Do we know they don't have anything similar? PS4 Pro had checkerboarding support in hardware; might PS5 have, I don't know, checkerboarding 2.0?
-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
Digital Foundry || Death Stranding PC DLSS 2.0 vs PS4 Pro Checkerboarding: Image Reconstruction Analysis
- Thread starter ILikeFeet
- Start date
Do we know they don't have anything similar? PS4 Pro had checkerboarding support in hardware; might PS5 have, I don't know, checkerboarding 2.0?
The way checkerboard was handled in Horizon (the game still looked great) is likely what was bugging me big time with the way whenever the grass would undulate or the camera was adjusted on the Y axis, it would cause some massive shimmering to happen. When I asked about it around launch, most said they didn't notice what I was talking about, but the problem is described in this video. Will be interesting to see how it ends up looking on PC in comparison.
Last edited:
OP
OP
there's no reason to believe they won't advance their upscaling solutions. they might pale in comparison to hardware solutions, but no reason to think their current stuff won't get better
I'm sure the studios have better up scaling techniques as well...just gotta wait for it.
Also, always a possible PS5 Pro
I think when people say better than native they are talking about how you can get more detail in some edge case scenarios. I remember seeing a comparison somewhere of wolfenstein 2 having window ledges visible far in the distance with dlss 2 but native missing parts of the ledges. I'm guessing its going to be very person specific though in a more general sense. I hate how soft modern aa techniques make stuff to the point I would suffer shimmering artifacts and turn off fxaa/smaa/taa.
There's also the motion clarity/temporal resolution argument that could be made but I don't think i've seen that really be what people are talking about when they say better than native.
Neither the PS5 or the XSX will be using something like DLSS 2.0 because it requires specialized machine learning hardware. Doing something like this in software is out of the question, I imagine.
Who knows what new techniques or improvements will be done to checkerboard rendering however.
Mostly agree if you're just doing a flat 1:1 comparison, but it's more than just a static image. DLSS allows you to get a near equivalent image, but also at a huge performance cost reduction, which also allows you to up other graphical enhancing effects if you choose, for a net improvement overall.
So basically, these next gen consoles don't have any form of machine learning right?
So hyperthetically, has pc absolutely leapfrogged next gen consoles before they have even launched?
I feel like this happens every time. We just need devs to utilise dlss in every game?
So hyperthetically, has pc absolutely leapfrogged next gen consoles before they have even launched?
I feel like this happens every time. We just need devs to utilise dlss in every game?
True, I did pick that up afterwards. Hopefully though it means that Sony and MS are working to improve on current methods. I'd be surprised if they weren't as it seems like it's going to be future of rendering.
Lol this intro was amazing, good video as well. About to experience DLSS for myself on Control.
In fairness to AMD, DLSS is a pretty new technology that requires dedicated hardware to use. And it's first implementation was pretty bad in DLSS 1.0. This is the sort of thing Nvidia is pioneering to sell their particular hardware. Equivalent solutions will probably end up on AMD hardware at some point, but the tech wasn't at the forefront when the silicon for these consoles was on development.
I don't disagree. The results are so close that it's not worth *not* using it.
And 4k is nearly always a waste of performance to begin with to me, with few exceptions like extremely high frequency details (and UI which should always be native).
DLSS is an amazing piece of tech which greatly improves the biggest issue with reconstruction techniques: how to appropriately reject bad samples from the history, while keeping good ones.
It's absolutely always worth having imo.
But in a direct comparison vs. native 4k I wouldn't claim it wins and produces a better image, at least from the pure rendering correctness point of view.
There may be software based solutions that come along, but nothing as far as we know hardware wise.
It makes you wonder the point of the 16series cards. Hoping I can get a 2060 used when y'all rich people decide to upgrade.
How do you feel about the temporal injection technique that Insomniac and Bluepoint tend to use, compared to checkerboard rendering? If it's that different at all from a technological perspective.
That backpack comparison was the scene that really helped me understand and see the difference. DF continues to deliver great content! Thanks Alex.
