/vr/ - Retro Games

I must be finally going blind. I see literally zero difference between these.

16x is a mess, especially if you don't have a super high resolution screen.

>>5453596
Explain how higher anisotropic filtering makes image quality worse.

>>5453636
Have you ever played with anything above 4x on a lower res screen anon? Look up mip mapping examples or something shit

>>5453592
More detail

>>5453578

>>5453647
I always set it to 16x because I had no reason to think it would look better at 4x or 8x. Talking about playing on a 1080p display though.

>>5453660
1080p is fine for 16x. You get some bad artifacts and just general duckery at lower resolutions though. Especially in games without proper mip mapping

>>5453592
The trick is in understanding what theyre doing. Texture filtering is used by the rendering pipeline to answer the question "what color do i make this pixel when there are 10 texels (texture pixels) that all overlap on it" - eg, pixels where either very high detail textures are on the screen or where things in the scene are far away. (Imagine a high def texture on something so far away the object appears as 1 pixel.)
tldr look at the distant ground, the hill, the trees.

>>5453676
So to elaborate further on this, looking at the characters in the foreground you'd see no difference, since it's close to the camera anyway. The bilinear/trilinear filters don't take orientation of the textured surface into account whereas anisotopic filters do, so to compare the two types you want to look at surfaces that don't face the screen, eg, don't look at the stone column faces pointing toward us - look at the faces pointing sideways or upwards.

Just look at the stone texture on the ground. It's a huge difference

>>5453578
they're good?

>>5453578
Ansiotropy, mipmapping and filtering are different things.

Anyway, if you want crisp, sharp textures all the way then this is how you should set it up:
nearest textures + linear mipmap +16x anisotropy.
Surprisingly, linear mipmaps look crisper and preserve more detail at the distance
nearest neighbour is a pretty much a must for every game that was not designed for it (every pre-1998 game except N64 ones)
IIRC one of the old Doom threads had an extensive gallery comparing all possible combinations

>>5453592
Using a texture as-is makes it look like shit at an angle in the distance (like floor or a wall)
with nearest-neighbour it will turn into a shimmering mess as texels (as >>5453676 explained) would constantly fight for dominance)
with linear/trilinear it will turn into a blurry shitty mess
Anisotropy fights that by subdividing mipmapped textures which results in a much better filtering and crisper texture.
Watch the distant textures on the OP.

>>5453647
>>5453660
The difference between 8x and 16x would be more visible in the distance - 16x is crisper, and the performance impact is absolutely neglegible on any PC made in the last 15 years or so.

>>5453667
You don't get any fuckery at lower resolutions, unless you checked some stupid shit like "trilinear optimisation" or "anisotropy optimnisation" in the hardware drivers, which supposedly makes it run like 0.1% faster, but if you are running at lower resolution you dont need that in the first place.
Anisotropy includes mipmapping in itself, as it is a mipmapping technique (it uses generated mipmaps as input) and separating the two would require some trickery, though I have no idea why would anyone deliberately do that.

>>5453596
>16x is a mess
Do you actually believe anisotropic filtering is a full-screen filter or something?

>>5454451
>with nearest-neighbour it will turn into a shimmering mess as texel
That's actually what mipmapping prevents. When you do anisotropic filtering these days it's trilinear + anisotropic, not just anisotropic.

With mipmapping but no linear filtering you get visible miplines. Without anisotropic you get incorrect sampling at angles. It's possible to have anisotropic without linear filtering too. In fact a famous limitation of the old Radeon 8500 GPU was that it couldn't do anistropic and linear filtering on the one texture, either one or the other, while its competition the GeForce 3 could do both on the one.

>>5454487
>That's actually what mipmapping prevents
I was talking about how it looks without mipmapping, as I said
>Using a texture as-is
aka not mip-mapped.
Also in most instances linear mipmap + anisotropic is crisper than bilinear or trilinear mipmap with anisotropic due to how they interact with each other.

>>5453647
>mipmapping is good for any other reason than to save on texture filling rate

I hate this meme.

>>5454503
>no mip map
>distant textures a mess

All of them look objectively worse than simply having no filtering at all.

>>5454503
>>5454512
>mipmap, mp anisotropy
>distant textures fine and sharp, textures at an angle a mess

>>5454495
>Also in most instances linear mipmap + anisotropic is crisper than bilinear or trilinear mipmap with anisotropic due to how they interact with each other.
Not really. Trilinear + anisotropic is the gold standard. Without it you still get aliasing and noisy fake detail.

