Nvidia introduces DLSS 3: New upscaling technology and performance improvements spotted

Nvidia introduced its DLSS 3 technology at the Geforce Beyond special broadcast at GTC. The new temporal upscaling technology was demonstrated by the company's CEO, Jensen Huang. This iteration will bring huge advancements to the technology and will focus on even higher performance gains.

Some gameplay footage showcasing the performance improvements was also presented. These included Cyberpunk 2077 and Microsoft Flight Simulator, some of the most demanding titles on PC.

DLSS 3 is being introduced as a part of the Ada Lovelace lineup of GPUs, also called RTX 40 series. The new technology does not solely rely on AI upscaling and predicting the final image. Instead, DLSS 3 uses sequential frames to detect the changes in the output and render the game accordingly.

Nvidia has defined the upgraded upscaling technology to be "A Revolution in Neural Rendering."

---

How does DLSS 3 work?

Evolution of Deep Learning Super Sampling (DLSS) (Image via Nvidia)

Deep Learning Super Sampling (DLSS) was introduced alongside RTX 20 series GPUs back in 2018. The initial version of this technology would take a low-resolution image and predict a high-res version of it via the neural network. This output would then be used to train the AI in predicting better images.

The older technology did not prove to be effective in several demanding games. Thus, Nvidia had to come up with a better innovation that could push the boundaries of temporal upscaling to a whole new level.

Jensen Huang, CEO at Nvidia, said:

"DLSS 3 is one of our greatest neural rendering inventions."

DLSS 3 combines four technologies into one. These include a new optical flow accelerator, game engine motion vectors, a convolutional auto-encoder AI frame generator, and Nvidia's Reflex super-low latency pipeline.

Processing pipeline of DLSS 3 (Image via Nvidia)

Instead of computing high-resolution pixels, DLSS 3 has been designed to analyze the changes in successive frames. The technology registers the changes made in each scene. This can be achieved with the Optical Flow Accelerator.

This data, including the visual geometry of the scene and the pixel motion vectors, are then fed to the neural network. Then, the intermediate frames are generated.

Increase in ray tracing workloads (Image via Nvidia)

This technology is capable of producing completely new frames without interfering in the graphics processing pipeline unlike the previous two iterations of it. The third version of DLSS, according to Huang, can help in "boosting game performance by up to 4x over brute-force rendering."

This approach is for both GPU and CPU-limited games. If performance is limited by the CPU, the process can be divided with the Ada Lovelace GPU. This will allow the game to be rendered much faster than the CPU's capabilities.

---

Performance boosts with the all-new upscaling technology

Performance gains in Cyberpunk 2077 (Image via Nvidia)

Nvidia showcased the benefits of the Cyberpunk 2077 performance at the event. While the game was only running at 22-24 FPS at native resolution with ray tracing turned on, turning on DLSS 3 and SER bumps the framerate to over 90 FPS.

Performance gains in Microsoft Flight Simulator (Image via Nvidia)

In Microsoft Flight Simulator, viewers got to witness a similar performance bump. While the game was achieving around 50 frames at native resolution, the framerate jumped to 105+ FPS with the latest version of DLSS turned on.

Microsoft Flight Simulator is not a GPU-intensive title. As the game has complex physics and a vast open-world, it is limited by the CPU. DLSS manages to double the performance in this scenario.

The upcoming temporal upscaling technology will revolutionize the performance of AAA titles. However, since it will rely on Optical Flow Accelerators, it will only be limited to RTX 40 series GPUs. The older RTX 30 series and 20 series graphics cards cannot benefit from it.