2K27 folder. Within hours, modders in the RenoDX community hot-wired it into Control, Skyrim, and Final Fantasy VII Rebirth. What they found is now making the argument for them.
For a Decade, Turning On NVIDIA’s Feature Made Games Faster. This Time It Makes Them Slower.
DLSS built its reputation on giving something for nearly nothing: render a game at a lower resolution, run it through an AI upscaler, and pocket the frame difference. Neural Rendering, the headline addition in DLSS 5, flips that bargain.
Instead of reconstructing cheap images, the GPU now applies a real-time AI pass on top of whatever it has already drawn. Faces, textures, lighting—all rewritten on the spot. This does not reduce work. It adds to it.
The cost appears the moment the toggle is flipped. On an RTX 5060, Control drops from 86 frames per second to 34. An RTX 5070 Ti falls from roughly 71 fps to 35 at 4K. Turn the slider all the way up in Grand Theft Auto V and one system went from 90 fps to 29. In Final Fantasy VII Rebirth, frame rates approximately halved while video memory usage spiked. The pattern is consistent across cards and titles: enable the feature and you pay nearly half your performance back.
The Feature Is Narrower Still.
The leak arrived in a form that deserves a second look. This build demands RTX 50 series hardware. One modder barely got it running on a previous-generation RTX 40 series card by patching CUDA calls, and even then performance suffered. On AMD and Intel GPUs, the feature does not run at all. The honest version is narrow: it works well on NVIDIA’s newest cards, works poorly on last generation’s, and does not work on anyone else’s.
That exclusivity is not new for NVIDIA. What is unusual is that the premium feature attached to the 50 series—the one the company will point to when trying to move hardware—now carries a performance penalty that most buyers will notice immediately.
The Visuals Raise Questions NVIDIA Has Not Answered Yet.
With the sliders left at default, Neural Rendering gives Skyrim characters a flat, lifeless quality to their eyes and reconstructs Final Fantasy VII’s Cloud into someone his own animators might struggle to recognize. NVIDIA has said the technology will come with a set of developer controls—intensity, style, skin masking—so that artistic direction stays in creators’ hands. But the leaked build shows what happens when nobody adjusts those controls. It is off-key. It is not subtle.
This is the detail that turns the leak from curiosity into a problem. NVIDIA still plans to ship Neural Rendering this fall. It is positioned as a core selling point for the 50 series. The leak lets buyers see in actual games what the feature costs, what it looks like at rest, and what hardware it requires—without waiting for a marketing deck.
What This Means for the GPU Upgrade Cycle
For years, the upgrade pitch was simple: buy the new card, turn on the feature, gain frames. Neural Rendering inverts that script. The pitch is now: buy the new card, turn on the feature, lose frames—and hope the visual upgrade justifies it.
That changes the calculus for two groups. First, the people already holding RTX 40 series cards. They cannot run the feature properly even if they wanted to. Second, the people considering the RTX 5060 or 5070 Ti as an upgrade from anything older. The promise attached to those cards is no longer purely additive; it carries a visible performance tax that must be weighed against whatever aesthetic gain the sliders eventually provide.
The leak also makes the upgrade decision more immediate. When a feature arrives locked behind specific hardware and then costs nearly half your framerate on that same hardware, the question is no longer when to upgrade but whether the upgrade was worth the premise. That creates pressure on launch pricing and on the framing of the 50 series.
AMD’s FSR 5 Is Already Being Shaped by This Leak
AMD does not have a Neural Rendering equivalent to point at today. But the leak has given its engineers and marketers a template to react against. FSR 5 will likely double down on the older DLSS bargain—preserving or gaining frames while improving image quality—because that is where the competitive high ground sits after this revelation. If NVIDIA’s new feature is the one that slows games down, AMD’s answer can be the one that does not.
The pressure is also strategic. Every month the 50 series sits in stores with a headline feature that halves performance by default, the longer AMD has to refine FSR 5 without being forced into a race to match an AI-pass approach it may not want to join. The leak has effectively handed AMD the opening it needed.
Who Wins, Who Loses
NVIDIA wins on exclusivity. The feature reinforces the RTX 50 series as a destination purchase and keeps the ecosystem locked to its latest silicon. But it also risks trust. For a decade, the implicit contract was that enabling an NVIDIA feature made things better for the player. This breaks that contract in real time.
Gamers lose on performance and gain on information. They now know, from an actual build, what the trade-off looks like before they buy. Modders and hardware reviewers win on timing—they have material to publish while NVIDIA is still waiting for official controls.
Developers face an open question: how much control will NVIDIA actually let them exercise over intensity and masking once the feature ships? If the defaults remain as aggressive as the leak suggests, the developer toolkit becomes less about artistry and more about damage management.
What Happens Next
NVIDIA has not commented on the leak itself. The company is still on track to release Neural Rendering this fall. When it arrives, the relevant question is not whether the feature exists—leakers have already proven that—but whether the shipped version gives developers enough knobs to make the performance hit worth it for the average buyer.
Until then, the leak has done something unusual: it has let the market test a flagship feature before the launch window closes. That is rare. And it means the narrative around the RTX 50 series will be shaped less by NVIDIA’s keynote and more by the data sitting in a leaked DLL right now.
Ten years of turning on an NVIDIA feature meant your game ran faster. For the first time, it may mean your game runs slower. A basketball game leaked the file. Gamers are already doing the math.