Following the many GeForce Beyond hardware and software announcements, some NVIDIA engineers took to Twitter to discuss them with the community at large.
DLSS 3 has been a labor of love in Applied Deep Learning Research since the group was founded. I can’t wait for people to play with it.
In the same Twitter thread, he then explained why the technology will be exclusive to the upcoming NVIDIA GeForce RTX 4000 series.
DLSS 3 relies on the optical flow accelerator, which has been significantly improved in Ada over Ampere - it’s both faster and higher quality.
The OFA has existed in GPUs since Turing. However, it is significantly faster and higher quality in Ada, and we rely on it for DLSS3. [RTX 2000 and 3000] customers would feel that DLSS 3 is laggy, has bad image quality, and doesn’t boost FPS.
That said, Catanzaro left a door open to NVIDIA DLSS 3 potentially becoming compatible with GeForce RTX 2000 and 3000 series in the future, although he stressed it wouldn't yield the same benefits seen with the new graphics cards. As a reminder, DLSS 3 games will still provide DLSS 2 + Reflex support for GeForce RTX 2000 and 3000 owners.
It’s theoretically possible that with additional research and engineering that we could get this technology working on other cards, although it wouldn’t provide as much benefit. The current version only works on 4000-series cards.
Catanzaro also commented on the potential latency issue potentially introduced by the new DLSS 3 method. As we had surmised, NVIDIA plans to circumvent the problem by bundling their Reflex anti-system latency technology.
NVIDIA Reflex removes significant latency from the game rendering pipeline by removing the render queue and more tightly synchronizing the CPU and GPU. The combination of NVIDIA Reflex and DLSS3 provides much faster FPS at about the same system latency.
Bryan Catanzaro was even kind enough to check with another NVIDIA team on the status of the RTX I/O project, which we haven't heard about for a while. He relayed that people interested in RTX I/O should stay tuned since 'cool stuff has happened' and NVIDIA is looking forward to sharing more with the public.
Catanzaro wasn't the only NVIDIA engineer to share some info after the GeForce Beyond presentation. Alexey Panteleev (who previously worked on projects such as Quake2RTX, RTXDI, and Portal RTX) also engaged with Twitter to explain a bit of how the RTX Remix tool is going to work.
The Remix runtime is a pretty complicated "reverse game engine" that finds objects in draw calls, which allows us to derive motion vectors. The runtime doesn't make any artistic decisions, it works with the original and replacement assets that it's given. Modders can choose to use AI tools for like material upscaling or do everything by hand.
Particles and decals "just" work. Some DX hacks need special processing or should be disabled.
We'll have more information from NVIDIA regarding DLSS 3, RTX Remix and more soon; stay tuned.