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!

[Spook]

Αναρτήθηκε αρχικά από Gizzmoe:

Αναρτήθηκε αρχικά από Spook:

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.

I realise this, but depending on your GPU it may also increase the cost of LSFG beyond a point that's positively offset by the use of DLSS.

I know LSFG's resolution scale allows you to decrease this cost, but it also noticeably increases artifacting below ~75%.


Not disputing btw that highest ingame-res+DLSS looks better.

[Gizzmoe]

Αναρτήθηκε αρχικά από Spook:

I know LSFG's resolution scale allows you to decrease this cost, but it also noticeably increases artifacting below ~75%.

65% should be fine though at 1600p, that's 1040p for the framegen.

[Spook]

Αναρτήθηκε αρχικά από Gizzmoe:

65% should be fine though at 1600p, that's 1040p for the framegen.

Personal preference really, maybe base-resolution dependant as well. When testing 3.0 at 4k, I found >75% had a negligible effect on artifacting.

[Gizzmoe]

Αναρτήθηκε αρχικά από Spook:

maybe base-resolution dependant as well

Yes, recommended is 50% for 4k, 75% for 1440p and 100% for 1080p, but I don't care, I also go for at least 75% at 4k and above if my GPU can handle it.

[Sh. Fred]

Okay, I'm here to report on some findings, fresh out of the oven.

The following results are from the opening area where the game is supposedly the most intensive, (which might be true because the system load seems to have only gone slightly down the longer I've advanced into the game, P.S: I genuinely can't NOT type too much details!)

Throughout all the tests, all the graphical settings are completely maxed out, and the CPU is sitting around mid 70s while GPU is around mid 80s.

The "Resolution Scalability" option in advanced graphical settings is also set to 200% and I couldn't tell you what it does to save my life, so after too many tests, I've just left it there.

Also these framerates are almost completely stable so there's no accounting for stutters and frametime spikes since I'm not leaving the area for UE5 to ♥♥♥♥ me over:


- 1200P with 0 upscaling, fullscreen, (and I'm guessing with the baked in anti-aliasing): low 20s.
It looks like a painting, photographed with a $5000 camera.
Sadly at this framerate, it is a photograph since I can't properly tell you what it looks like in motion.

- 1200P windowed (not borderless), no in-game upscaling, LS1, 1.3 factor, LSFG at 70% and X2, DXGI API: around 15.
Massive amounts of warping around James himself when moving the camera.

- 1200P, DLSS ultra performance, no LS, fullscreen: high 80s.
Looks almost fine aside from jaggies on all vertical and horizontal lines and smearing in distance details.

- 1200P, DLSS performance, no LS, fullscreen: mid 70s.
Less smearing, same jaggies.

- 1200P, DLSS balanced, no LS, fullscreen: high 60s,
No smearing, jaggies more prominent on vertical lines (♥♥♥♥♥♥ hate those more), with a few on the much longer horizontal ones.

- 1200P, DLSS quality, no LS, fullscreen: mid 60s.
Slightly more tolerable jaggies.

- 1600P, DLSS ultra performance, no LS, fullscreen: high 70s.
No smearing I can see, same annoying jaggies more prominent.

- 1600P, DLSS performance, no LS, fullscreen: low 60s.
Tolerable jaggies.

- 1600P, DLSS balanced, no LS, fullscreen: mid-high 50s.
Very tolerable jaggies.

- 1600P, DLSS quality, no LS, fullscreen: low-mid 50s.
I wish I could tell you there were no jaggies.
Except they still ♥♥♥♥♥♥ exist, they're still there, just much harder to spot unless I'm fixated on finding them (which I sometimes am).

1600P, DLSS quality, no LS upscaling, LSFG 3.0 at 70% and X2, DXGI API: high 40s base.
I "feel" like there's artifacts in distance when moving the camera, but can't exactly see them.
Like an invisible enemy I'm about to encounter in not too long from now, but can feel it presence there!

1600P, DLSS balanced, LS1, 1.3 factor, LSFG at 70% and X2, DXGI API: mid-high 50s base.
I would go with this if it weren't for the vertical jaggies taking away my attention from the game, it almost looks perfect.

1600P, DLSS quality, LS1, 1.3 factor, LSFG at 70% and X2, DXGI API: low 50s base.
What I originally settled with.
Unless I'm going insane (which is extremely likely and that's not even related to the game or Lossless Scaling) this looks perfect!
It looks like 1200P (no scaling) in terms of clarity, except it's no longer a photo and is fully animated.


These findings are just to have some more data out for the sake of having them, in case they help anyone at all at some point when they're googling around.

Coming full circle, I'm actually just gonna go back to what I had originally settled on (unless there's any suggestions you think could magically increase my performance while keeping visuals at the same level, which sounds dumb but I can totally imagine it's possible since Lossless Scaling runs off black magic!)

Somewhere along the line I completely lost track of what I was even trying to go for, which is why I made my last comment.
All the going up and down the scales and then compensating back for each one gives me an aneurysm sometimes.

Maybe in 10 years they'll finally optimize UE5's asset streaming so we can come back and experience these games perfectly.
Or if not, I guess RTX 9090 and DLSS 10 will do Epic's work for them way before Epic themselves do.

[(ノ°□°)ノ]

I didn't get a chance to read the full post due to time constraints, but the mention of SH2 and stutters caught my attention. Coincidentally, I've already tested this and even recorded a video showcasing the flawless performance of LS in this game (though there are some artifacts since the FPS is too low for X4):
https://drive.google.com/file/d/1yj1Wu36ifWPcf_G55FkE35bijOrXGwO5/view?usp=drive_link

[Sh. Fred]

Αναρτήθηκε αρχικά από (ノ°□°)ノ:

I didn't get a chance to read the full post due to time constraints, but the mention of SH2 and stutters caught my attention. Coincidentally, I've already tested this and even recorded a video showcasing the flawless performance of LS in this game (though there are some artifacts since the FPS is too low for X4):
https://drive.google.com/file/d/1yj1Wu36ifWPcf_G55FkE35bijOrXGwO5/view?usp=drive_link

Hey, thank you for tuning in!

I'll try X3 and X4 a bit later, but X2 is honestly just perfect.
Random question but do you think it would one day in the future be possible to run Lossless Scaling over fullscreen?

I don't know squat about programming but I know there's a reason that's not possible, but could that change down the road?

I'd die to an overdose of magic if that ended up happening someday.

[(ノ°□°)ノ]

It’s not possible to display anything over an exclusive fullscreen window, as this behavior is fundamental to the system and cannot be modified by any software. However, LS can capture fullscreen content and output it to a secondary display, if one is available.

[Sh. Fred]

Αναρτήθηκε αρχικά από (ノ°□°)ノ:

It’s not possible to display anything over an exclusive fullscreen window, as this behavior is fundamental to the system and cannot be modified by any software. However, LS can capture fullscreen content and output it to a secondary display, if one is available.

That'll come in handy.
Thanks again!