Aspect ratio: the original?

I feel like mentioning Dimahoo as a weird edge case. It looks like parts of the art were designed for one aspect while others were for the other.

This is it at 3:4 (MAME Standard aspect):

(I’m using stand-alone MAME for this comparison since it’s easier to switch aspect) With this you get correct looking circles for the craters in the background and the top of the tank, but the diamond shaped pickups, score at the top of the screen and the UI at the bottom look skewed.

In 7:12 (MAME Pixel aspect):

You get everything wrong with 3:4 right, but everything that was right before is now wrong. There’s no perfect aspect for this game that makes everything look right. But I think 3:4 looks best for most things in this case.

Obviously the circles are circular in one shot and not in the other. The one in which circles are circular is the one intended by the artists. Why is the height in your shots not an integer height, by the way? It’s windowed mode.

Awakaned, just what I said. Anyway I think neither are circular, the first shot circles are a bit too fat. I think you are getting misguided with those craters, I believe they aren’t supposed to be circular (maybe perspective wasn’t supposed to be 90º straight). If I were you I would keep 3:4 since that seems more correct.

Physics dictates that explosions and rotating turrets must be circular. Artists agree.

It’s clear which one has circles. The other does not.

(I used shift+drag to create circles with ms paint.)

Yes, doesn’t match as you see. Since I corrected so many AR on video processing I’m used to it and can tell without even checking, but perfect circles tend to look thinner than fatter to our eyes (despite being perfect circles), and for me those look a tad fat (1st image). Crater pixels out of the circle on the sides, and grass pixels inside circle on top and bottom. On tank drawing circle outside outline on top-bottom and inside outline on the sides. With such low resolutions you can’t really make cheap assumptions. 12:7 or 16:10 or 16:9 (or 3:4, 2:3, etc in your case), you are not gonna find out differences on such small margins.

What are you saying? What am I reading? I don’t even understand what you’re trying to say.

To clarify my previous post, I was saying the top screenshot shows circles, the bottom one does not, and that it’s obvious to anyone. There’s really no need to brag, since being able to recognize a circle visually is a basic human cognitive function that everyone posessess.

You are so wrong my friend. I know what you are saying and I keep saying you talk too much and without knowing. Let’s count pixels. Your 3:4 ratio (correct geometry by your untested methodology)

And this is perfect circle image (1.15 deviation from 3:4), which IMO isn’t the correct ratio for the game. I keep saying circular things (craters, etc) were supposed to be oval in this game (look that skewed tank).

This thread… is amazing.

These discussions are always hilariously contentious, and they’re constant in emulation forums. I’ve seen like 3 just like it in the past year on byuu’s forum.

The fact is: the displays that existed when the games were made were 4:3. Artists knew it, game designers knew it, everyone knew it. You can’t just base pixel aspect ratios on circles because, as we’ve seen, they’re not even consistent in a single game. This is because when you draw small things in pixels, the ratios of x-pixels to y-pixels are more important than the overall aspect, while in larger objects, the overall aspect is more important. Does that make sense?

LCD monitors have square pixels, CRT TVs didn’t. CRTs didn’t care what the pixel aspect was, though, because they didn’t “know” what a pixel is. They just take in a signal and spit out an electron beam against a roughly 4:3 tube.

If you absolutely need the sharpest pixel edges, set it to 1:1 PAR (i.e., assume square pixels everywhere) and circles/artists be damned. If you want an “authentic”/nostalgic look, set it to 4:3. At 4:3 on a square-pixeled LCD, you’ll get pixel warping with nearest neighbor, so if you don’t want that, use integer scaling on the y-axis and something like bilinear on the x-axis (see: pixellate.cg, etc.).

If you’re emulating on a CRT, set it to native resolution (or at least native on the y-axis and a super-wide resolution on the x-axis) and 1:1 PAR then use hardware knob adjustments to fill the screen.

That’s it. End of story.

