Today, Nvidia announced the release of a successor to the company’s anti-aliasing, sharpening, and neural-network learning tool, Deep Learning Super-Sampling (DLSS), dubbed DLSS 2.0. The update contains huge overhauls to the fundamental aspects of how DLSS functions, and it promises major leaps in performance that were previously unattainable due to the constraints of the old technology.
DLSS 2.0 features new pipelines, new learning techniques, new approaches, and a whole lot of new math. Nvidia has focused a huge amount of resources on DLSS over the past few months, and the company seems to have learned from the pain points that many users felt in the first generation.
What Does DLSS Do?
This 2.0 version of DLSS, in theory, will finally back up the two big original promises of DLSS: (1) help a large number of PC games run at higher frame rates in the same resolutions you’re used to, or (2) give you the option to increase in-game graphical quality while retaining the same frame rates without changing anything about your hardware. (This is granted you already have an Nvidia GeForce RTX card installed; employing DLSS relies on these cards’ tensor cores.)
In theory, the combination of these two benefits at once could mean better graphical fidelity at higher speeds, something of a Holy Grail in the world of PC gaming. Consider what the developers of MechWarrior 5 have to say. “Nvidia DLSS 2.0 basically gives our players a free performance boost, without sacrificing image quality,” says Russ Bullock, president at Piranha Games. “It was also super easy to implement with Nvidia’s new SDK, so it was a no-brainer for us to add it to MechWarrior 5.”
Not only does the DLSS 2.0 update promise better speeds, but the quality of the DLSS render itself is also said to have gotten an overhaul, thanks to an improved architecture that’s been integrated into the neural net algorithm, and it will be better equipped to handle situations that used to trip up DLSS 1.0.
In the example above, you can see where a fan behind a grate of metal might create too many lines for the original algorithm to render properly. DLSS 2.0 has a refined learning network that can spot these problems for what they are, and approach them with new math that helps to determine where certain objects end and others begin, resulting in less artifacting and crisper graphics overall.
Nvidia also promised a slider in DLSS 2.0, allowing users to choose among three performance profiles: Performance, Optimal, and Quality. These profiles can be customized to suit each game you’re playing and the hardware you’re playing it on.
What Does DLSS Run On?
As it was at the launch of DLSS 1.0 (and continues to be today), the DLSS-supporting game library is still somewhat limited with the release of DLSS 2.0. But according to Nvidia, the new learning model will make it much easier, by orders of magnitude, for developers to submit their games to the system.
Plus, with Nvidia’s announcement of its partnership with Unreal to build DLSS tools directly into a new Unreal Engine 4 branch, the number of developers that could build their games from the ground up with ray tracing and DLSS optimizations baked in just got a lot larger in an instant.
You can try DLSS 2.0 for yourself today in either Wolfenstein: Youngblood or Deliver Us to the Moon, while the patches for Mechwarrior 5 and Control DLC will be made available later this week.
This is also when we’re planning to release a deep-dive on everything new that DLSS 2.0 has to offer. We intend to include both performance and quality testing of how DLSS 2.0 holds up when pitted against AMD’s competing Radeon Image Sharpening (RIS) technology, so stay tuned for that to see which performance-saving graphics software reigns supreme.