Experiments and shenanigans: possible cure to the Unreal Engine 5 cancer?

[Sh. Fred]

So this is just a random post about some tests I was running last night and the results I got.

If anyone has any suggestions to tweak it or try something else, lemme know.
I know for sure there are a few spots that could use some tweaking in my settings across all the programs.
So feel free to comment on those.

I'm also posting it because I bumped into very few posts about running both LS and DLSS simultaneously and most of those posts say to not do it.
Yes, I know my formatting is ass, but I'm on the phone so please bear with me.
At the very end of the post, I'll arrive at the results so feel free to skip there if you want.

After screwing around increasing and decreasing the resolution in-game and in LS and trying various scaling methods quite a few times, this is what I ended up with in Silent Hill 2:

In-game:

- All graphical settings maxed out, not using the Epic preset but rather the custom one that then allows you to crank everything up to max one by one.

- Blur is off.
Also one of the options in the advanced graphics settings is set to HZB (Ryzen 9 7940HS so it makes sense?) But I don't remember what this option is called.

- Raytracing is off because ♥♥♥♥ it (will get to this later).

- Resolution is set to 1600P and it's in borderless mode.

- Upscaling is set to DLSS quality.

- There's no frame-cap and V-Sync is off.

In Nvidia Control Panel (Silent Hill 2 profile):

- V-Sync is off here too (default).

- Frame cap is set to 45 (I'll explain why later)

- Reflex is set to on.

- Everything else is left untouched.

- Also G-Sync is on and the refresh rate is 165 (but those are general settings anyway and I didn't wanna add another section just for this one)

In LS:

- Scaling mode(?) - And bear with me, I'm at work now and haven't memorized most of the LS options and feature names - is set to 1.3.
The option below it is set to fullscreen.
And the checkbox below them is also ticked to rescale the window or whatever.

- Scaling type(?) is set to LS1.
It's the most consistent one I found and I really wanted to try NIS but it looked straight up like the deep-fried meme.
FSR wasn't bad, but LS1 just... feels better!
.
And Anime4K looked the best I think, and I didn't get any performance hit because I've capped my frame so low (again, will explain below.)
I used the "small" option for Anime4K because the "ultra large" or whatever it's called made the game look like AI-generated nightmare with warps everywhere.
I think I only didn't use the Anime4K in the end just because I was told it was the most taxing one and LS1 looked perfect.
.
Also worth mentioning that I didn't use the performance mode or whatever the option for each of these scaling types is called.
The other scaling types I didn't mention didn't really do much of anything so I didn't bother.

- And finally there's Frame Generation.
I'm really struggling to remember all the options here but the mode is set to X2.
.
I lowered the resoultion scaling (or again, whatever it was called) down to 70 or 60% (don't exactly remember).
.
The API option(?) is set to the first one, even tho I went so low on the FPS cap to reach a very very stable number.
I think this might be a mistake because the tooltip for the first API option says to lock FPS to half my refresh rate, and the tooltip for the second one which I DIDN'T use said to cap it to a stable FPS.
Suggestions on this part?
.
Every other setting in this section below, I think I either didn't touch or really didn't think affected performance and changed it for personal preferences.
That's why I haven't really remembered anything from down here.


And now...

THE RESULTS

Lemme get a disclaimer out; I have bad eyesight and don't wear glasses, hence why I get very very close to my screen.
But I'm also extremely sensitive to small details and could spot even a single pixel acting weird in the corner of the screen.
I guess I am progressively making my eyesight worse in this cycle, but it is what it is.

I will eventually get around to testing Cyberpunk 2077 but that's a game I love AND hate visually.
I know it's random but it'll make sense with what I'm about to say next.

Cyberpunk looks amazing when going crazy with the settings, but unless I'm running at 10 FPS on DLAA and a virtual 4K resolution on DLDSR, the noise and artifacts drive me ♥♥♥♥♥♥ NUTS!
The noise and artifacts caused by DLSS, frame-generation and Raytracing in there are unbearable if I'm aiming for a playable experience at 1600P, no matter what I do with the baked-in upscaling settings.
And dropping to 1200P is straight up disgusting to look at with much artifacts there are.
I do admit I try to run all the graphical settings on max there to avoid anything looking rubbery or plastic-y, especially the textures.

Now with all that out of the way, lemme cycle back to LS and Silent Hill 2.
It looks... ♥♥♥♥♥♥ flawless!
Believe me I tried and really shoved my face into the screen and I searched and searched.
I genuinely can't see any artifacts, in fact I tried to find something so hard that it worried me a bit because I can't tell if LS is working or not, except I know it's working for a fact because I can see it scaling the window, I see the frame-rate input and output both on LS and Steam's counters, and I'm looking at the crispy (but not deep-fried!) textures!

I don't wanna go into any spoilers for Silent Hill but the 2 places I texted where the opening area and the hospital.
It ♥♥♥♥♥♥♥ is magic! And I've always been a non-believer, which is why I freaked out so much by this!
There is one very minor visual glitch which I know isn't actually caused by LS but I'll bring that up in a sec.

