So I was thinking about DLSS since it now supports VR. The possibilities of it are truly awesome. Especially when you use it with the 8k X. You could use DLSS to go from 1080p to 1440p. You’d be maxing out fps in upscale mode in a ton of scenarios!
The only bottleneck would be cpu. Same applies when you use it to go from 1440p to 4k! It would look nearly identical to native resolution, and you’d be almost always maxing out fps assuming you have a good rig.
Hmm, I wonder what might happen if you use DLSS 2.1 in combination with the dynamic foveated rendering. Would it smooth out the pixels and thus reduce shimmering? If it does, that might make foveated rendering much more appealing. It’ll be very interesting to see the results when it gets implemented into games.
Pretty sure DLSS 2.1 works generically and does not require a super computer to train the ai per game. Though the dev still has to implement it.
(Edit: Yup, no need to train per game. https://developer.nvidia.com/dlss )
I mean, to be fair this would help a large portion of users. It’s not like a super niche thing where you need eye tracking. You just need a 30 (or 20 probably) series gpu. So it makes sense that devs would implement it.
Sure I agree it would help out a ton of people but I just don’t see devs who have already abandoned their games coming back to help us out any. There’s tons of games out there that just need little tweaks here and there to become great but devs refuse to do any more work on them. Alot of VR games were quick one offs that will never be touched again by the developers. VR ports are even worse then this, we see devs being almost aggressive towards the VR community. I doubt they, who can barely do the bare minimum of support, will even care about this.
Games like SkyrimVR and FalloutVR will literally never be updated again, thats a fact.
Yes it need train per game its how DLSS works. Developer need train neural network to make DLSS working correctly. In last DLSS version NVIDIA making it easier for developer. But u still cant turn on DLSS for any game only for game support it.
theoretically, a tool is possible that would create a neural network for any game, on the user’s side. But I do not think that NVidia will bother with this. It is too difficult to implement.
In the past, yes it was required. However the new DLSS does not need to be trained on a per game basis. It’s pretty great. DLSS seems to still be in a beta of sorts, so maybe once they fully release it’ll be easier for devs to implement.
All Games On One Network: The original DLSS required training the AI network for each new game. DLSS 2.1 is different, in that it trains using non-game-specific content, distributing a generalised network that works across DLSS supported games. This means faster game response times, and ultimately the launch of more DLSS games in the future.
Then the words “DLSS supported games” seems to indicate some limitation maybe no training per games but specific call in the code for supporting DLSS
It’s on an engine limitation level, anything made in Unity, Unreal, CryEngine, Frostbyte or the other major AAA in-house game engines will be supporting DLSS. NVidia offers incentives to dev studios to support their hardware specific features and optimisations (speaking from personal experience).
On a driver level NVidia are years ahead of AMD, their direct support to dev studios is pretty good despite their reputation of being assholes to deal with.