1 Like

What I find hilarious is how on every other thread about aspect ratios we need to explain the basics over and over again because some people talk without knowing when the topic is way over their head. Some people should need to search about DAR and PAR before commenting in such threads.

That said in the end of the day, it’s about what you find more pleasant or think about designers original intention, so whether some sprites are perfectly round, square, or anything, what you have to weight is the overall look of the game and the sum of their sprites. So sorry for the guys that want to put everything into a maths equation, there’s not such thing.

Stand alone MAME will stretch vertically to the top and bottom of the screen in windowed mode by default instead of using an integer like RetroArch does. I have bitmap prescaling set to 4 which does some blurring similar to what pixellate does, so it doesn’t look too bad. Though I only use stand alone MAME for testing things anymore.

I used to use pixel aspect for everything until someone on GAF pointed out how the Saturn version of Megaman X3 used pixel aspect and was a bad port because of it. For some reason that got me to try 4:3 again (I had tried it before, but not for very long because it looked “weird”) for home consoles and arcade and I’ve stuck with it since. Now pixel aspect stuff looks off to me, even games like Super Mario World, since I’m used to 4:3.

Yep, ultimately it comes down to: whatever looks good to you, use it :slight_smile:

(Unless you want to stretch 4:3 content to 16:9. You monster.)

One thing worth mentioning is that for certain systems you can keep per-pixel accuracy, but use a custom aspect ratio to get you close to the original.

Running SNES games at 5x4 (1280x896) is a lot closer to a 4:3 image than 4x4 (1:1 PAR) scaling. (1024x896) With certain games that don’t fill the screen, you can get away with 6x5 (1536x1120, cropped to 1080)

There’s a lot of misinformation flying around in here, so I’m going to try to briefly cover the basics. Hello, boulder.

Classic TVs were 4:3. Does this mean that classic games consoles displayed a 4:3 picture? No. The games were designed to be played on a 4:3 set, but that doesn’t mean the console’s aspect ratio was 4:3. For example, the majority of Sega Master System games had an active area of 256*192 pixels and the signal is padded out to 240 lines with background color. It’s kind of like watching a movie in letterboxed format. Then you have to factor in the pixel width (even though the TV doesn’t know or care what a pixel is, the pixel timings determine how long before the next pixel starts) which determines the aspect ratio in concert with that of your TV.

What did games look like on my old TV as a kid? How can I replicate that accurately? Nobody knows what your TV was like, even you. This question comes down to what your TV was like when you bought it, what you did to the settings yourself, how old the TV was and how much use it got. You can replicate it accurately to whatever extent you remember it by doing whatever you feel makes it look right. If you’re happy at this point, you can stop reading.

So what’s the correct aspect ratio for all consoles? Pff, no, not even … what. It’d be really, really nice if this was something that had an answer, there just isn’t one correct aspect ratio. Even if you assume the display is 4:3 (this is fine for most consoles and horizontally-oriented arcade games), the console is a major variable factor. The resolution and pixel timing of each individual platform is what will determine this, and there are almost as many aspect ratios as there are consoles.

Bleh, this sucks. What do I do then? Yeah, it does suck. The easy answer? Quit caring about it, just use 4:3 and call it close enough. It’s usually going to look pretty OK. But it’ll almost never be “correct” (whatever that means). Plus, emulators introduce problems here; e.g. most SNES emulators will hand the user a 256224 image, even though the actual signal produced by the console was 240 lines tall (usually the remaining 16 lines were automatically filled in black). If you were going to force anything to 4:3, it would be that 256240 image, not the cropped 224-line picture the emulator gave you.

That’s not good enough, I’m a big asshole and I want the right answer. What is it? Welcome to the asshole club. At this point, you need to run around finding the pixel timing for each platform you want to run at the “right” aspect ratio, keeping in mind that this will be correct in a mathematical sense based on the assumption of a perfectly calibrated contemporary television, which none of us had. The Pin Eight wiki has an excellent resource on dot clock rates, which can be used to determine the console’s pixel aspect ratio. You can multiply the PAR by the width of the image in pixels to find a width that matches the “correct” aspect ratio of the original platform (again, on your magical perfect TV that you don’t own).

