Understanding the design of the the Super Nintendo video system

https://fabiensanglard.net/snes_video/index.html

by guiambros • 2 years ago

273 83 2 years ago

aidenn0• 2 years ago

I think TFA has a typo; it suggests the aspect ratio is 8:6 which would be the same as 4:3, my math says 8:7

Bluecobra• 2 years ago

You're right, the aspect ratio internally is 8:7. Emulators like Snes9x default to this. This effectively is rendered on a CRT display at 4:3 due to how CRT's work and mangle perfect square pixels to rectangles. At least that is my understanding.

fabiensanglard5• 2 years ago

Thank you, I fixed the TFA.

bombcar• 2 years ago

TFA here, of course, being The Fabien Article. ;)

rf15• 2 years ago

European TVs, especially those in France, came with SCART connectors

This is, as far as I can tell, an understatement - almost all TVs in europe offered SCART ports, the standard just originated from france.

ZFH• 2 years ago

Yes, SCART was for all of Europe. Mass adoption was kind of slow though, low end TV models often lacked the port well into the early 90s.

actionfromafar• 2 years ago

Older TVs did obviously not have SCART, and it took a while after 1976 when SCART was created for it to be widespread. Also, a lot of cheaper TVs had only RF in, and a lot of smaller TVs had RF only or RF + composite input only, even in Europe.

meindnoch• 2 years ago

I've never seen anyone use the SCART port on their TV. Component video or S-Video was much more widespread.

BlaDeKke• 2 years ago

Here in Belgium almost everyone used scart cables for game consoles, vcr and dvd players. We had 2 scart input but 3 scart devices. PSX vcr and dvd. So we had to switch from time to time. There were even scart switches, but they all sucked.

HeckFeck2• 2 years ago

Growing up in the UK, everyone I knew used it for VCRs, DVD players and digiboxes. It always made a notable improvement over the RCA jacks, and I longed to get the SCART cable for my PS2 (never did). Famously bulky and stubborn with their wires joining the connectors at an aggressive 45° angle; I never had one go bad on me.

I've read that it could even do HDTV in theory because it had YPbPr lines, but this was was never seriously attempted/rare in practice.

kalleboo• 2 years ago

In Sweden SCART was the default - your VHS deck or satellite TV box would come with a SCART cable. I don't think our 1995 TV even had composite or S-video ports, you had two SCART and RF[0]. For connecting up PCs or video cameras, we used a SCART to composite adapter plug (with a little switch to choose if it was in or out).

[0] found a photo of a period TV, not specifically ours http://www5.picturepush.com/photo/a/5316953/640/5316953.jpg

amelius2• 2 years ago

Why does the SCART connection use capacitors in the RGB lines? (last picture)

Doesn't that block low frequency signals (e.g. an all-blue screen)?

userbinator• 2 years ago

There's still horizontal and vertical blanking intervals.

amelius2• 2 years ago

Yes, but are the RGB signals required to be zero there?

MisterTea• 2 years ago

They appear to be there for high pass filtering but the 220uF value appears to be incorrect and is likely 220pF.

RetroSpark• 2 years ago

220uF is correct. Along with the 75Ω terminating resistor inside the TV, that forms a high-pass filter with a cutoff frequency of ~10Hz.

crtasm• 2 years ago

Meanwhile I was using the RF output and a switch box to pick between SNES and TV antenna.

mrandish4• 2 years ago

Oh my! The only way to get even worse video quality (and extra noise) was to add a two-wire UHF screw terminal adapter between the RF switch box and the TV. So of course that's what I had as kid on my 1970s knock-off Pong game console (we couldn't afford the real Atari). Later in life, video engineer me could laugh about this. I'm just glad that at the time kid me had no idea how awful it was. :-)

anthk• 2 years ago

game running slower

Eh, jus try Super Mario World with an emulator on PAL settings with an NTSC ROM.

