Engadget Explains How NVIDIA DLSS Uses AI Upscaling to Boost Game Performance

When DLSS is enabled, the GPU first renders at a lower resolution before upscaling to whatever output resolution the user has set, typically the monitor's native resolution. A user with a 1440p monitor would set the output to 1440p in most games. DLSS then uses a machine-learning algorithm trained by NVIDIA to upsample the image. When implemented well, the technology allows higher graphical settings than a GPU could normally handle at native resolution while maintaining a playable frame rate, or it can push frame rates higher at lower settings, which can be useful for high-refresh-rate monitors.

The result involves a compromise. The machine-learning models behind DLSS are trained to interpret motion vectors, depth and brightness information during upscaling. NVIDIA's GPUs and models have improved over the years, but Engadget reported that some fidelity is lost whenever an image is upscaled. The amount depends on the DLSS mode, which sets the internal render resolution relative to the target output. In Quality mode, the GPU renders at 66.7 percent of the target output resolution: a 4K monitor starts with a 1440p image, and a 1440p monitor starts at 960p. Quality offers the best image quality among the presets but only a moderate frame-rate increase. Balanced mode renders at 58 percent of target output, translating to 1253p at 4K and 835p at 1440p, and is intended as a middle ground between image quality and higher frame rates. Performance mode renders at 50 percent of target output, or 720p internally for a 1440p monitor and 1080p internally for 4K output. Ultra Performance mode renders at 33.3 percent of output resolution and is intended for 8K monitors; Engadget said users should expect a blurry image if it is used with a 1440p or 4K monitor.

As a rule of thumb, Engadget said DLSS Quality is a good starting point for most games on a 1440p monitor, while Balanced is a good starting point for 4K. Users who need or want more frames can drop to Performance, though image quality can become quite blurry beyond that.

The DLSS branding covers several related technologies that can be used together. The initial version of DLSS arrived in 2019, following the release of NVIDIA's RTX 20 series GPUs. At launch, only a handful of games supported it, including Battlefield V and Metro Exodus, because NVIDIA had to train the algorithm specifically for each game it was meant to enhance. That limitation no longer exists after the release of DLSS 2 in 2020. According to NVIDIA, more than 1,000 games and apps offered DLSS support as of the start of 2025.

The biggest upgrade after 2020 came at the start of 2025, when NVIDIA reworked the algorithm to take advantage of a new transformer-based machine-learning architecture and released DLSS 4. Engadget reported that the new model offers better image quality, reduced ghosting and improved image stability. On older RTX GPUs, running DLSS 4 can result in worse performance relative to previous versions of DLSS because of higher overhead. For that reason, NVIDIA gives users access to different DLSS presets, separate from the modes.