Let’s use the NES as an example. You’re playing Kirby’s Adventure (because there’s circles in it!), at the NES’s 256240. Your monitor’s pretty big, so you’re displaying at like 5 scale, 12801200. You need to do 8/71280, which rounds out to about 1463. This gives you an exact (display) aspect ratio of 128:105, which is actually quite a bit slimmer than 4:3.

You big dumbass, how come all my old NES games fit exactly on my 4:3 if the picture’s not 4:3? Two answers.

  1. You or somebody else set up your TV to make it look that way by twiddling some knobs somewhere. This is fine, I’m not critiquing your choice, but whatever you did required you to distort the image more horizontally than you did vertically (it looked wider than it “should”). I want to stress that this is a fine decision, but pushes the picture away from the expected ratio on the last question’s magic TV.

  2. It didn’t, you just thought it did. The most obvious reason for this is that some of the picture was actually off the edge of the screen. This is totally normal and even expected behavior: witness the leftmost column on scrolling NES games or the black emptiness on the top/bottom of many games; not to mention the large gap between the edge of the screen and where things like healthbars are usually placed. They’re not right on the edge because nobody would have ever seen them off in the overscan area (one of the old NES Castlevania games famously screwed this up and the HUD is a bit illegible).

Anything else? Probably. There’s always something else.

So the idea is that it’s correct because it’s based on math using a hypothetical display that reflects no display/experience anyone ever had or has (except, I guess, broadcast monitors that haven’t knobbed the image out to 4:3)? If so, that’s the same reasoning byuu used for his obviously-not-quite-right NES palette, which is another highly contentious topic. He wanted something that was objectively irrefutable, even if it subjectively failed to match anything anyone remembers. I certainly understand this reasoning, but I think it’s specious to call it “correct.” Perhaps “idealized” would be a better descriptor.

I agree 100% about the SNES’ 16 px padding needing to be factored into 4:3 stretching, though, if you’re wanting to be a stickler. That’s how higan does it, for example.

Pretty much, and that’s why I went to such lengths to put “correct” and “right” in scare quotes all over the place. I can see using “idealized” for the same reason. I think it makes more sense to look at this approach as hitting the median, rather than being what’s right. While the odds of home users experiencing their games in exactly this way are low, this should hit about the midpoint for home experience. Some sets would be calibrated slimmer and others wider, so this marks a solid compromise between the expectations of different users, which as a bonus can be backed up by the fact that it aligns with the mythical perfectly calibrated set.

For what it’s worth, Nintendo shoots for around the same values as suggested above when emulating classic games. On the Wii Virtual Console, NES and SNES games are scaled from 512 pixels wide up to 640. Wii pixels are 10:11, so the resulting PAR is 25:22 (7.95:7). The “idealized”/“median”/etc. is 8:7. That is to say, Nintendo is 0.57% off from the “idealized” value. Anything close to this area seems quite reasonable, but obviously individual preference trumps math any day.

Yeah, that’s fair, I think. :slight_smile:

Having a value that’s backed by math and reasoning rather than a subjectively reached, what-I-remember-from-25-yrs-ago shot-in-the-dark definitely has value when dealing with these endlessly debated topics. At the very least, it lends people a sense of closure to an otherwise endless obsession of tweaking.

Vague Rant, you are adding even more confuse and misinformation to the people saying half truths, and truths explained in half. You are the reason these threads get longer and longer and longer. You make it sound more complicated than it really is. Not care for Aspect Ratio? Really?

Classic TVs were 4:3. Does this mean that classic games consoles displayed a 4:3 picture? YES. Everything was scaled to fit your 4:3 TV, all the games regardless of their internal PAR (pixel size) were scaled to be 4:3 DAR (although this might not be correct).

