The latest iteration of NVIDIA’s machine learning-based upscaling technology, DLSS 4.5 (Deep Learning Super Sampling), debuted in January 2026 for all GeForce RTX GPU users. This upgrade is available through the NVIDIA app updates and is supported by the new Game Ready driver. DLSS has become an essential tool for PC gamers, aiming for high performance while maintaining stunning visuals by rendering games at lower resolutions and employing a deep neural network to enhance images, ensuring minimal loss to visual fidelity.
A key feature of DLSS 4.5 is its second-generation transformer model for Super Resolution, designed for enhanced temporal upscaling. This model is trained on an extensive dataset and promises to provide significant improvements in image reconstruction, anti-aliasing, and temporal stability. The enhancements include sharper details, smoother edges, and a reduction in visual artifacts such as shimmering and ghosting. This model allows for better preservation of high-contrast lighting and complex details compared to its predecessors.
In addition to Super Resolution, DLSS 4.5 introduces significant upgrades to frame interpolation technology with Dynamic Multi-Frame Generation (MFG) and the new 6X MFG mode, specially tailored for the RTX 50 series. Set to release in spring 2026, this feature can generate up to five additional frames per rendered frame and intelligently adjust the number of created frames dynamically to achieve target framerates—a function especially beneficial for high refresh rate displays.
DLSS 4.5 Super Resolution is currently available in hundreds of supported games. It marks a substantial advancement in ML-based temporal upscaling quality and adaptability. This article will delve deeper into Super Resolution, exploring its visual differences from previous versions, how to activate it for testing, the anticipated image quality impacts in selected games, and its performance implications compared to the older generation model.
Understanding DLSS 4.5: A Technical Overview
DLSS 4.5 enhances the existing DLSS 4 framework by upgrading to a second-generation transformer model that advances temporal upscaling fidelity while improving Multi-Frame Generation technology, allowing for a variable number of generated frames—up to five—to optimize performance.
DLSS 4.5 Super Resolution
The heart of DLSS 4.5 lies within its enhanced transformer model designed for Super Resolution. This update aims to elevate temporal upscaling to new heights.
Second-Generation Transformer Model Unlike previous iterations that utilized convolutional neural networks, the DLSS 4.5 model has been retrained on a more extensive, high-fidelity dataset, leveraging vastly superior compute resources. This leads to:
- Enhanced spatial and temporal detail understanding, facilitating finer detail reconstruction including lighting and edges.
- Greater temporal stability, resulting in diminished shimmer and ghosting artifacts, with improved image quality even during motion.
- Better handling of high-contrast elements, accurately reproducing light reflections and highlights through direct linear space processing.
NVIDIA engineers estimate that the new transformer model requires roughly five times the inference compute power of its predecessor. It also optimizes pixel sampling and motion vector data from the game engine, resulting in higher-quality images characterized by sharper anti-aliasing, improved motion clarity, and superior lighting fidelity when compared to the former DLSS 4 SR model.
Dynamic Multi-Frame Generation
Beyond the upscaling capabilities, DLSS 4.5 also refines NVIDIA’s Multi-Frame Generation technology. This innovative feature is designed to interpolate additional frames between rendered ones, which enhances overall smoothness at the expense of latency and potential visual artifacts. While the complete rollout is scheduled for spring 2026, the principles of Dynamic MFG are well established.
Introducing 6X Multi-Frame Generation Mode The update allows for a new 6X Multi-Frame Generation mode—increasing the maximum from the previous 4X mode—that can generate up to five extra frames for each rendered frame, significantly boosting perceived smoothness on monitors with high refresh rates. This is particularly advantageous in graphically demanding scenarios, provided the pre-MFG frame rate is sufficiently high to offset latency and visual artifact concerns.
Dynamic Multi-Frame Generation Unlike its static predecessor, DLSS 4.5 incorporates a dynamic approach to frame generation. This system adjusts the number of generated frames in real-time based on GPU performance and display refresh rate (or a chosen framerate limit).Practically, this means:
- When a GPU struggles to meet a set target refresh rate, Dynamic MFG compensates by generating additional frames to enhance smoothness.
- As GPU workload eases, the system scales back generated frames, optimizing the balance of smoothness, system latency, and visual clarity.
DLSS 4.5 dynamic MFG technology functions synergistically with NVIDIA’s Reflex Low Latency technology to maintain gameplay responsiveness while active, offering a responsive and flexible approach to frame interpolation adapted to the player’s hardware and the game being played.
Activating DLSS 4.5 Super Resolution
Users can enable DLSS 4.5 Super Resolution via the NVIDIA App’s latest update alongside the latest GeForce Game Ready driver. This update refines how Super Resolution Presets are applied, simplifying the access to the advancements offered by the new transformer model across games and applications.
Updating NVIDIA App and Graphics Driver
Ensure that both NVIDIA App and your graphics driver are updated before activating DLSS 4.5 Super Resolution in your games. This can be done through the app, which should automatically refresh upon launch. The current version supports Presets—M and L—which can be configured in the “DLSS Override – Model Presets“in NVIDIA App settings, applicable either globally or per game.
Follow these steps to update your settings:
- Open the NVIDIA App from a desktop shortcut, system tray icon, or via a right-click on the desktop to access the Graphics tab.
- Select a specific game from the “Program Settings“section or opt for “Global Settings“to apply a Super Resolution Preset override across all compatible games.
- Navigate to the “DLSS Override – Model Presets“under the “Driver Settings“section and select your preferred SR model.
New Recommended Super Resolution Setting
The NVIDIA App previously utilized a “Latest”setting to select the most recent DLSS SR model. However, with DLSS 4.5, this has been replaced with a “Recommended”setting that aligns the optimal Super Resolution model with your chosen quality mode. Its operation includes:
- Model M is used for Performance mode (50% resolution axis).
