- AMD is rumored to be working on its own version of DLSS 5
- A leaker says that the neural rendering tech will be launched with next-gen RDNA 5 GPUs
- That means we could be waiting until late 2027 or 2028 to see this debut, which might leave AMD lagging behind Nvidia considerably
AMD could be working on its own take on Nvidia's DLSS 5, and apparently this alternative attempt at neural rendering will be for its next-gen RDNA 5 graphics cards.
VideoCardz spotted that a known leaker in the GPU field, Kepler L2, posted on the Anandtech forum in a discussion around RDNA 5 speculation to say that: "AMD will have their own version of 'Neural Lighting' like DLSS5 next-gen."
Kepler then clarified that he didn't mean a version of Ray Reconstruction, but a take on DLSS 5 (neural rendering), and that next-gen RDNA 5 GPUs would be required because "this stuff is pretty heavy" (as in resource-intensive).
In a further post, Kepler elaborated: "There's a lot of issues with Nvidia's approach [with DLSS 5], but it's possible to have a version of this with minimal hallucinations and 'AI slop' look."
He then suggests a way that AMD could work its version to improve on DLSS 5, including giving the AI "direct access to 3D game data such as geometry, ray data, depth buffers, materials, etc so it doesn't have to guess and can accurately know how each object in a scene should interact with light."
Kepler's outline of a proposed alternative for AMD would have a separate model (AI) for lighting improvements, and another different model solely for the shading of complex surfaces. DLSS 5 bundles these together with its neural rendering system, and adds color-grading on top — to adopt what the AI judges as more natural colors — which Kepler advises that AMD should dispense with entirely.
Analysis: a race against time – or a chance to get it right?
I won't pretend to be able to critique what Kepler is suggesting as an alternative for AMD and the technical nuances therein, except to say that the idea is clearly to achieve a result that looks less like it's been 'messed' about with by AI and minimizes any 'slopification' therein — preserving more of the original artistic intent.
And that seems a laudable goal, of course, and it also brings up how AMD is in an interesting position here in lagging behind Nvidia with its graphics tech (which has, of course, been the case from the start, with FSR following DLSS — version 2.0 of Nvidia's upscaling tech was already in play, in fact, when FSR 1.0 arrived). Team Red can sit back for now and see how DLSS 5 plays out, as more games emerge that use it, and gauge the reaction of gamers, and forge its own neural rendering tech to tackle any perceived weaknesses.
The performance hit will be an obvious area to target, and if AMD can bring that down — it's substantial with DLSS 5 as it is now — that will obviously be a big win. That said, I'd expect Nvidia will be working with some haste to make DLSS 5 more performant, and as with DLSS 1, the tech will be improved considerably by the time Team Red has its own version out.
This is the other side of the equation here, where Nvidia could win out. With RDNA 5 GPUs not expected to arrive for at least a year, and quite possibly longer (the RAM crisis isn't helping), there's quite a lengthy timeframe on any AMD rival to DLSS 5. By the time it arrives, DLSS 5.5 may be in town with considerable improvements in tow. There's already been a notable turnaround in the perception of DLSS 5 with its official launch (compared to Nvidia's sneak preview earlier this year).
This is, of course, assuming AMD is indeed working on its own neural rendering technology, but I'd be surprised if this wasn't the case. For now, though, the word of a single leaker needs to be thoroughly seasoned, as ever, even though it's believable enough that Team Red would be aiming to use this on RDNA 5 graphics cards.
from Latest from TechRadar https://ift.tt/8ZHC9mW
No comments:
Post a Comment