For the first time in many years, Nvidia refused to present new models of GeForce video cards at CES. The extended and under-rehearsed behavior of CEO Jensen Huang is typical. 90-minute talk focused almost entirely on the company's dominant artificial intelligence business, relegating gaming-related announcements to separate video published later in the evening.
Instead, the company focused on improving software for existing hardware. The biggest announcement in this direction is DLSS 4.5, which adds several new features to Nvidia's basket of upscaling and frame generation technologies.
DLSS scaling is improved with a new “second-generation transformer model” that Nvidia says is “trained on an expanded dataset” to improve predictions when generating new pixels. According to Nvidia's Brian Catanzaro, this is especially useful for image quality in Performance and Ultra Performance modes, where the upscaler has to make more assumptions because it's working with a lower-resolution source image.
DLSS multi-frame generation is also improved, with the number of AI-generated frames per rendered frame increasing from three to five. This new 6x mode for DLSS MFG is combined with what's called dynamic multi-frame generation, where the number of frames the AI ​​generates can dynamically change, increasing the number of frames generated during “demanding scenes” and decreasing the number of frames generated during simpler scenes, “so it only calculates what's needed.”
The standard caveats regarding multi-frame generation still apply: it still requires a 50-series RTX GPU (the 40-series can still only generate one frame for each frame rendered, and older cards can't generate additional frames at all), and the game still needs to run at a high enough base frame rate to minimize lag and weird rendering artifacts. It remains a useful tool for speeding up fast-paced games, but it won't help turn non-game framerates into playable ones.