Yeah. This is where I am. I only understand things on a pretty high level, but I never pass on a new Digital Foundry video. They're always fascinating to watch.
This has been the case for a while sadly, AMD is pretty boring when it comes to features on PC.
That games gonna help nvidia sell some gpu's that's for sure. Wonder what the chances are nvidia will throw in a copy of it with the 3000 series cards.
Honestly, watching the Digital Foundry videos on DLS 2.0 have made getting a 3080 Ti a must now. Plus, since I'm still gaming at 1080p 144hz, I could probably put most Ray Tracing on at Max with DLS 2.0 and not have to worry about getting sub 60 fps.
Honestly, proprietary software solutions haven't been something I've appreciated in the past from any vendor; stuff like tress-fx or whatever the equivalent was just seems like a waste as those were often game specific and performance hogs. I shut all of that stuff off.
This particular innovation though presents a pretty clear and vision for improving performance and image quality for games. I hope this is the direction chipmakers think in the future.
It's definitely a better alternative imo.
Produces more pleasing results (its artifacts are more visually pleasing noise rather than the checkerboard pattern) and it's easier to "manage" and integrate at the engine level.
They are related tho, temporal supersampling/injection is essentially a generalised version of checkerboard rendering without the fixed pattern.
Same goes for DLSS, it's essentially temporal supersampling/injection with machine learning-based rejection test at its core.
Interesting that it's actually easier to integrate, I'd imagine the opposite, just because it's less common. I don't have the technical knowledge, but to my eyes it always produces better results as well. Shadow of the Colossus is particularly impressive, imo.
This is what I was wondering. Do you think now that insomniac is under sony other sony first party studios will use their tech over CB?
Pretty sure they will, but if i will get one is another question. People are so thirsty for the new GPU's it seems like (me included).
Really solid comparison!
DLSS 2.0 continue to impress.
It will be very interesting seeing what solutions win out among developers on next gen consoles.
DLSS 2.0 continue to impress.
It will be very interesting seeing what solutions win out among developers on next gen consoles.
I have a 1080p 50" screen and DS in "4K" mode looked like ass on Pro... almost more aliasing then 1080p native mode (not counting the slowdown like when using the mirror and such).
Days Gone is on another league for example in terms of IQ.
Days Gone is on another league for example in terms of IQ.
There are GPU brand fanboys? Wtf lolHad to google this one and this is the first thing that came up.
LMAO.
I'd be surprised to see anyone use CB in the next generation.
Pretty much everyone that used it at some point removed it in favour of temporal supersampling.
The last example of it was probably Death Stranding, but at a quick look it seems Horizon 2 is not using it, so it's probably gone from Guerilla's engine as well.
One crucial advantage of temporal supersampling vs. checkerboard rendering is that the latter relies at a fundamental level on the checkerboard structure to work and so the underlying resolution is fixed.
Temporal supersampling makes it easier to integrate a history of frame with different resolutions and therefore plays much better with dynamic resolution.
In modern games the rendering workload can vary dramatically in different scenes so limiting the entire game to a lower resolution because there's an especially intensive spot is not a worthy trade-off (and dropping frames is less than ideal as well).
I think saying that dynamic resolution + temporal supersampling (ML-based or not) + VRS are gonna play a huge role in future games on all platforms is a pretty safe bet.
INT-8 performance on the next-gen consoles is simply too low in comparison to the current Turing gaming GPUs, so even if they do manage to implement a software-based solution, it would not only take longer to process, but it would also take up rendering resources as they would be using the compute cores rather than dedicated hardware* (tensor cores, in the case of RTX).
*I'm not completely sure on that last part, though. I think I read somewhere before (may have even been on Resetera itself) that DLSS is processed at a certain point in the rendering pipeline--not parallel to the rest of the rendering pipeline. If that's the case, then a solution utilizing the compute cores would not take up rendering resources, but the cores themselves are not nearly as well-equipped to perform INT-8 operations as the tensor cores are, so it would still be slower than Nvidia's current solution.
We may see DLSS on the next switch, though.
It does allow you to do that, yeah. Gotta find a DLSS 2.0 game like you mentioned though.