What's the difference between DLSS 3 and DLSS 2?

Nvidia answered players' confusion about DLSS 3 in the latest blog post. According to NVIDIA, DLSS 3 includes 3 technologies: DLSS Frame Generation, DLSS Super Resolution (also known as DLSS 2), and NVIDIA Reflex. Among them, DLSS frame generation requires hardware support from the latest RTX 40 graphics card, and DLSS 2 and NVIDIA Reflex current graphics cards can be supported.

According to Nvidia, the DLSS frame generation function uses a high-speed RTX 40-series optical flow accelerator to calculate the motion flow for the AI ​​network and then executes the network on the 4th generation Tensor Core. Further innovation and optimization of optical flow algorithms and AI models is required to support previous GPU architectures.

Of course, existing hardware will still support DLSS Super Resolution and NVIDIA Reflex, so current GeForce gamers and creators alike will benefit from games that integrate DLSS 3. NVIDIA will continue to research and train AI for DLSS super-resolution and will provide model updates to all RTX customers as it has since DLSS was first released.

Q: DLSS 3.0 looks great and impressive on a technical level. Is an engine-level update required to experience the improvements in DLSS 3.0 over 2.0? Or, can DLSS 3.0 be easily implemented in a game that already supports DLSS 2.0 without a lot of development work?

A lighter and simpler way of integration has always been the design goal of DLSS 3. With over 35 games and apps coming soon, DLSS 3 has become one of Nvidia's fastest-growing technologies. The first games to support DLSS 3 will launch in October.

DLSS 3 leverages the same integration points as DLSS 2 and NVIDIA Reflex, making it easy to upgrade existing SDKs through the DLSS 3 Streamline plugin.

DLSS 3 is also coming to the world's most popular game engines, including Unity, Unreal Engine, and Frostbite, making it easy for games based on these engines to switch to DLSS 3.

Q: How does optical flow fit into the model? If DLSS 2 is a spatial reconstruction of the next frame, does that mean we do a temporal reconstruction of the previous frames? Also, can it achieve sub-25% pixel rendering (DLSS performance mode)?

There are two AI models in DLSS 3: DLSS Super Resolution (also known as DLSS 2) and DLSS Frame Generation. DLSS super-resolution increases frame rates by rendering fewer pixels and then using AI to build sharper images at higher resolutions. DLSS frame generation generates additional high-quality frames by analyzing sequential frame and motion data from the Optical Flow Accelerator in GeForce RTX 40-series GPUs, improving performance while maintaining excellent image quality and responsiveness. When DLSS 3 is enabled, the first frame is first reconstructed by DLSS super-resolution, and subsequent frames are reconstructed by DLSS frame generation. Overall, DLSS 3 is able to reconstruct 7/8 of the total display pixels.

Post a Comment