>>5454503
Pic related. Without mipmaps altogether you just get a different kind of noisy fake detail. Looks alright in stills, but shimmers like a bitch when the camera is moved.

>>5454503
>>5454512
>>5454516
>mipmap, anisotropy
>all textures are fine

>>5454503
>>5454512
>>5454516
>>5454518

>>5454517
>trilinear filter/mipmap, anisotropy
distant textures are slightly blurrier than with linear mipmap

Also that's not called "fake detail" you fag, thats shimmer from texel overlap. "Fake detail" would be adding shit that isnt there.

is this fake detail?

>>5454520
>Also that's not called "fake detail" you fag, thats shimmer from texel overlap. "Fake detail" would be adding shit that isnt there.
It's fake detail from aliasing which is the outcome of nearest neighbor sampling.

>>5454529
Yes. The aliasing generates a noise pattern which is often mistake for actual detail by the misinformed.

>>5454565
So, we are inventing new terms again, fine, you bumwhacked libberming

>>5454568
All of those filtering methods (mipmap, bilinear, trilinear, anisotropic, etc) were all designed to combat aliasing. Aliasing is high frequency noise. In this context noise is extraneous detail because it emerges from improper sampling.

>>5454512
>>5454516
>>5454517
>>5454518
I LIKE THE NOISY PATTERN

IF I WANTED TO SEE THINGS WITH LESS OVERLAPPING SCALING ARTIFACTS I WOULD USE SSAA OR RENDER AT HIGHER FUCKHUGE RESOLUTIONS, NOT EMULATE MY SHORTSIGHTEDNESS

Am I an heretic for playing old games this way?

>>5454618
indeed

>>5454518
>>5454520
>>5454529
Point sampling has its charm, but more objectively, it hides low color accuracy, either in textures, or in the frame buffer.
Try running a game in 16-bit with texture filtering and no dithering. All I can say is yikes.
For your standard glorious 24-bit colors, you're a drooling brainlet for using nearest neighbor.

>>5455041
this is /vr/ board
we are talking retro games

>>5454618
Opposite, it's the smart way. There's so many choices in shaders, why bother with wasting space and money? Most TVs these day eliminate any practical input lag with game mode and you can hang out on the couch playing the best zelda game (my favorite at least) on a large television.

>>5455046
Many /vr/ approved games are 24-bit throughout, most notably Quake 3, which even supports texture compression.

>>5454503
Is this because hardware works with four samples concurrently, and it falls back to working with one when the texels are too close together?

>>5453578
Soulless | More Soulless | Even More Soulless | The Most Soulless

>>5454512
>>5454516
>>5454518
>>5454520

GL_NEAREST_MIPMAP_LINEAR master race reporting in

>>5454585
If you prefer the noisy pattern there’s nothing wrong with that. But don’t pretend it’s the ‘objective’ best option or something like that.

>>5453578
Which game is this?

>>5455505
Let me get this straight, the amount of linears refers to how many dimensions of filtering are going on, independent of which ones are actually in play?
So "trilinear" mipmapping with nearest neighbour texels is actually just linear, and ordinary texel interpolation with and without mipmapping are both bilinear?

Yeah, the OpenGL designation is far less confusing.

>>5456505
You can use bilinear with or without mipmapping, and you can use trilinear with or without anisotropic...

>>5456514
You need an entire matrix of screenshots to demonstrate every single combination. It is more elegant to demonstrate an array of incremental improvements.

>>5456524
I don't see any difference between bilinear and trilinear in that pic. If it's properly staged, the bilinear should have a visible mip-line while it should be smoothed out with trilinear.

>>5456529
I see it easily. Look closer at the alpha texture in the distance.

>>5455838
So Crt was the original filter for pixels.

>>5456534
Yeah I see it now.

>>5456537
>Filtered 2D sprites

>>5453596
>>5453596
brainlet

Thread is full of brainlets who can't into graphics

>>5456546
(You) all the brainlet posts in this thread, excluding this one, of course ;^)

>>5456542
wtf is this from

>>5456791
looks like that yoshi's island 2 commercial

dreamcast ftw

>>5453578
Bilinear for performance. I don't think /vr/ even had Anisotropic???

>>5456814
Trilinear is relatively expensive on Dreamcast. PowerVR2 and N64's GPU both require the same number of cycles to do it, despite N64 being a last-gen system.