- Model L stands as the preferred option for Ultra Performance mode (33% resolution axis).
- Model K will be utilized across other SR modes, including DLAA (100% resolution axis), Quality (67% resolution axis), and Balanced (58% resolution axis).
Using the Recommended option ensures that the NVIDIA App chooses the ideal DLSS Super Resolution model based on your upscaling requirements, striking a favorable balance between quality and performance efficiency. Note that earlier RTX GPUs (20 Series and 30 Series) may struggle with the new DLSS 4.5 models, primarily due to their tensor cores’ inability to support ML inference with FP8 precision.
After configuring your SR overrides, you can verify the active upscaling Preset by enabling the in-game performance overlay in the NVIDIA App, accessed by the hotkey ALT+Z. Subsequently, navigate to the “Statistics“menu at the bottom of the overlay settings.
In the “Statistics“menu, toggle on the “Show statistics in heads up display“setting and select either “DLSS“or “Custom“in the “Statistics view“option. Ensure that the “Super Resolution Model Override (SR Model OVR)“indicator is active to confirm which SR Preset is currently implemented.
Finally, launch a game to test the DLSS Super Resolution settings you have configured, ensuring the NVIDIA overlay displays the expected SR model.
Evaluating Visual Performance: DLSS Super Resolution 4.5
To assess DLSS 4.5 Super Resolution’s capabilities, we conducted visual comparisons involving both static screenshots and side-by-side video footage—with slow-motion playback at 50% speed—to identify performance at various resolutions and upscaling modes. As static images often overlook motion-related artifacts, our evaluations encompass gameplay dynamics, focusing on aspects like shimmering, ghosting, and temporal instability.
This analysis evaluates five modern titles: Assassin’s Creed Shadows, Black Myth: Wukong, Grand Theft Auto V Enhanced, Cyberpunk 2077, and Horizon Zero Dawn Complete Edition—with each utilizing a different game engine to reflect the strengths and weaknesses of DLSS SR’s temporal upscaling.
Cyberpunk 2077 (1080p)
Static Screenshots:
Video Footage:
Watch Video
Assassin’s Creed Shadows (1440p)
Static Screenshots:
Video Footage:
Watch Video
Black Myth: Wukong (1440p)
Video Footage:
Watch Video
Grand Theft Auto V Enhanced (4K)
Static Screenshots:
Video Footage:
Watch Video
Horizon Zero Dawn Complete Edition (4K)
Static Screenshots:
Video Footage:
Watch Video
Analysis of Image Quality Comparisons
Visual quality comparisons reveal that the DLSS SR 4.5 model enhances detail and sharpness, frequently resulting in exceptionally crisp images, albeit at times overly sharpened. The model’s strengths become evident when upscaling from lower resolutions, as NVIDIA indicates Preset M for Performance mode and the more visually impressive Preset L for Ultra Performance mode, both showcasing commendable performance at 4K and higher resolutions.
In video comparison, DLSS 4.5 presents significant visual improvements during motion, diminishing ghosting, shimmering, and disocclusion artifacts while enhancing temporal stability. However, it does not entirely eradicate existing rendering artifacts, particularly in foliage rendering, which may appear exacerbated. Overall, the new models excel in high-performance modes, allowing visual standards to rival traditional higher-quality upscaling methods.
Performance Implications
Community feedback and initial evaluations indicate that DLSS 4.5 Super Resolution incurs a higher performance hit than the earlier DLSS 4 SR model. The intensified compute demands of the second-generation model can lead to performance declines, particularly on older RTX GPUs that are less capable of handling the increased workload effectively.
To quantify the performance variations between DLSS SR 4.5 and 4 across all modes, including the native resolution DLAA mode, we conducted benchmarks using CapFrameX in Black Myth: Wukong at a resolution of 2560×1440 on an updated NVIDIA GeForce RTX 4090 system. The following results outline the findings:
DLAA Mode Performance Comparison (Preset M vs Preset K)
Quality Mode Performance Comparison (Preset M vs Preset K)
Balanced Mode Performance Comparison (Preset M vs Preset K)
Performance Mode Performance Comparison (Preset M vs Preset K)
Ultra Performance Mode Performance Comparison (Preset L vs Preset K)
The performance metric comparisons indicate that the newly enhanced upscaling model significantly increases GPU demand, with performance drops becoming more pronounced at higher rendering resolutions.
Final Thoughts
DLSS 4.5 Super Resolution yields significant visual enhancements over previous versions. Gamers may notice sharper anti-aliasing, better temporal stability, and fewer artifacts in high-contrast or motion-intensive scenes. However, the model is not without flaws, as certain instances of oversharpening and foliage flickering remain prevalent in select titles.
Performance-wise, the newer Super Resolution model demands more resources than earlier models, particularly on older GPUs that exhibit pronounced performance drops. Community feedback highlights considerable debilitation on RTX 20 and 30 Series GPUs, especially when deployed in demanding settings, raising questions on whether the improved image quality justifies the greater hardware burden.
Ultimately, while DLSS 4.5 Super Resolution signifies a notable advancement in visual quality—offering clearer images and improved motion fidelity—it inflicts performance costs and can lead to occasional visual anomalies, making it prudent to evaluate its application on a case-by-case basis.
Related Articles:
Microsoft Introduces Xbox Mode for All Windows 11 PCs, Merging Console and Desktop Gaming Experiences
20:22
Samsung Ends LPDDR4 Memory Production Amid 50x Profit Surge, Predicts Increasing Memory Shortages by 2027
19:07
Ultimate Guide to Windrose PC Performance Analysis and Tuning for Optimal PC Experience
18:18
Leave a Reply