Nvidia RTX DLSS/Ray Tracing Discussion

[NIGELG]

Originally Posted by SIrPauly

Finally seen one of those trailing artifacts in Metro.


Imho,


Some may disagree, but to gauge quality, one must start at a lower resolution and work your way up. Also, not every one can afford the latest-and-greatest gaming hardware -- and considering 1080p is the most popular resolution -- and considering many are gaming on laptops -- can one have a good Rtx experience at a lower resolution?


This is the team that created Dlss 2.0 and there was some nice nuggets of data even for lower resolutions here:


http://behindthepixels.io/assets/files/DLSS2.0.pdf

I see them on extremely rare occasions but ultimately DLSS does the job on a 3070.

[Exposed]

Are we sure it's DLSS 2.2? From reading the reddit it looks like a driver hack.

[Exposed]

Well's that certainly unexpected and surprising. I would have figured Nvidia to make a bigger deal about DLSS 2.2

Sounds like DF will do an indepth analysis soon so looking forward to that.

[acroig]

Originally Posted by Exposed

Well's that certainly unexpected and surprising. I would have figured Nvidia to make a bigger deal about DLSS 2.2

Sounds like DF will do an indepth analysis soon so looking forward to that.

Seconded

[acroig]

Originally Posted by SIrPauly

Just changed the nvngx_dlss.dll in Metro Exodus Enhanced Edition to version 2.2 and that trailing artifact is basically gone.

Is there a link to that file?

[andino]

I wonder if this will work in Control.

[demo]

I read Cyberpunk performance is better and has far less artifacts - anyone tried it?

/edit, sorry Pauly, missed your initial post, seems it does indeed work but performance improvements have been exaggerated by some users.

[acroig]

Originally Posted by SIrPauly

Here you go:



https://drive.google.com/file/d/1Piw...ew?usp=sharing

TY.

[clerick]

Swapped dll to 2.2 in death stranding, fixed the effect and particle artifacts, also backpack appears sharper. Really works.

[acroig]

Why has nV not made a huge deal about 2.2?

[Exposed]

This is more mysterious than the time Bill dissed the 2080ti and bought it anyway (insert cutaway to 2018).

[mizzer]

Originally Posted by Exposed

This is more mysterious than the time Bill dissed the 2080ti and bought it anyway (insert cutaway to 2018).

[Seyiji]

Originally Posted by clerick

Swapped dll to 2.2 in death stranding, fixed the effect and particle artifacts, also backpack appears sharper. Really works.

Now drop it into a DLSS 1.0 game and nut?

[clerick]

Originally Posted by Seyiji

Now drop it into a DLSS 1.0 game and nut?

Seems like it only works with games that utilize at least dlss 2.0. DLSS didn't use tensor cores, and was much slower too. Games that use 1.0 need an overhaul to work with 2.0, like metro/control did.

[Trunks0]

Originally Posted by Seyiji

Now drop it into a DLSS 1.0 game and nut?

Originally Posted by clerick

Seems like it only works with games that utilize at least dlss 2.0. DLSS didn't use tensor cores, and was much slower too. Games that use 1.0 need an overhaul to work with 2.0, like metro/control did.

DLSS 1.9 in control didn't use the tensor cores, pretty sure other 1.0 implementations did no?

That said though, DLSS 2.0 needs a very different set of data from the game to function. Most notably temporal information, which DLSS 1.0 wasn't making use of at all.

[clerick]

Originally Posted by Trunks0

DLSS 1.9 in control didn't use the tensor cores, pretty sure other 1.0 implementations did no?

That said though, DLSS 2.0 needs a very different set of data from the game to function. Most notably temporal information, which DLSS 1.0 wasn't making use of at all.

Control didn't originally, but in the ultimate update it was using 2.0 yeah. 1.0 was a different beast altogether pretty much.

[acroig]

https://wccftech.com/nvidia-dlss-2-2...-improvements/

[Trunks0]

One of the few YT videos still up comparing and talking about this. I have to wonder why we are not seeing more... this is pretty cool.

[Megaman]

Are we at the point where we can safely discard all other AA and filtering options and just stick to latest and greatest?

Becoming difficult keeping on top of all these.

[Trunks0]

Originally Posted by Megaman

Are we at the point where we can safely discard all other AA and filtering options and just stick to latest and greatest?

Becoming difficult keeping on top of all these.

No. Temporal methods are the current trend, but they all? produce some artifacting or ghosting.

[Megaman]

Well that sucks.

[Munkus]

Originally Posted by Megaman

Are we at the point where we can safely discard all other AA and filtering options and just stick to latest and greatest?

