Leaked Build of NVIDIA's DLSS 5 Neural Rendering Tech Shows Uncanny Visuals and Performance Dip

An unreleased version of Nvidia Corp.’s upcoming Deep Learning Super Sampling (DLSS) technology has leaked online, providing an early look at the graphics giant's latest direction in artificial intelligence-driven real-time rendering. The pre-release software build, which was extracted from an early-access PC release of NBA 2K27, has been modified by software enthusiasts and grafted onto several existing video games. However, initial tests of the leaked DLSS 5 files have yielded unspectacular and often surreal visual outcomes alongside significant hardware performance drops, raising questions about the early state of the software.

First reported by Engadget, the unreleased DLSS 5 iteration marks a fundamental departure from Nvidia’s previous image reconstruction techniques. Rather than simply scaling up lower-resolution source images into sharper output frames through conventional temporal upscaling algorithms, the updated software introduces a generative neural rendering framework. According to official Nvidia product language cited in the report, the neural rendering system "infuses pixels with photoreal lighting and materials" directly during the frame generation process, generating stylized surface detail rather than merely sharpening existing geometry.

In practice, early attempts to implement the leaked build into older titles have produced results characterized by a pronounced uncanny valley effect. When applied to existing game assets, the technology attempts to apply hyper-realistic skin textures, surface material details, and complex lighting dynamics onto legacy character models. However, because the technology operates as a rendering overlay without modifying the underlying character models, polygonal geometry, or animation structures, the visual output frequently presents a jarring contrast between realistic surface imagery and outdated engine assets.

Specific examples documented by users highlight the stark visual changes resulting from the unofficial implementations. In Remedy Entertainment’s action game Control, the algorithm was applied to lead character Jesse Faden. While the protagonist is designed in-game as approximately 28 years old, the DLSS 5 neural rendering model generated facial textures that made the character appear middle-aged. Observers noted that while the underlying artificial intelligence logic behind the facial transformation was discernible from the original source face, the final output simultaneously appeared more realistic and more synthetic, failing to match the character's original design intent.

Similar visual discrepancies were documented in Bethesda Game Studios’ open-world title The Elder Scrolls V: Skyrim. When running with the leaked DLSS 5 build, the game's lower-polygon character models appeared as if a high-definition 3D scan of an actual human face had been stretched over a fifteen-year-old digital figure. The unnatural contrast was exacerbated by the fact that the underlying facial expressions and body movements remained completely unmodified, leaving hyper-detailed facial renderings to execute stiff, mechanical animations and stare blankly into the distance.

Beyond visual anomalies, the leaked software build appears to impose a severe computational penalty on system performance, conflicting with the traditional purpose of upscaling technologies. In testing on Remedy Entertainment's Control at 4K resolution using an Nvidia GeForce RTX 5070 Ti graphics card, enabling the DLSS 5 Neural Rendering feature caused frame rates to plummet from 71 frames per second down to 35 frames per second. For an upscaling feature historically designed to improve performance efficiency on consumer hardware, the steep drop in frame rate indicates that the generative lighting and pixel synthesis models require vast amounts of compute power in their current state.

The leak comes as Nvidia increasingly positions itself around enterprise artificial intelligence infrastructure and advanced generative models. While the company built its core business on consumer graphics hardware, its push toward generative pixel synthesis reflects its broader focus on AI technology. Industry observers emphasize, however, that the leaked files represent an unoptimized and unauthorized software extraction. Nvidia is officially scheduled to launch DLSS 5 this fall, at which point developer-sanctioned integrations using finalized production code will demonstrate whether the technology can deliver natural lighting enhancements without sacrificing rendering performance.