What did games look like on my old TV as a kid? How can I replicate that accurately? Easy. Set up emulator to display 4:3 DAR. As Vague Rant explained, some emus output a different resolution than original console (emus occasionally remove some padding), so double check that before, and re-adapt target DAR if necessary.

So what’s the correct aspect ratio for all consoles? What do you mean by “correct”, how you played as a kid? Read above. Or geometry correct? this is another (and long) story. Every console (and even games) had a different resolution, and hence a different PAR (Pixel Aspect Ratio). Designers had to account for this PAR to make things look correct (eg. a perfect square or circle) AFTER the TV scaled content to 4:3. This means that if you saw the content as 1:1 PAR (straight from emu without resize) you should see perfect circles and squares vertically stretched. Problem is, MOST designers didn’t even care for this (and you thinking a beautifully crafted 60$ game couldn’t have such blatant fails), they drew everything in their correct look at 1:1 PAR and there be dragons in 4:3. And this is my friend why you never saw Sonic in TV as a perfect circle when jumping. So yes, geometry correct is a PER GAME thing, and not console, it depends on designers. Some designers accounted for the 4:3 stretch (F-Zero) and some didn’t (Super Mario World), yes, even within same company. Now were coin block in SMW supposed to be square? hmmmmm that’s a good question.

Bleh, this sucks. What do I do then? THAT IS what this thread IS ABOUT. Looking at a game and decide what looks (geometry) correct or not. Mind you, some designers mixed elements with different inherent PAR in the same game, so some sprites will never look good in either DAR (you will have to choose one). So this thread is not ABOUT discussing superfluous and known facts, but judging supposedly correct looks.

That’s not good enough, I’m a big asshole and I want the right answer. What is it? Right answer is “What do you think designers thought as good aspect ratio” (rather than “what do you think is good AR”). Think this way to decide what aspect ratio to use for each game. Think as a designer to evaluate your game’s DAR.

You big dumbass, how come all my old NES games fit exactly on my 4:3 if the picture’s not 4:3? Yo, mothafucka, first respect, do not insult me. Now, as I said before “scaling”, resizing, or whatever you wanna call it. All TV’s have a scaling chip embedded for this.

Anything else? I have no answers to no questions.

edit: oh, yes, and I wouldn’t base my conclusions on what Nintendo does for the VC. Yes, coin block are square with 8:7 DAR but they are the same people that still keep releasing mangled PAL games to VC, you know those that played 16% slower and more horribly things people rage so much.

Meh. My scenario involves magic TVs which are tuned perfectly, your scenario involves magic TVs which are all tuned wrong but in exactly the same way. Your understanding of pixel aspect ratios is faulty. Perhaps you should have a look at something like this Commodore 64 doc (under the Aspect Ratio heading). It specifically mentions calculating PARs based on clock timing (and the necessary clock speeds for square pixels on PAL and NTSC), then also throws in that a less accurate way of making this calculation is to use the resolution and assume it fits a 4:3 area and work backward from there. It’s an OK way to operate and the results end up pretty similar, but ultimately, pixel clocks matter. They’re not there for fun, they determine how long each pixel is within a scanline.

Resolution is not the only factor in PARs, as you seem to be convinced. Hell, there are a billion arcade games that run at 256*240, but they’ve practically all got different pixel clocks, as seen on Pin Eight. To be clear, I don’t like that this is complicated, either. I think it’d be great if it was as simple as “slap it on 4:3 and that’s what every TV looks like all the time”. But it doesn’t work like that because there’s different consoles, TVs, user settings, etc. There’s not one answer, there’s as many answers as there are combinations of those three things. I’m not saying any of those answers are wrong, but none of them is right for everybody, including “it’s always 4:3” and also including “it’s whatever the pixel clock dictates”. It’s a messy issue.

