We reached out to Nvidia not too long ago with DLSS 3.5 on our minds. The release of the Cyberpunk 2077 2.0 patch ushered in the latest update to Team Green’s AI upscaling software, and in the run-up to its launch we were excited to find out a little more about how it’s going to change gaming hardware in the future.
The latest version of Deep Learning Super Sampling (DLSS) works on a host of new graphics cards, extending compatibility beyond DLSS 3’s limitations to the RTX 40-series. With extended compatibility, features, and performance comes a host of questions – so here they are, alongside Nvidia’s anticipated answers.
A specific curiosity we were looking to itch was how the development of hardware may be reaching the point of diminishing returns. While it’s true that each generation of Nvidia graphics cards has technically improved on the last, this has also been paired with significant increases in price. Critical receptions, while all pointing to improvements in performance, have also all picked up on the rising costs of graphical hardware. Thankfully, Nvidia may have touched on this in their responses – telling us that ‘the best gaming experience’ is provided by more than just the hardware.
“With the slowing of Moore’s Law, we’re now at a point where the GPU has to offer a combination of features across hardware and software to provide the best gaming experience. We believe this is the RTX ecosystem – Providing GeForce RTX hardware, DLSS, Ray Tracing, Path Tracing, Game Ready Drivers, NVIDIA Reflex, GeForce Experience and more.”
Moore’s Law, perhaps now irrelevant in the graphics card discourse, has been replaced by something slightly more topical: Huang’s Law, which plays into Nvidia’s comments on DLSS’s place in the future of gaming. While Moore’s Law might be slowing, this is supplemented by growth in other areas – AI especially. This is the essence of Huang’s Law, which generally suggests that the growth of graphics components is vastly outpacing the growth of computing components, mostly thanks to advancements in AI. DLSS is just one example of that.
“NVIDIA will be using generative AI more and more for the graphics process. DLSS paves the way, by reconstructing more pixels with Super Resolution, generating more pixels with Frame Generation, and making ray tracing look even better with Ray Reconstruction. 79% of 40-series gamers turn on DLSS, a testament to the technology’s benefits.”
We asked them how technologies such as frame generation will change the future of what Nvidia GPUs could look like.
“Frame generation is available today with DLSS 3.0, which uses AI-powered graphics to massively boost performance, while maintaining great image quality and responsiveness.
AI is also improving experiences within games themselves. At Computex this year we announced the future of NPCs with the NVIDIA Avatar Cloud Engine (ACE) for Games. NVIDIA ACE for Games is a custom AI model foundry service that aims to transform games by bringing intelligence to non-playable characters (NPCs) through AI-powered natural language interactions.”
At the moment, there’s not a whole lot of games that support DLSS 3.5. Given that it’s a very new feature, can we expect a mass adoption of games supporting it soon?
“Developers are excited about the technology and the next title to support DLSS 3.5 will be Alan Wake 2, releasing on October 27th. While our current focus is on intensive ray-traced titles, we do plan to expand to light-use ray-traced titles in the future.”
Introducing DLSS 3.5 compatibility to ‘light-use ray-traced’ games would offer a much lower entry point for people to try the next generation of graphics. At the moment, the reality is that you’re going to need a top tier graphics card in order to fully enjoy ray tracing. Add path tracing into the mix, and it gets a whole lot tougher.
Following on from Starfield’s launch, it was clear that players want DLSS in their games. Developers listening, and DLSS is coming for Starfield at some point – however, we were curious if DLSS 3.5 is going to be limited to Nvidia sponsored games moving forward?
“We provide the support and tools for all game developers to easily integrate DLSS if they choose and even created NVIDIA Streamline to make it easier for game developers to add competitive technologies to their games. DLSS 2 and 3 are available in over 300+ titles with many more to come. We are committed to providing this technology to the industry.”
With the bulk of these AI image upscaling techniques being first generated from a server side technique, we wanted to know if there was an opportunity to create something similar to be utilized by non-RTX cards.
“While the NVIDIA Supercomputer trains the AI models for neural rendering, you will need a GPU that’s capable of inferencing and executing these models in real-time. NVIDIA DLSS has the benefit of being powered by Tensor-Cores on the GPU. The GeForce RTX 40 Series introduced fourth-gen Tensor Cores as well as the Optical Flow Accelerator which powers DLSS Frame Generation.”
And finally, we wanted to understand why DLSS 3.5 and DLSS 3 have the differences they do.
“NVIDIA DLSS 3.5 features Ray Reconstruction, a new AI model that creates higher quality ray-traced images for intensive ray-traced games and apps.
“DLSS Ray Reconstruction adds additional AI to the ray-tracing lighting pipeline by replacing multiple hand-tuned denoisers and adding a combined AI model for Super Resolution and Ray Reconstruction, addressing image quality challenges associated with a denoiser and high frequency information loss during upscaling.
Much like DLSS 3, DLSS 3.5 is a suite of AI rendering technologies powered by Tensor Cores on GeForce RTX GPUs for faster frame rates, better image quality, and great responsiveness. DLSS 3.5 includes Super Resolution & DLAA (available for all RTX GPUs), Frame Generation (RTX 40 Series GPUs), and Ray Reconstruction (available for all RTX GPUs).”
A couple of months ago, artificial intelligence seemed like this flash in the pan trend closer to a dystopian fiction than the future of gaming hardware. Not for Nvidia, it seems. We’ve known for a while that AI has been a big part of Nvidia’s future – in fact, it’s more likely that Team Green are an AI company over a graphics hardware company now. Their processing chips are used by global AI superpowers – OpenAI a notable partner, with other companies such as Meta and Google trying to wrestle a spot alongside Jensen Huang’s company.
We’ve also known that AI is going to be a big part of the future of gaming. Nvidia ACE for Games was announced earlier on this year, and DLSS has been around since 2019. It has paved the way for Nvidia to be “using generative AI more and more for the graphics process,” a pretty significant statement, might I add. It seems as though Nvidia’s trajectory will be headed towards developing AI graphical processes over creating more powerful hardware on its own. This isn’t necessarily an issue for the consumer in-terms of performance, though in the long-run it could change things. The question is whether AMD are going to be able to follow Nvidia down their artificial intelligence path, and if they’re not, this could lead to big changes in the market.
