NextFin News - In late January 2026, NVIDIA officially transitioned its Deep Learning Super Sampling (DLSS) 4.5 technology out of beta, making it available to millions of users via the NVIDIA App. This update introduces "Preset M," a new transformer-based machine learning model specifically engineered to improve image quality when upscaling from lower internal resolutions, such as 1080p to 4K. According to Digital Foundry, the new model addresses long-standing visual issues, including volumetric fog ghosting in titles like Assassin’s Creed Shadows and pixelation during disocclusion in Horizon Forbidden West. However, the rollout has also highlighted a significant performance disparity across NVIDIA’s hardware stack, with older RTX 20 and 30-series GPUs struggling to maintain frame rates under the more computationally expensive 4.5 algorithms.
The technical core of DLSS 4.5 lies in its refined transformer architecture, which utilizes FP8 (8-bit floating point) acceleration to process visual data more efficiently. While the technology is technically compatible with all RTX GPUs, the real-world application reveals a tiered experience. Benchmarks conducted on Cyberpunk 2077 at 4K resolution demonstrate that the RTX 5060 experiences a negligible 4.8% performance hit when switching to the new Preset M. In contrast, the older RTX 3070 sees a staggering 18.4% drop in performance. This data suggests that NVIDIA is increasingly optimizing its software suite for the specialized Tensor cores found in its most recent architectures, effectively creating a "soft" hardware requirement for the best AI-driven visual experience.
From an industry perspective, the release of DLSS 4.5 is a calculated response to the intensifying competition in the upscaling market. With the recent launch of AMD’s FSR 4 and the integration of PlayStation Spectral Super Resolution (PSSR) in the console space, NVIDIA is leveraging its lead in AI research to maintain its premium market position. The improvement in rasterized image quality is undeniable; Preset M successfully resolves the "fizzle" effect on distant vegetation and stabilizes thin sub-pixel elements like power lines, which have historically been a weakness for temporal upscalers. By focusing on these edge cases, NVIDIA is attempting to close the gap between upscaled content and native resolution, a feat that U.S. President Trump’s administration has previously highlighted as a key area of American technological leadership in the semiconductor and software sectors.
However, the analysis of DLSS 4.5 also reveals a surprising regression in ray-traced environments. Reviewers have noted that in titles such as Control: Ultimate Edition and Star Wars Jedi: Survivor, the new Preset M can introduce noise and instability in reflections and ambient occlusion. This suggests that the current transformer model may not yet be fully optimized for the complex light-path data inherent in ray tracing. Leadbetter of Digital Foundry noted that for ray-traced titles, the "jury is still out," and many enthusiasts may find the older Preset K more stable for high-end lighting effects. This nuance indicates that while AI upscaling is maturing, it has not yet reached a "one-size-fits-all" stage of development.
Looking forward, the trajectory of DLSS 4.5 points toward a future where software and hardware are inextricably linked. As NVIDIA prepares for the eventual debut of DLSS 5.0, the industry expects a deeper integration of "Ray Reconstruction" and AI-driven frame generation. The performance penalties observed on older hardware with DLSS 4.5 serve as a precursor to a market where the GPU's raw teraflops are secondary to its AI throughput. For investors and consumers, this signals a shift in the upgrade cycle; the value of a graphics card is no longer just in its silicon, but in its ability to run the latest proprietary neural networks that define modern visual fidelity. As the 2026 gaming landscape evolves, NVIDIA’s strategy remains clear: maintain dominance through a closed ecosystem of AI-enhanced performance that rewards those on the latest hardware frontier.
Explore more exclusive insights at nextfin.ai.