WYP
News Guru
Does DLSS make Shadow of the Tomb Raider look better?
Read our Shadow of the Tomb Raider RTX Analysis.
Read our Shadow of the Tomb Raider RTX Analysis.
Last edited:
Shadow of the Tomb Raider RTX Performance Review - Ray Tracing and DLSS Analysed
- Thread starter WYP
- Start date
WYP
News Guru
ISeeWhatYouDidThere
WYP
News Guru
I can confirm, Lara's bottom looks just as good with DLSS on
Thanks for this dude. I love that slider thing, it's sweet !
Whenever I was putting the screenshots together I was wondering who would make a comment about Lara's bottom first.
Yeah, needed proper proof of how DLSS looks in this. You know what the internet is like. Most people will assume it is the same as in Metro Exodus ETC ETC
NeverBackDown
AMD Enthusiast
DLSS looks pretty good. Not blurry!!
Ray tracing on the other hand imo makes it look either the same or worse. Not much improved and certainly not enough for the performance hit.
Ray tracing on the other hand imo makes it look either the same or worse. Not much improved and certainly not enough for the performance hit.
WYP
News Guru
Yeah, I was very surprised. Had to double and triple check to make sure my screenshots were in the right order.
NeverBackDown
AMD Enthusiast
WYP
News Guru
I'll be honest I questioned if you did it right as well
That is basically why this wasn't finished on Wednesday. Extra time to be 100% sure. That and GDC becoming way more of a press event than ever before.
Can be scene dependent, but across all the areas I played DLSS looks better more often than not. Mostly scenes which show Lara's clothing up close look worse, such as the example in the screenshots shown.
When playing Tomb Raider, I look more at the environment and distant objects, that tend to look better with DLSS. The game is about exploration after all.
WYP
News Guru
I will soon be setting off for a weekend in the mountains, so not at the moment. Internet dead zone over there, so can't do anything until Monday or Sunday evening at the earliest.
As for evaluating DLSS i think it really makes sense to ask game developers and especially they graphics people what they think.
Anyone else may have their personal opinion - and for instance like that something may look sharper with DLSS on. But if that was not the intent of the developers then it becomes more like a 'loudness' button on your old hifi. You may like the result, but what is playing is now quite far from the artists intentions.
It may well be the other way around - that DLSS on come closer to what was intended.
Only way to know, is to ask the developers.
Anyone else may have their personal opinion - and for instance like that something may look sharper with DLSS on. But if that was not the intent of the developers then it becomes more like a 'loudness' button on your old hifi. You may like the result, but what is playing is now quite far from the artists intentions.
It may well be the other way around - that DLSS on come closer to what was intended.
Only way to know, is to ask the developers.
SPS
Moderator
It's because we're so used to seeing bad shadowing hacks to add some level of occlusion. Those soft shadows from those characters ooft.
You don't really have developer intent with DLSS, it's a neural network based algorithm with the training based on the game itself ran at an extreme resolution, with the desired input as the game at a normal resplution, there's no fine tweaking on the individual basis(while a developer could manually play with the training weights in the neural network this shouldn't be desirable or necessary), really you want the back propagation algorithm to work all the weights out itself from the input. The reason DLSS improves with time is because either
A) The algorithm has spent more time training (refining it's internal weights to produce the desired output from input)
B ) The algorithm was originally trained to work with a notably different input (different resolution, aspect ratio or even certain settings to the original training input could feasibly have an impact)
Neural networks are cool when they work right, you just feed in an already sorted or correct data set and let the weights adjust themselves till the network fires out the right outputs for the inputs you want, but the lack of ability to easily manipulate that algorithms functionality directly, particularly with deeper networks, without knowing and understanding how all the internal weights would impact the result, which is near impossible for stuff neural networks are best used on(very abstract wide data sets), makes them a real pain(or practically impossible) to try and "quickfix" manually or add functionality or intent in later(think accidentally racist neural networks as a result of oversights during training and stuff) so often they just get retrained which is a pretty long compute intensive process.
A) The algorithm has spent more time training (refining it's internal weights to produce the desired output from input)
B ) The algorithm was originally trained to work with a notably different input (different resolution, aspect ratio or even certain settings to the original training input could feasibly have an impact)
Neural networks are cool when they work right, you just feed in an already sorted or correct data set and let the weights adjust themselves till the network fires out the right outputs for the inputs you want, but the lack of ability to easily manipulate that algorithms functionality directly, particularly with deeper networks, without knowing and understanding how all the internal weights would impact the result, which is near impossible for stuff neural networks are best used on(very abstract wide data sets), makes them a real pain(or practically impossible) to try and "quickfix" manually or add functionality or intent in later(think accidentally racist neural networks as a result of oversights during training and stuff) so often they just get retrained which is a pretty long compute intensive process.
Last edited:
#14 if your reply was to me, then I maintain that the developers are the best people to judge the end result after DLSS. Everyone else is entitled to their opinion, but it remains only an opinion based on personal preferences.
Developers can say if the result is better, worse or about the same compared to traditional rendering. Only they know what they are aiming for - and even traditional rendering has it limitations so it is not necessarily the perfect picture that everything else should be compared to.
How it works is meaningless in evaluating the results (although interesting)
Developers can say if the result is better, worse or about the same compared to traditional rendering. Only they know what they are aiming for - and even traditional rendering has it limitations so it is not necessarily the perfect picture that everything else should be compared to.
How it works is meaningless in evaluating the results (although interesting)
Yeah I just mean, the game developers have no influence on the results of DLSS, the training process is done by NVidia on their own supercomputers, and the algorithms themselves have no developer influence, it's kinda a black box to the gaming devs, but yeah I agree they should have the ultimate opinion on how their artwork is meant to be viewed (Although I think ultimately beauty is in the eye of the beholder regardless of whether that beauty is an accurate portrayal of the art as intended).
NeverBackDown
AMD Enthusiast
Similar threads