>>5455050
>TVs these day eliminate any practical input lag with game mode
>tv
>game mode.

I guess pc monitor master race as usual.

>>5456814
Wats that sweetie

>>5453578
They change the gameplay immensely
This intense obsesssion with graphics unironically and singlehandedly killed gaming. It's a soulless corporate platform and even the oldies are being beaten to death with this stick. It's a plague that annihilated all things constructive within gaming and its communities.

>>5456991
the hardware/technology is the only thing interesting about video games. If you want arbitrary challenge just interact with real life.

>>5456991
>Yikes. This intense obsesssion with graphics unironically and singlehandedly killed gaming. Caring about looks is so fucking problematic and gross. It's a plague that annihilated all things constructive within gaming and its communities.
Please leave.

>>5453596
Only on /vr/ will you find people who actually defend blurrier less defined textures. This place never ceases to amaze me.

>>5456832
https://www.khronos.org/registry/OpenGL/extensions/EXT/EXT_texture_filter_anisotropic.txt

Geforce 256 had anisotropy, though it wasn't very good.

ive got this projected on to the wall
>>5456989
nice

>>5456991
https://www.youtube.com/watch?v=uB24HiY9A1A

why the fuck do textures look shittier as they go along anyway?

>>5457226
see
>>5453676

>>5457226
Because textures are squeezed vertically, the hardware looks up low-res versions of the texture to avoid aliasing. The texture however isn't squeezed horizontally, and so the low-res textures being used become apparent.

The real question is: Trilinear or bilinear AF x2? Certain hardware can do either for free.

>>5457202
Dreamcast’s PowerVR2 also supports it.

>>5457792
is there anything the powervr can't do?

>>5457817
Can’t do HW T&L or complex pixel shader effects like EMBM or DOT3 but otherwise can do pretty much everything.

>>5458147
>Can't do T&L
Moot point since dedicated DSPs filled this role
>No dot3
What is this notion about it being able to do normal mapping, then? Granted, it was not usable in practice.

>>5458228
>Moot point since dedicated DSPs filled this role
Yeah, but it’s better if it’s on the same die as the pixel pipeline.

>What is this notion about it being able to do normal mapping, then?
It can do “DOT2”

>>5453592
>I see literally zero difference between these.
click the picture to enlarge

>>5458245
What is the advantage of on-die TnL? Fewer pipeline bubbles?

>>5458287
An internal connection between one part of the die and another is super fast compared to an external connection. Not to mention that in consoles like the Dreamcast and PS1, the vertex data has to share the same bus as the CPU.

>>5458287
legs are better than ass

>>5458315
Well, it only has to be as fast as it can spit out data, but of course, if the bus is shared, it will inevitably be hindered .
Geforce 256 does something like 15M vertices per second. Whether it's capped by the T&L or the triangle setup, I don't know. It claims to have six polygons "in-flight", whatever that means.

>>5457994

wut filter?

>>5458725
one of the crt-royale versions

>>5457792
>>5457817

I doubt anisotropic is gonna help much on native 480p. Maybe Dreamcast emus can make better use of that.

>>5459818
What are you talking about? Low resolutions especially benefit from anisotropic filtering.
Tons of modern games forego it on consoles, and it looks mostly fine since the blur doesn't do much to the already super sharp image.
Dreamcast games suffer heavily from muddy textures just meters away when they are seen from oblique angles.

>>5453578
I literally cannot tell the difference in any of these pictures
What game is this?

this will be the sixth time we've had this thread and we've become exceedingly efficient at it.gif

>>5453578
I can't stand when people obsess over minute differences in graphics or audio. Art-style is, and always will be, 10x more important than any filter or graphical improvements you can make to a game. In my experience, the types of people that obsess over filters won't even touch retro games, because anything less than a modern console/PC "looks like shit" in their eyes.

If you want to use a filter, sure, go for it. But don't claim that they're mandatory, or start berating others for not using them.

>>5460484

>>5462067
texture filtering is not the same as a post-processing "filter"
if there is a technological improvement - why not use it? why argue for not using it?

artstyle is subjective and arguing about it is like arguing which cheese is the best

>>5462387
But gorgonzola is undeniably best.

>>5453578

No filter is best. Otherwise, settle for bilinear.

>>5460457
Many Dreamcast textures don't look blurry due to a lack of anisotropic but a lack of trilinear filtering causing unseemly mip-lines.