Last year marked a significant milestone in the battle of upscaling technologies, as AMD introduced FSR 4, finally offering a competitive response to Nvidia’s DLSS (Deep Learning Super Sampling) technology after nearly six years. FSR 4 garnered positive feedback from reviewers, who generally placed it in between DLSS 3 “CNN”and DLSS 4 “Transformer”regarding image quality, albeit at a performance deficit when compared to DLSS 4. The recent unveiling of DLSS 4.5 at CES 2026 suggests that Nvidia has regained a competitive edge in upscaled image quality.
Exploring DLSS 4.5: The Next Generation Transformer
DLSS 4.5 brings several enhancements over its predecessor, DLSS 4, resulting in improved visual fidelity. Key upgrades include:
- Utilization of a significantly advanced second-generation Transformer model
- A larger and more refined training dataset for improved image quality
- An increased computational cost that is five times that of DLSS 4
- Support for FP8 precision on RTX 4000 and 5000 series, which helps mitigate performance impacts
To leverage DLSS 4.5, users can access the Nvidia App, select their game, navigate to the Driver Settings, and adjust the ‘DLSS Override – Model Presets’ to either ‘Preset M’ (Performance) or ‘Preset L’ (Ultra Performance mode).In this review, focus will be on Preset M.

Testing Framework
To accurately evaluate the latest advancements in upscaling technologies from Nvidia and AMD, I used the RX 9060XT 8GB for FSR 4 and the RTX 5060 Ti 16GB for DLSS 4.5. Both GPUs were inserted into a testing rig boasting the following specifications:
- Intel i7 13700k (optimized on MSI Performance Settings)
- MSI Z790 MPower motherboard
- 32GB (2x16GB) DDR5 7200 RAM
- Seasonic Focus V3-GX 1000W Power Supply Unit
- Arctic Liquid Freezer II 420MM All-in-One cooler
- Two SN850X 1TB SSDs
- Windows 11 24H2
- AMD Radeon Adrenaline Driver 25.12.1
- NVIDIA GeForce Game Ready Driver 591.74
The games selected for testing included native implementations of FSR 4, found either through in-game menus (e.g., Cyberpunk 2077) or via driver updates like in Clair Obscur: Expedition 33. Additionally, I conducted an experimental review of Forza Motorsport with FSR 4 enabled through Optiscaler using DLSS inputs.
It is important to note that Black Ops 7 does not show Driver Settings options on the Nvidia App when accessed via Game Pass. To enable the M preset in this case, I utilized DLSS Swapper. However, using such workarounds for multiplayer titles is not generally advisable.
Footage was captured using the built-in hardware encoders of the respective graphics cards, as I do not currently possess a dedicated capture card. Please keep in mind that the captured footage from the 5060 Ti might appear sharper in certain scenes due to the superior hardware encoding capabilities of that card. All games were tested at 1440p with performance mode upscaling enabled, which corresponds to a 720p internal resolution.
Performance Insights
Preset M can be quite taxing on pre-Lovelace generation hardware as it uses a native FP8 model, necessitating conversions to a different precision format for inference on older GPUs. For the 5060 Ti, the overhead is comparable to that of standard Transformer presets. FSR 4 exhibits similar performance characteristics. Given that both the RX 9060XT and 5060 Ti possess similar raw rasterization performance, neither GPU is unduly favored by a higher FPS, providing a more accurate comparison for the upscaling models.
Cyberpunk 2077: A Comparative Analysis
When testing Cyberpunk 2077, I found both upscaling technologies produced comparable overall image quality. While performance mode upscaling at 1440p pales in contrast to native quality, both FSR and DLSS managed to deliver a usable image, albeit with some artifacts.
View footage here
Elements such as neon lights and textured walls exhibited noticeable shimmering, though edge aliasing was largely controlled. Notably, DLSS demonstrated superior handling of disocclusion artifacts, particularly surrounding foliage. While DLSS 4.5 made these artifacts less noticeable, FSR 4 rendered them more apparent even during quick movements. Consequently, I would categorize this matchup as a tie overall.
View additional footage here
Hogwarts Legacy: Upscaling Performance
Shifting focus to Hogwarts Legacy, Nvidia’s DLSS solution emerges as the clear winner. Its clarity exceeds that of FSR due to the more aggressive sharpening applied by the M preset. While both upscalers handle ghosting well, FSR struggles with foliage, where panning the camera—even slightly—introduces distracting background noise that takes a moment to settle.
View gameplay here
This issue has been persistent for FSR 4 in various games, indicating a less adaptable upscaler. While DLSS also has difficulty with moving foliage, it is significantly less pronounced compared to FSR. Overall, DLSS 4.5 provides a better experience in this title.
View further gameplay here
Chiaroscuro: Expedition 33 Evaluation
Next in line is Clair Obscur: Expedition 33, which has historically suffered from poor native visuals due to extensive processing effects that blur the image. Hence, the quality of the upscaler is critical for this game.
View more footage here
Here, FSR 4 yields a somewhat softer image, while the M preset’s heavy sharpening leads to a near-native visual quality, considering that the native representation was already quite poor. Both upscalers show no substantial ghosting, shimmering, or jaggies, aside from visible reflections that result from Lumen’s low resolution.
View additional gameplay here
Despite both facing the same challenges related to native image quality, including issues like disocclusion artifacts and shimmering around textures, the additional sharpening from Preset M provides a slight edge in this context, leading to a close win for DLSS.
Call of Duty: Black Ops 7 Insights
Testing Call of Duty: Black Ops 7 presented challenges due to its fast-paced nature, requiring frame-by-frame analysis to discern quality differences in motion clarity. Ultimately, my findings suggested the two upscalers are closely matched in this context, as both solutions effectively maintain sharpness and motion clarity despite operating at a low internal resolution of 720p.
Watch my frame-by-frame analysis here
Footage confirmed that both upscaling technologies preserve impressive levels of clarity and motion during demanding gameplay sequences.
Explore further analysis here
Forza Motorsport: An Experimental Approach
Forza Motorsport served as an experimental case since it does not natively support FSR 4. I employed Optiscaler to leverage FSR 4 through the game’s DLSS inputs. This method has previously worked well while playing at 1440p with FSR 4 Quality using a different GPU.
View my experiment here
Surprisingly, using 1440p FSR 4 Performance mode resulted in a notable decline in image quality. This was unexpected, as FSR 4 generally excels even in Performance mode. Anomalies such as ghosting from taillights of cars were present, which is atypical for FSR 4. Overall, this experience highlighted the significance of official support for upscaling technologies, as community-driven solutions can lead to inconsistent results.
Watch more on this experiment here
It’s essential to acknowledge the results of this title are somewhat subjective, as the FSR 4 version implemented was not officially supported.
Conclusion: DLSS 4.5 vs. FSR 4
After evaluating five games using 1440p Performance upscaling with both the DLSS 4.5 M preset and FSR 4, it is evident that DLSS offers a slight edge in many scenarios. The aggressive sharpening employed by Preset M enhances games with softer native visuals, giving DLSS an advantage.
Still, it’s important to note that DLSS is not a comprehensive solution across all games; image quality remains generally comparable to FSR 4. A significant factor favoring DLSS remains its robust game support. Although tools like Optiscaler allow FSR 4 integration in titles lacking native support, as evidenced by the Forza Motorsport experiment, the experience can vary widely and may not meet expectations.
Leave a Reply