The video evaluates NVIDIA’s new DLSS Transformer model, preset K, comparing its performance and visual quality against older presets J and E, particularly at 4K resolution. While preset K offers improved image quality, it generally results in lower performance across various GPU generations, with minimal differences noted between newer and older cards, leading to a recommendation for using the performance version of preset K for the best experience.
The video discusses the latest updates to NVIDIA’s Deep Learning Super Sampling (DLSS), particularly focusing on the new Transformer model, preset K, which is compared to the older presets, including E and J. The presenter acknowledges that they initially overlooked the existence of preset K and emphasizes the need to test under dynamic conditions rather than static scenes to better assess the performance differences. They highlight the importance of user feedback and comments, which prompted further exploration into various aspects of DLSS, including VRAM usage and visual fidelity in different scenarios.
The comparison between presets K, J, and E is made with a focus on performance and quality at 4K resolution. The presenter mentions that while both J and K offer improvements over E, they struggle to identify significant differences between J and K in terms of visual clarity. They demonstrate the performance of these presets through gameplay footage, noting that the latest Transformer model (K) shows some advantages in terms of image quality but also some visual artifacts, particularly in fog and lighting effects.
Performance testing reveals that while preset K generally provides better image quality than E, it consistently yields lower performance across various GPUs, including the new 580 and older 3000 series cards. The presenter illustrates how using preset K can enhance the gaming experience, especially in demanding scenarios such as Quantum travel, where it allows for smoother frame rates even on lower-end hardware. They recommend using the performance version of preset K for optimal results.
The video also addresses concerns regarding the impact of different GPU generations on DLSS performance. Despite NVIDIA’s claims that the newer 5000 series cards would experience a smaller performance hit with the Transformer model, the presenter finds that the performance differences are minimal across all generations tested. They conclude that users with RTX 2000 series and higher should benefit from using preset K without significant drawbacks, regardless of their GPU model.
Finally, the video touches on issues with Vulkan performance following recent NVIDIA driver updates, indicating that problems may have arisen specifically due to changes made for the latest 5090 series cards. The presenter assures viewers that they will conduct further testing once the Vulkan issues are resolved. Overall, they express optimism about the Transformer model, suggesting that it represents a significant improvement over previous iterations of DLSS, particularly in high-demand scenarios, despite the noted trade-offs.