Jump to content

DLSS Revisited - Hardware Unboxed


Recommended Posts

 

 

God damn. DLSS :bow: 

 

Some interesting tidbits in this video.  Hardware Unboxed have spoken with Nvidia about DLSS and this newest version, which shipped in Wolfenstein and Deliver Us the Moon, is the version which will be used and improved upon going forward. 

 

Hardware Unboxed are calling it DLSS 2.0 because it's a substantial improvement over the previous implementation which is removing a lot of the restrictions the previous implementation had.  They say Nvidia has told them that this new version is more general purposed and doesn't require per-game training..so devs aren't required to patch their games to implement improvements, they can simply release them by driver updates alone.  All the data they collect from all sorts of games can be used to improve the entire AI model to benefit any game that supports it in the future.  It also isn't dependent on resolution on specific hardware which means that you can use it across all resolutions on all RTX GPUs top to bottom.  This new version also runs faster on the tensor cores than the previous version, so performance improvements relative to native res has improved further as well.

 

Those are some huge improvements.  Of course, we'll have to see more games and more implementations to see how everything holds up, but with native or better than native IQ, and 30-40% improvement to performance (60% in Deliver Us the Moon) ... DL AI models are definitely going to be prominent in the future of graphics rendering.

 

Now this right here... could be very good news for Nintendo with a future Switch console... and considering the rumors that Nintendo is working on a Switch console with a Volta-based GPU, that would give them access to tensor cores and the ability to run DLSS.  Could be game changing for performance in a handheld. :whew: 

Link to post
Share on other sites
3 hours ago, Team 2019 said:

Newer version implemented in games is amazing tech and is here to stay. It will also get only Better.

 

 

Reconstruction techniques have come a LONG way since 2016 and are going to be the norm.

It is super impressive, and it's definitely going to get better.  You can already tell their plan is to make it as easy as possible for devs to implement, making it a no-brainer to support.. possibly even eventually be something that can simply be toggled on the driver and work for all games new and old which support TAA. (which let's face is.. is most new games) Also now that they have it to the point where it's general purpose, Nvidia themselves can update the AI model through drivers so that all games with the new DLSS 2.0 and better support will be improved due to the better training.

 

When Nvidia started with DLSS, it was just one guy... then 3, and now that people are seeing the potential, the team has grown considerably.  So while it definitely had a rough start, the people working on it knew the potential, they just needed time to get it all sorted.  Also, the 20-series only has so many tensor cores, so they have to work within a certain budget.  It's only going to get better with more tensor cores.

 

Also, on Twitter, members of the DLSS team at Nvidia said that this is just the beginning of what Deep Learning is going to do for graphics rendering.  They also mention that it could be used for things like animation interpolation, making animation frames themselves better and smoother.  If games happen to drop FPS under 60, they could potentially fill in those missing frames smoothing the game out and you wouldn't really even know.  This could be a huge deal in the future. 

 

Also, as I said before... it's going to have HUUUUUUGE implications for the actual development of games, more than anything else.  Devs using it to develop things at a rate they could never achieve without it... things from the artwork, to level design and so on.  Using DL to program better AI and physics models.

 

It's going to play such a huge part in the future.  Will be awesome to see where it leads.

  • Upvote 1
Link to post
Share on other sites
  • 1 month later...

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Loading...
  • Recently Browsing   0 members

    No registered users viewing this page.

×
×
  • Create New...