The counter and music will go much faster.

criddell• 2 years ago

I haven’t seen carving used like this before. Is it a common usage?

jihadjihad• 2 years ago

Author is French; the French word for "carve" could translate to something like "sculpting/tailoring" in English, maybe that is what he meant? Tailoring the video system to the proper settings/resolution, or something?

fabiensanglard5• 2 years ago

Yes, that is what I meant. 20 years in the country and still not fluent :!. Anyway, I changed the title to "Designing the SNES video system".

theandrewbailey2• 2 years ago

No. Carving doesn't make sense here. He's not writing about cutting something for artistic purposes. I would use building or designing here.

I think jihadjihad might be on to something. Maybe this is a European English usage?

(I'm an American English speaker.)

rand0mfacts• 2 years ago

The power grid in Japan runs at 50hz in half the country and 60hz in the other half.

bluedino• 2 years ago

They probably did not pick that number at random. 224 is a number evenly divisible by 16 (224/16 = 14) which means it plays nicely with the graphic rendering pipeline tilemaps.

This was something that took a bit to figure out, but made so much sense to me after I had been playing around with trying to learn game programming when I was a kid.

CGA/EGA/VGA all had popular 320x200 modes.

The NES was 256x224, as was the SNES (although it did have higher resolution modes), and that was really a TV limitation.

Meanwhile, Pac-man was 288x224 in the arcade.

So none of the Pacman clones on the PC would ever look 'right', and even the Pacman games on the NES that were made by Namco didn't look right either. There were always hacks like giant characters because the tiles for the map were smaller, or you'd get a scrolling world (Gameboy, Tengen versions), other kinds of distortion, non-original maps...it was all just weird and frustrating when you're trying to play the 'arcade' game at home.

But after learning the details of the machines, how sprites worked (and then coming to the conclusion that they just didn't have any other choice), was such a huge 'a-ha!' moment for me. Let's not even get into the fact that pixels aren't square on those resolutions on the PC.

And then it became almost an instant reaction when I'd see a Pacman port or clone, and try to figure out what size the world was, what size the tiles were, what size the sprites were....

NobodyNada5• 2 years ago

The NES was actually 256x240, using all 240 lines of an NTSC field, but many TVs would cut off part of the picture on the top and bottom -- thus, 256x224 was the "usable" space that was safe on most TVs. But this overscan was inconsistent from one TV to another, and modern TVs and emulators will usually show you all 240 lines.

The SNES's vertical resolution was configurable to either 224 or 240 lines, as the article mentions. Most games stuck with 224, as the longer vertical blanking interval gives you more time to transfer graphics to the PPU.

mrandish4• 2 years ago

256x224 was the "usable" space that was safe on most TVs

Adding further complication, although most arcade cabinet games also used 15Khz CRTs similar to de-cased televisions, since all the cabinets were being assembled by the manufacturer using CRTs they specified, designers could take some liberties with varying the resolution, frame rate, scan lines and scanning frequency of the video signal generated by their game's hardware circuit board. Being analog devices, CRTs of this era were generally tolerant of such variations within specified ranges since they had to sync up with inputs from disparate over-the-air television channels, cable boxes, VCRs or even live cameras. This allowed arcade hardware designers to optimize their circuits either for slightly better resolution and frame rates or alternatively reduce them somewhat in cases where their hardware wasn't quite able to generate enough pixels in real-time. For example, the Mortal Kombat 1, 2 and 3 cabinets displayed video at 54 Hz (instead of NTSC-standard 59.94 Hz) enabling higher horizontal resolution. They could also optionally choose to use the entire overscan safe area for active game pixels since they knew the CRT's width and height adjustments could be dialed on the cabinet manufacturing line to ensure the entire active picture area was visible inside the bezels - whereas few consumer TVs exposed all of these adjustments externally to users.

All this subtle variation in classic arcade hardware makes clock-for-clock, line-for-line accurate emulation especially challenging. Fortunately, the emulation community has solved this thorny set of problems with a special version of MAME called GroovyMAME which is designed specifically to generate precisely accurate emulated output signals so these emulated classic arcade games can be perfectly displayed on analog CRTs. This requires using one of the many PC graphics cards with native analog RGB output, which was most graphics cards made up to 2015 - but sadly none since. GroovyMAME works with specially modified Windows graphics card drivers to generate correct signal ranges from the analog output hardware of most off-the-shelf, pre-2015 Radeon and NVidia cards - which are still widely available on eBay ($10-$50).