Let's get into a few little things.

Stutters... Yes, ♥♥♥♥♥♥ stutters, especially of the traversal variety.
Silent Hill 2 stutters, it stutters like a motherf*cker, just like any Unreal Engine 5 game before it.
I'm going to deliberately avoid talking about Timmeh Swine and his pesky little engine because if I did, then I'd completely derail my own train of thoughts.

Regardless, Silent Hill 2 stutters and I'm extremely sensitive to those, especially because I game on mouse and keyboard with a very high DPI.
And yes, sure there are naysayers denying the stutters because the game supposedly runs perfectly fine on their setups and everyone else is lying, but those usually disappear into the shadowrealm as soon as you ask them for footage with their frame times visible.

The reason why I've gone so low on the damn FPS cap is actually the stutters.


Alright, deep breath, forget Timmeh, forget Timmeh...

Input delay was one thing I was concerned about, but going from 45 to 90 and still playing on the same high-DPI mouse, I really couldn't tell you if I feel any different.
James' movement feels just as heavy as it was before I started using LS, and that's just a gameplay thing because James' both legs are apparently broken and in a full [invisible] cast.

Oh, forgot to mention, I wasn't using the built-in frame-generation before because it actually made my performance worse.

The tiny visual glitch I noticed was the shadows being cast on tiny objects such as vents with a lot of horizontal lines would flicker when looked at from a long distance.
This was already happening before LS while using DLSS and is also resolved when you get closer to the said object (6 feet and closer).

Might have to do with the draw distance or shadows, but remember, everything is maxed out in game.
You really wouldn't be seeing this unless you were literally standing in a ♥♥♥*ked up room looking at tiny little things with everything else out there coming for your ass, at least not often, not in Silent Hill anyway.

I guess I also forgot to say anything about raytracing but it was causing some weird glitches with lighting in some places so I figured I might as well get back the performance without it.
The baked lighting and reflections are perfectly fine anyway.

I don't have anything installed atm to measure the delay with, so I might have to give that a go.
But if it helps, I brawled 3 nurses in a tight corner with nothing but my trusty pipe and downed all 3, without taking a single hit.
Oh, also they really love to get back up after you've vigintuple-tapped them not even a minute earlier!
Now I can continue holding onto all the cough syrup I can find like the absoulute hoarder goblin I am.
IT'S MY COUGH SYRUP!

And I can do all that while locked onto a very very stable 90 FPS that doesn't seem to move up or down at all.
In fact I can now be that ♥♥♥♥♥♥♥ that claims my game is perfectly stutter-free and it must be everyone else' system or drivers causing issues for them.
Oh, don't ask me to provide any evidence tho, that'll make me vanish into thin air!

Sorry for any typos or bad formatting again.
We'll fix those in post!

[Gizzmoe]

I'm not going to read this, it's a f*cking book... Maybe some tldr about the questions you have?

[Sh. Fred]

Nah, it's fine.
It was a waste of time anyway.

[Gizzmoe]

Originally posted by Sh. Fred:

It was a waste of time anyway.

So you don't want any help and wrote this whole book for nothing?

[Spook]

Originally posted by Sh. Fred:

I'm also posting it because I bumped into very few posts about running both LS and DLSS simultaneously and most of those posts say to not do it.

The point to this is that if you set your game to say [2160p@50% ingame res-scale], your game is calculating only 1080p, but you are making LSFG process a 4k frame. Which is waaay more demanding than a 1080p, or even 1440p frame.

If you set your game to [1080p@100% ingame res-scale], your game requires the same or less GPU-power, and LSFG only has to process 25% of the pixels it would have at 4k. You would combine this with using LS to scale this 1080p frame up to 4k. Drastically improving performance, at only a minor loss in image quality.

LSFG's "Resolution scale"-slider now allows you to mitigate some of the cost of LSFG at high resolutions, but it's still an extra step and noticeably increases artifacting at lower settings.

[Sh. Fred]

Originally posted by Gizzmoe:

Originally posted by Sh. Fred:

It was a waste of time anyway.

So you don't want any help and wrote this whole book for nothing?

I wrote it as a report on some experiments I did that worked out well for me, with possible tweaks being offered on some points I missed.

Also to help anyone possibly going on the same path as me; just more data.

If no one's gonna read it, then so be it, I'll enjoy myself I guess.

[Gizzmoe]

Originally posted by Sh. Fred:

I wrote it as a report on some experiments I did that worked out well for me, with possible tweaks being offered on some points I missed.

And it's a good thing you posted that, but you could have fitted all those info into muuuch less text. Then more people would actually read it.

[Sh. Fred]

Originally posted by Spook:

Originally posted by Sh. Fred:

I'm also posting it because I bumped into very few posts about running both LS and DLSS simultaneously and most of those posts say to not do it.

The point to this is that if you set your game to say [2160p@50% ingame res-scale], your game is calculating only 1080p, but you are making LSFG process a 4k frame. Which is waaay more demanding than a 1080p, or even 1440p frame.

