What is NVIDIA DLSS 3? | The Performance Multiplier
Introduction to DLSS 3
In this section, Bryan Catanzaro introduces DLSS 3 and explains how NVIDIA has been working hard to bring neural rendering to the next level.
DLSS 3 Components
- Making DLSS 3 required all of NVIDIA working together because it has hardware co-design with software in order to create this new capability.
- DLSS 3 has three main software components:
- DLSS Super Resolution
- DLSS Frame Generation
- NVIDIA Reflex
- DLSS 3 is also powered by three hardware components:
- Fourth-generation Tensor Cores
- Optical Flow Accelerator
- NVIDIA AI supercomputer
How DLSS 3 Works
- DLSS 3 can boost performance by an additional 2X through frame generation over DLSS 2 or 4X compared with brute-force rendering alone.
- NVIDIA Reflex synchronizes the CPU and GPU in order to keep FPS high while significantly lowering latency.
- The Optical Flow Accelerator analyzes two sequential frames and computes an optical flow field that describes how pixels in the first frame are moving in order to create images in the second frame.
- To solve this, we train the DLSS Frame Generation AI to make use of both optical flow and motion vectors in order to generate the highest quality intermediate frames.
Benefits of Using DLSS 3
- Between Super Resolution and Frame Generation, DLSS 3 uses neural rendering for seven out of every eight pixels on screen.
- DLSS 3 can also overcome CPU boundedness. GPUs are getting so fast that sometimes the FPS becomes limited by the speed of the CPU. But DLSS 3 can generate new frames independently of the CPU, which means that we can get up to 2X speedup even in CPU bounded situations.
- DLSS 3 further pushes the limits of real-time graphics with amazing FPS, low latency, and great image quality.
Conclusion
- With DLSS 3 neural graphics is really hitting its stride, enabling new frontiers of amazing ray-trace graphics and improving system efficiency.