Incidentally, while I wouldn’t use Nintendo as a confirmed standard either, I don’t think your reasoning is sound. Nintendo has said the reason they still release games running in letterboxed 50Hz for PAL territories is because that’s what is accurate to users’ experiences. I’m Australian; I know all about how annoying this is, but Nintendo is intent on making the games look like they did the first time around, even to the detriment of those games (in my opinion). If anything, this reaffirms Nintendo as jerky sticklers. But I still don’t think Nintendo knows best when it comes to emulation. Their NES palette is awful and doesn’t match any console or TV I’ve ever seen and they’ve even sometimes used non-integer scaling with no filtering (see Kirby’s Dream Collection). Those are great examples of Nintendo not making things look the way most people remember them, so I’d use those to justify Nintendo’s efforts being unreliable over them accurately displaying games the way they ran in markets they didn’t care about then and now.

EDIT: I suppose another question that might help to ponder is “Why do pixel clock-calculated PARs tend to result in such reasonable pictures?” Not even correct pictures, let’s just stick with “reasonable” for now. If pixel clocks are irrelevant, why bother using clock speeds which correlate fairly linearly along with the resolution, with a resulting PAR that’s in the ball park of creating a 4:3 image? The only major exception I’m aware of are the occasional arcade game with an unusually high pixel clock, where proprietors were directly instructed to calibrate their monitors using the test menu to make the picture fit. If all that mattered was resolution, a) arcade proprietors wouldn’t need to adjust anything because everything is always 4:3 and b) the clock speeds wouldn’t follow such a smooth curve, they’d just wildly fluctuate because they don’t matter.

With what we have, that is pretty much what we can do (explained above). I already made some petitions that got no attention. Although this gets into a development topic.

I pretty much test with 1:1 PAR, 4:3 DAR, 7:5 for some arcades…

I say this because normally designers would pixel-draw in 1:1 par, so most likely that is the good ratio. Sometimes is not and I would default to 4:3 since RetroArch doesn’t give any more options. To judge by your clock rate link let’s take CPS3 9:11 PAR, what this means is that in order to show a perfect square perfectly square one would need to multiply the width with this ratio, so 384x224 becomes 314.18x224. This makes 1.40 DAR, so your average display TV would then resize it to 298x224 (4:3). So this is no different than directly let your old TV scale the signal (224 by (4/3) = 298x224)

But now we can do things that weren’t possible back then and truly show the desired(?) 1.40 ratio.

I made some tests. CPS3 in 4:3 1194x896 CPS3 in 1.4026:1 (9:11 PAR) 1257x896 How do you see it?

Here a proof at ideal 9:11 PAR. What I said above (mixing sprites with different PAR in mind). The circle looks fine, the in-game drawing looks horizontally stretched, the drawing would match the in-game one if this was stretched to 4:3 (1194x896).

Anyways this is pure theory, we can’t do anything until the RetroArch guys either put a meaningful DAR (based on clock rate) available to choose or let us make a real custom DAR (instead of the current custom resolution thing).

edit: btw here’s a lot of talk on clock rate and stuff if you are interested.

You’re forgetting the extra 16 lines that are in the real signal. The active area on CPS3 is indeed 224 lines tall, but the full signal is 240. So using the “always 4:3” policy, the TV would "scale this to 320240" (scaling is kind of a digital process and not really the right word, but that’s largely semantics). Adding a 4 scale onto this as in your screenshots, you’d be looking at 1280960. In practice, emulators don’t show all 240 lines, so the equivalent for RetroArch purposes would be 1280896, and there’s your 4:3 display aspect ratio. Your screenshot is quite slim relative to this, which is what the game really looks like at 4:3.

Using 9:11 PAR, that would be 314240, or including the 4 scale, 1257960. Again, dropping the empty area, in RetroArch that’s 1257896, as in your image, which is dead on accurate. But when you include the cropped area (as is correct), you get a DAR of 72:55. Perhaps a more useful way to describe that is 3.93:3, which is just 1.82% slimmer than 4:3, hence why there’s only a 23 pixel difference from 4:3. And there’s the ball park.