If you set your game to [1080p@100% ingame res-scale], your game requires the same or less GPU-power, and LSFG only has to process 25% of the pixels it would have at 4k. You would combine this with using LS to scale this 1080p frame up to 4k. Drastically improving performance, at only a minor loss in image quality.

LSFG's "Resolution scale"-slider now allows you to mitigate some of the cost of LSFG at high resolutions, but it's still an extra step and noticeably increases artifacting at lower settings.

This is interesting.

Does DLSS quality on 1600P render on 1200P or is it slightly higher?

I have to admit I've never actually bothered to test how the game runs on 1200P without any upscaling at all, so I'll try this today.

Although the main reason I use DLSS is actually the anti aliasing it comes with, so I don't know what the visual would look without it.

I guess I didn't make it very clear, but main intent with this whole experiment was to be able to run Silent Hill 2 about as smooth as I could and eliminate all stutters at no cost to visuals.

I don't intend on getting more performance out of it at any cost since it's perfectly playable at a stable 45 | 90.

I'll give your suggestion a try and see how it'll look.

[Gizzmoe]

Originally posted by Sh. Fred:

Does DLSS quality on 1600P render on 1200P or is it slightly higher?

The internal render resolution is 1067p with DLSS Quality on 1600p.

[Spook]

Originally posted by Sh. Fred:

Originally posted by Spook:

The point to this is [...] at lower settings.

[...] I'll give your suggestion a try and see how it'll look.

Does DLSS quality on 1600P render on 1200P or is it slightly higher?

https://www.reddit.com/r/nvidia/comments/oymxyp/comment/h7txr8n/
DLSS settings correspond to a percent-of-axis value.
I believe the above values still hold true for DLSS 3.

Although the main reason I use DLSS is actually the anti aliasing it comes with, so I don't know what the visual would look without it.

Your'e going to run DLAA if possible, TAA otherwise. The quality of TAA differs per implementation and ranges from way better, to way worse then DLAA. One major difference though is that you're less likely to encounter smearing/ghosting in TAA.

I guess I didn't make it very clear, but main intent with this whole experiment was to be able to run Silent Hill 2 about as smooth as I could and eliminate all stutters at no cost to visuals.

See, this is information someone can do something with.

In this case i'd start by establishing a framerate at which you are effectively CPU-"unlimited". I'd then settle for a ratio of 2x, because; "at no cost to visuals". I'd then figure out the best trade-off between in-game DLSS setting and a Custom "Scale factor" in LS.

[Sh. Fred]

Thanks for the suggestions.

I guess with 1200P at 100% resolution with no scaling, I'd have a small performance hit compared to DLSS Quality on 1600P.

I'll see how it goes.

Regarding the anti-aliasing, DLAA has always looked great to me in every game, but I'll have to see if it'd tank my frames too much without DLSS alongside it.

I'll run some tests today.

[Gizzmoe]

Originally posted by Sh. Fred:

I guess with 1200P at 100% resolution with no scaling, I'd have a small performance hit compared to DLSS Quality on 1600P.

Stay at 1600p with DLSS, it makes no sense to switch to 1200p without DLSS. Less fps, less quality.

[Spook]

Originally posted by Gizzmoe:

Stay at 1600p with DLSS, it makes no sense to switch to 1200p without DLSS. Less fps, less quality.

Less resources required for LSFG if your GPU is struggling, more than 40% less pixels to process. But if your GPU can handle it, run max-res and DLSS for sure.

[Sh. Fred]

Originally posted by Gizzmoe:

Originally posted by Sh. Fred:

I guess with 1200P at 100% resolution with no scaling, I'd have a small performance hit compared to DLSS Quality on 1600P.

Stay at 1600p with DLSS, it makes no sense to switch to 1200p without DLSS. Less fps, less quality.

Wait, so do I just turn off scaling in LS?

Because at the moment I'm running two layers of scaling on top of each other which I guess sounds dumb, but I thought it helped with the overall image stability and perfecting any possible DLSS artifacts.

If I did turn off DLSS and went down to 1200P, I'll see if LS does a better job than DLSS...

Wait, wtf am I saying.
My brain don't brain no more.
I'll get back to this later.

[Spook]

Originally posted by Sh. Fred:

Originally posted by Gizzmoe:

Stay at 1600p with DLSS, it makes no sense to switch to 1200p without DLSS. Less fps, less quality.

Wait, so do I just turn off scaling in LS?

If you are satisfied with the FPS you are getting when running your preferred level of DLSS with your game outputting 1600p, don't change anything.

[Gizzmoe]

Originally posted by Spook:

Originally posted by Gizzmoe:

Stay at 1600p with DLSS, it makes no sense to switch to 1200p without DLSS. Less fps, less quality.

Less resources required for LSFG if your GPU is struggling, more than 40% less pixels to process. But if your GPU can handle it, run max-res and DLSS for sure.

DLSS Quality is only 1067p at 1600p, even Balanced would look good and use much less GPU resources than 1200p native plus TAA.