DLSS 3.0: What is it? How does it work?DLSS 3.0 is the new version of NVIDIA technology for the new batch of high-performance graphics cards that are going to be launched on the market. This new revision is a huge improvement over the DLSS 2.0 version and will delight lucky gamers with a GeForce RTX 4000 Series . But… will it be able to compete with the new version of AMD FSR ?
The combination of GeForce RTX GPUs’ Tensor Cores, AI-enabled super resolution, and DLSS technology have revolutionized graphics by increasing FPS and delivering crisp, high-quality visuals that are ever more realistic. Since the adoption of DLSS, 216 video games and applications have been able to increase frame rates and provide the performance headroom needed to make real-time ray tracing a reality in gaming.
Index of contents
- What is DLSS 3.0?
- How does it work?
- Results obtained and differences with DLSS 2.0
What is DLSS 3.0?
The release of NVIDIA DLSS 3.0 Neural Graphics , a new cutting-edge technology, has now been announced. Using DLSS Super Resolution, DLSS Frame Generation, and NVIDIA Reflex, a new technology running on GeForce RTX 40 GPUs, DLSS 3 improves performance by up to four times over brute force rendering. In less than a month, more than 35 games and applications will use DLSS 3 technology, the first of which will be released.
In short, a technology to improve the sharpness and quality of the graphics of your favorite video games, among others, and that uses the potential of artificial neural networks to operate on the image. And all without sacrificing FPS speed or responsiveness.
How does it work?
With DLSS 3.0, you can get the performance and image quality of DLSS Super Resolution without sacrificing speed or responsiveness. DLSS 3 creates new frames using DLSS Optical Multi Frame Generation technology and low latency NVIDIA Reflex . Tensor cores and optical flow accelerators from the new NVIDIA Ada Lovelace architecture power GeForce RTX 40 series graphics cards.
Ada Lovelace is a fourth-generation convolutional autoencoder that creates frames using game engine data such as motion vectors and depth. It creates these frames by encoding 4 inputs: a current and previous game image, an optical flow field generated by the optical flow accelerator, and an optical flow field generated by Ada Lovelace.
Ada ‘s Optical Flow Accelerator measures the optical flow field between two game frames to capture the direction and speed at which pixels move from frame to frame. Optical Flow Accelerator is capable of detecting pixel-level information such as particles, reflections, shadows, and lighting that are not measured by game engines, as well as motion flow. The biker’s movement in the example below accurately represents that the shadow stays at roughly the same point on the screen relative to his bike.
Using the game engine’s motion vectors , DLSS 3 accurately tracks pixel-level effects like reflections, as well as important visual details like the position of geometry in the scene. The road, for example, moves through the scene in a way that matches the biker’s path, but the shadows do not. Stuttering shadows would be an example of the visual anomalies that could occur if frames were generated using only the motion vectors of the engine.
The DLSS Frame Generation AI network creates intermediate frames using game motion vectors, optical flow fields, and sequential game frames, in addition to engine motion vectors, to generate the image. As shown in the image, the DLSS Frame Generation network is capable of accurately recreating both geometry and effects using optical flow to track motion, in addition to motor motion vectors.
With DLSS Super Resolution enabled, the first frame is reconstructed with seven-eighths of its pixels , and the entire second frame is reconstructed using DLSS Frame Generation. By reconstructing seven-eighths of the total displayed pixels, DLSS 3 greatly improves performance. NVIDIA Reflex is also included, which synchronizes the GPU and CPU and ensures optimal responsiveness and low system latency. Compared to the native system, DLSS 3 can reduce latency up to 2 times. Improved responsiveness of in-game controls and near-instant on-screen actions are some of the benefits.
Results obtained and differences with DLSS 2.0
NVIDIA DLSS was originally launched as an AI-based real-time super-resolution rendering technique that produces sharper, higher-quality images by reducing the number of pixels and then applying AI to them. Later, NVIDIA DLSS 2.0 would be announced , which improves image quality and performance using a general neural network that can be used with all games and scenes without prior training. DLSS 2 is already supported by 216 games and apps, as well as Unity and Unreal Engine. NVIDIA’s AI supercomputer continues to improve DLSS 2 through continuous training, and so far four updates have been released that have improved image quality.
When the game’s CPU is throttled but the graphics card is not, DLSS framing works as a post-processing step that allows the AI network to improve the frame rate. With DLSS 3-enabled CPUs like the GeForce RTX 40-series, CPU-constrained games like physics-intensive games or large worlds can be rendered at up to twice the frame rate of what the CPU is able to calculate. In Microsoft Flight Simulator, for example, DLSS 3 increases FPS up to 2x by recreating our planet in real life.
By employing DLSS 3, GeForce RTX 40 series graphics cards can increase performance by up to 4x when running games on specific engines. Reflex software algorithms are supported by dedicated Tensor Core and Optical Flow hardware, as well as NVIDIA’s supercomputer, which continuously trains and improves AI networks. This stack offers full performance, advanced AI networking, and Reflex software algorithms, along with fast frame rates, fast responsiveness, and high-quality images.
- RTX 4090 24GB VRAM
- RTX 4080 16GB and 12GB VRAM
Now we have to wait until October, when the first NVIDIA GeForce RTX 40 graphics cards with DLSS 3.0 will be launched…