Becoming difficult keeping on top of all these.

If you're gaming in 4K, I'd argue that AA is mostly unnecessary, IMO

[Trunks0]

Originally Posted by SIrPauly

You really need to investigate Dlss first hand.

I've had very little hands on. But I've said it before and will say it again, if you have the option you should probably use it. Especially 2.0 and newer, as it looks pretty great. Perfect? No it is not, but if you need the performance go for it.

*late edit*
(or if the AA options are limited, you should probably use it then as well)

[Exposed]

FSR previews are out and it's pretty much what's expected, with Hardware Unboxed seemingly going out of their way to be lenient on AMD despite being very hard on Nvidia in the past even regarding DLSS 2.0

It seems like since their Nvidia review fiasco they've taken a stance to be deliberately negative towards Nvidia. You can see here they try their best to avoid stating DLSS 2.0 as superior to FSR despite admitting FSR to be noticeably blurrier and inferior in most cases.






Gamer's Nexus was far more damning of FSR. So much to state that AMD probably avoided DLSS 2.0 in direct gaming comparisons for a reason.





Digital Foundry's take as well.





Basically FSR falls short of DLSS 2.0 especially in 1440p/1080p upscaling. No surprise here.

While AMD can continue to improve there's only so much they can do with their method of upscaling. DLSS 2.0 will always have access to much higher pixel data and temporal data to make a better quality image compared to FSR. So I see this as really as a stop gap until AMD has its own resources to do their own AI based upscaling sometime in the future.

BTW I didn't post this in the AMD section because I wanted this to include discussions with DLSS 2.0, and now DLSS 2.2 as well.

[Trunks0]

Yup, it's not better than DLSS 2.0 & newer that's for damn sure. Also doesn't seems like it's better than EPIC's TSR(Temporal Super Resolution), but thats exclusive to UE5 based games. It will make for a nice hardware agnostic alternative though.

DF's comparison to UE older temporal upscaler was interesting, I wish they took a deeper look on that. That comparison was super shallow

[Exposed]

Originally Posted by Trunks0

Yup, it's not better than DLSS 2.0 & newer that's for damn sure. Also doesn't seems like it's better than EPIC's TSR(Temporal Super Resolution), but thats exclusive to UE5 based games. It will make for a nice hardware agnostic alternative though.

DF's comparison to UE older temporal upscaler was interesting, I wish they took a deeper look on that. That comparison was super shallow

They did mention that TAA upscaling looks better than FSR, and that it's disabled in Godfall for some reason despite being available as a command line option.

In fact that expanded on HU's findings of FSR's best case scenarios, showing that it was really the lack of contrast and 4k textures that hid FSR's flaws in those best case scenarios.

FSR is basically only really usable for 4k users with the highest quality setting and only under those circumstances. Which is better than nothing I guess, and good news for players that have those supported games on those weaker GPU's. I think AMD may have rushed this just to have something to show, I hope they are in fact working on a temporal AI based solution just like DLSS 2.0 and put this out as a stopgap. FSR may show incredible weakness in games like Cyberpunk where high quality assets are used compared to Godfall / Terminator Genesis.

[Greasy]

Well to be fair it's their first go at it. We all know how crappy DLSS was at launch. What I would really love to know is do the tensor cores actually do squat for DLSS? Remember how Nv said gsync required a special $100 chip in the monitor?

[clerick]

Originally Posted by Greasy

Well to be fair it's their first go at it. We all know how crappy DLSS was at launch. What I would really love to know is do the tensor cores actually do squat for DLSS? Remember how Nv said gsync required a special $100 chip in the monitor?

DLSS 1.0-1.x didn't use tensor cores and had a minute boost of performance and quality. 2.0 onward is using them and we have the IQ that we do. Basically to make DLSS work (or anything neural net), it requires very fast matrices compute which tensor cores basically are solely made for.

Bigger breakdown here.

https://www.techspot.com/article/204...-tensor-cores/

[Trunks0]

Actually DLSS 1.9 in control had a better FPS bump than it's DLSS 2.0 update(YT it). Doesn't look as good though. Also other DLSS 1 titles did use the tensor cores. However as Control proved, it was not needed.

DLSS 2.0 may be a different story... but UE5's Temporal Super Resolution(TSR) is kinda making it look like once again it may not have been needed. It's going to be very interesting once we can start looking at direct comparisons between TSR and DLSS 2.0.

[Exposed]

They weren't kidding when Nvidia stated DLSS is always constantly improving and evolving.

We're up to DLSS version 2.2.9 now?

Meanwhile: