NVIDIA has officially confirmed that DLSS 5 neural rendering, previously believed to be exclusive to the RTX 50-Series "Blackwell" lineup, will eventually be extended to the previous-generation RTX 40-Series "Ada Lovelace" GPUs. DLSS 5 officially launches today with 2K Games' NBA 2K27, supported on all GeForce RTX 50-Series desktop and laptop GPUs as well as GeForce NOW. According to NVIDIA, the technology is optimized for Blackwell's fifth-generation Tensor cores, while support for Ada Lovelace's fourth-generation Tensor cores is still being worked on.
NVIDIA describes DLSS 5 as its most computationally demanding model to date. The company says that since announcing the technology in March, it has achieved a 5X performance gain through hardware, software kernel, and co-design optimizations, and expects further gains. The demo hardware also evolved during development, starting on dual RTX 5090 GPUs before NVIDIA optimized the model to run on a single RTX 5090. NVIDIA states its current focus is tuning performance for the RTX 50-Series, with model updates expected later this fall; only after that work is more complete will the company begin expanding official support to RTX 40-Series cards. One report notes that RTX 40 GPUs lack Multi-Frame Generation (MFG) support, which is part of why additional tuning is needed before the model becomes viable on that hardware, and suggests RTX 30-series support could potentially follow down the road.
Sources differ slightly on the performance impact of DLSS 5. One report cites an existing average performance hit of 50-60%, with expectations that ongoing optimizations could reduce this to a 10-15% impact by the end of the year, potentially removing the need to pair it with MFG modes or Performance-mode upscaling. Another report simply notes that gamers should expect a noticeable drop in frame rates, especially with the initial release, given how resource-intensive the technology is.
DLSS 5 centers on 3D-guided neural rendering, which NVIDIA says is designed to deliver movie-grade visual realism and photorealistic lighting in games. The AI model is trained to infer how complex scene elements—such as characters, skin, hair, and environmental lighting—should appear based on inputs from the game engine, including color buffer and motion vector data, combined with an input frame. NVIDIA says the technology offers different AI models and controls that let developers decide what gets enhanced, where effects appear, and how strongly they are applied, while for players it is designed to function as a simple on/off graphics option.
Ahead of the official launch, the DLSS 5 DLL reportedly leaked, allowing modders to port the technology to various games last week, including onto older RTX 40-series cards not yet officially supported. Some testers also ran the leaked DLL on RPCS3, the open-source PlayStation 3 emulator. In response, the official RPCS3 account criticized the technology, calling it an "AI-slop generator" and stating that while the gaming industry pushes more upscalers and frame generation to "hallucinate games and hide their lack of optimisation," the project does not ship that kind of technology. The comment drew criticism from some users, with one calling it "possibly the dumbest" response the project could have given, arguing that DLSS and FSR frame generation can complement game-engine optimization rather than simply mask poor performance, and that NVIDIA's Reflex technology can help offset the latency penalty associated with frame generation.
RPCS3 developers further explained that emulators pursue open-ended goals rather than simply targeting playable status on high-end hardware, and that implementing modern temporal upscalers is difficult in their case because PS3-era games do not expose the motion-vector data such upscalers require. The developers added that RPCS3's low-level RSX emulation approach makes reconstructing that data impractical, which is why the emulator currently relies on simpler spatial upscaling methods such as FSR 1, bilinear, and nearest-neighbor scaling.