By using this site, you agree to our Privacy Policy and our Terms of Use. Close
Soundwave 12 hours ago
Edit Thread
Otter said:
Soundwave said:

I do wonder if a future version of DLSS5 (I really wish they had given this a unique name) if you'll be able to feed it data on what you want the final result to look like to iron out that problem. 

Like if the the algorithm could understand that you want Donkey Kong to look like this (which is art from your own artists). 

Particularly with the fur, then maybe it can understand and zero in on doing that up from a model that actually has a rendering cost of this:

Even this DK render is well below the Switch 2 capabilities. It's just a case of the team actually building an asset from the ground up for Switch 2 and utilising the right tech/rendering techniques. If Nintendo is not implementing DLSS2 at the moment to free up resources, I don't think there's much use in speculating their use of DLSS5 lol

DK Bananza was a Switch 1 game that was moved to Switch 2 (probably late into dev) so that explains a lot on that. 

Nintendo even showed the Switch 1 version here:

I think they will definitely use DLSS5/6/7 whatever by the time Switch 3 is out, people doubted the Switch 2 would use DLSS too in the early days and how'd that go. 

The fact that you can just offload most of the lighting compute to Tensor cores/NPU cores alone will be a huge benefit to a Switch 3. No point in bothering with path tracing/ray tracing really if you just just have that done on Tensor cores and not bog down the GPU with that. The lighting side of it will likely get better with future iterations too, this is just the first attempt. Right now it looks like the DLSS5 algo is trained to basically add perfect Hollywood style lighting to each scene (so a key light, fill light, back light, and I even notice an eye light too) and for some people that's jarring, they can likely adjust this for future revisions, though I think some people do like this look even as is. 

How they feel about actual generative AI over art assets themselves ... that could be a different story, I think they'll be initially cautious about using that but eventually probably will cave on that too. It really depends on what the broader industry does. They have a few years now to just kinda sit back and see how this whole thing shakes out, Switch 3 you're probably talking about a chip decision on that still be 3-4 years out from today (2030), by then Nvidia will be well past the 50 or even 60 series and you'll probably have DLSS7 or something that is a lot better than DLSS5. 

Last edited by Soundwave - 12 hours ago