Written by Henry Lin and Andrew Burnes on January 6, 2025 | Featured Stories
CES 2025: Frame Generation, GeForce RTX 50 Series, and DLSS Advancements
NVIDIA DLSS (Deep Learning Super Sampling) is a suite of neural rendering technologies that harness the power of GeForce RTX Tensor Cores to enhance frame rates while maintaining crisp, high-quality visuals comparable to native resolution rendering. It is featured in over 700 RTX games and applications.
At CES 2025, we are taking DLSS to the next level with the launch of DLSS 4, which introduces Multi Frame Generation for the GeForce RTX 50 Series graphics cards and laptops. This feature will be supported by 75 games and apps upon release.
DLSS Multi Frame Generation can produce up to three additional frames for every traditionally rendered frame, working seamlessly with the full suite of DLSS technologies to boost frame rates by up to 8X compared to traditional rendering methods. This dramatic performance improvement on the GeForce RTX 5090 graphics card unlocks stunning 4K 240 FPS fully ray-traced gaming.
NVIDIA's Bryan Catanzaro and Edward Liu Discuss DLSS 4 Features
With the launch of the GeForce RTX 50 Series, NVIDIA app users will have the opportunity to enhance games and apps by upgrading them to utilize the new DLSS 4 features.
A total of 75 DLSS-enabled games and apps, which already feature Frame Generation, will be upgraded to support Multi Frame Generation on GeForce RTX 50 Series GPUs.
For the same games, Frame Generation will see improvements for both GeForce RTX 50 and 40 Series GPUs, offering better performance while reducing VRAM usage.
DLSS Multi Frame Generation: Boost Performance on GeForce RTX 50 Series GPUs
DLSS 3 Frame Generation uses game data such as motion vectors, depth, and an optical flow field from the GeForce RTX 40 Series Optical Flow Accelerator to create a single additional frame. Generating multiple frames was previously inefficient because both the Optical Flow Accelerator and AI model would need to be applied to every new generated frame, which would strain the GPU and lower input frame rates.
The new AI model for frame generation is 40% faster, uses 30% less VRAM, and only requires one pass per rendered frame to generate multiple frames. For example, in Warhammer 40,000: Darktide, this model resulted in a 10% faster frame rate and saved 400MB of memory at 4K, max settings with DLSS Frame Generation.
Additionally, we’ve optimized the optical flow field generation by replacing hardware-based optical flow with a highly efficient AI model. These innovations reduce the computational cost of generating extra frames. Despite these efficiencies, the GPU still needs to run five AI models—covering Super Resolution, Ray Reconstruction, and Multi Frame Generation—for each rendered frame, all within a few milliseconds. Without this optimization, DLSS Multi Frame Generation could slow down performance. To handle this, the GeForce RTX 50 Series includes 5th Generation Tensor Cores, offering up to 2.5 times more AI processing performance.
Once the new frames are generated, they are paced evenly to ensure a smooth experience. Previous versions of DLSS 3 Frame Generation used CPU-based pacing, which introduced variability in the pacing that worsened with additional frames, causing less consistency and smoother gameplay.
To address these challenges, Blackwell introduces hardware Flip Metering, moving the frame pacing logic to the display engine for more precise control over display timing. The Blackwell display engine is also enhanced with double the pixel processing power to support higher resolutions and refresh rates for hardware Flip Metering with DLSS 4.
Together, these hardware and software advancements allow DLSS 4 to generate 15 out of every 16 pixels, delivering high image quality, smoothness, and low latency.
0 Comments