NVIDIA’s DLSS 4.5 Update

A modern graphics card does not draw every frame you see from scratch. NVIDIA’s Deep Learning Super Sampling, short for Deep Learning Super Sampling, already renders fewer pixels than your monitor needs and then uses an artificial intelligence model to fill in the missing detail. (nvidia.com) Frame generation adds another shortcut on top of that. Instead of only guessing missing pixels inside one frame, it also guesses entire in-between frames, which is how NVIDIA says Deep Learning Super Sampling 4.5 can now reach a 6-times mode on supported apps. (nvidia.com) That 6-times number does not mean the game suddenly computes six real frames for every one it used to render. NVIDIA’s own description says Multi Frame Generation on GeForce RTX 50 Series hardware can generate up to five extra frames for each rendered frame using fifth-generation Tensor Cores, which are the chip blocks built for artificial intelligence math. (developer.nvidia.com) The new part in April 2026 is that the multiplier no longer has to stay fixed. NVIDIA says Dynamic Multi Frame Generation acts like an automatic transmission, shifting between frame multipliers during play so it can chase either a custom frame-rate target or your display’s refresh rate. (nvidia.com) That matters because a game scene is never equally hard to draw from one second to the next. A hallway with static walls needs less work than an explosion full of particles, so a fixed setting can overshoot in easy scenes and add too much delay in hard ones. (nvidia.com) NVIDIA is also updating the image-upscaling model underneath all of this. At CES 2026, the company said Deep Learning Super Sampling 4.5 Super Resolution uses a second-generation transformer model across all GeForce RTX graphics cards, with transformers being the same class of pattern-finding neural network now common in language and image systems. (nvidia.com) The rollout is partly inside games and partly inside NVIDIA’s own control software. NVIDIA says the full Deep Learning Super Sampling 4.5 feature set is now available through a new NVIDIA app beta, where users can turn on Dynamic mode globally or game by game in the Graphics tab. (nvidia.com) There are limits baked into the first release. NVIDIA says Dynamic mode is currently not compatible with frame-rate limiters or Vertical Sync, which is the display-timing feature that tries to stop screen tearing by matching frame output to monitor refresh. (nvidia.com) The other new feature sounds smaller, but anyone who has stared at a loading screen will recognize the problem it targets. NVIDIA’s Auto Shader Compilation beta rebuilds DirectX 12 shaders after a driver update while the system is idle, instead of forcing that work to happen the next time you launch a game. (nvidia.com) A shader is a tiny program that tells the graphics chip how to draw light, shadows, and materials, and compiling it is like translating a recipe into the exact machine instructions your card understands. NVIDIA says moving that translation into the background can cut load times after driver updates and reduce the stutter tied to on-demand shader work. (nvidia.com) The hardware split is important. NVIDIA says the new second-generation Super Resolution model works across GeForce RTX graphics cards, but Multi Frame Generation and Dynamic Multi Frame Generation are tied to GeForce RTX 50 Series products and RTX PRO Blackwell Generation GPUs. (nvidia.com, developer.nvidia.com) So this update is less about raw silicon getting faster than about software getting more aggressive with prediction. If NVIDIA’s models can convincingly invent more of what you see between the frames the chip actually renders, then a workstation, a game rig, or a real-time visualization app can look smoother without waiting for a new graphics card. (nvidia.com, nvidia.com)