For arcade preservationists and retro gaming purists, the resulting output to a CRT is sublime perfection, identical to the original cabinet hardware circuit boards, many of which are now dead or dying. This enables creating an emulation arcade cabinet either using a period-correct CRT matching the traits of a certain series of original cabinets, or alternatively, using a special tri-sync or quad-sync analog RGB CRT which is able to display a wide variety of signals in the 15Khz, 25Khz, 31Khz and 38Khz ranges. This is what I have in my dedicated analog CRT cabinet and using GroovyMAME along with a 2015 Radeon GPU it can precisely emulate 99+% of raster CRT arcade cabinets released from 1975 up to the early 2000s accurately (and automatically) recreating hundreds of different native resolutions, frame rates, pixel aspect ratios and scanning frequencies on my cabinet's 25-inch CRT. For more info on GroovyMAME and accurate CRT emulation visit this forum: http://forum.arcadecontrols.com/index.php/board,52.0.html.

postexitus• 2 years ago

Was it? Amiga used 320x200 (NTSC) and 320*256 (PAL) on standard TVs.

progmetaldev2• 2 years ago

I love this comment, because I always saw the different ports as stretched compared to arcade versions, and I honestly had no idea why. That includes even old ports to Atari 8-bit computers, and the various versions, like Ms. Pac-Man and Super Pac-Man.

theandrewbailey2• 2 years ago

59.94Hz is such a weird number. Isn't the power grid running at 30Hz and TVs used to double it?

No power grid I know of runs at 30Hz. North America (where NTSC was designed) and a few other places[0] run at 60Hz.

[0] https://en.wikipedia.org/wiki/Mains_electricity_by_country

fabiensanglard5• 2 years ago

Oops :)! Thank you (fixed now).

ibobev• 2 years ago

Hi Fabien,

A bit unrelated, but the links to your books at your website are no longer working. I tried to connect to you via email for this issue. Is this intentional or you will fix them?

Best regards, Ivan Bobev

beefok• 2 years ago

I love all of your projects, Fabien!

This SNES video analysis one is incredible. I've always had all of this stuff running around in my head for how to explain how weirdly cool video generation for NTSC is, and you have done an incredible job finding a way to do so.

There is yet another reason for the weird frame and horizontal scan rate. When NTSC was originally introduced as a broadcast standard over a single RF modulated signal, the sound carrier and signals were also embedded in the signal as well. [1] Actually, I just found that Wikipedia does a good job of describing this on the NTSC page [2]:

    When a transmitter broadcasts an NTSC signal, it amplitude-modulates a radio-frequency carrier with the NTSC signal just described, while it frequency-modulates a carrier 4.5 MHz higher with the audio signal. If non-linear distortion happens to the broadcast signal, the 3.579545 MHz color carrier may beat with the sound carrier to produce a dot pattern on the screen. To make the resulting pattern less noticeable, designers adjusted the original 15,750 Hz scanline rate down by a factor of 1.001 (0.1%) to match the audio carrier frequency divided by the factor 286, resulting in a field rate of approximately 59.94 Hz.

So yes, yet another difficulty with NTSC -- sound actually splattered visual noise on the screen as well!

[1] https://upload.wikimedia.org/wikipedia/commons/f/fd/Ntsc_cha... The combined spectrum of video, sync, and audio all on a single RF broadcast signal.

[2] Search for 'sound carrier' in https://en.wikipedia.org/wiki/NTSC#Color_encoding

dlcarrier• 2 years ago

It was originally 60 fields per second (30 interlaced frames per second), on black and white TVs.

The highest frequency generated in a black and white TV was the horizontal scan rate, which was a multiple of the frame rate. With the addition of the NTSC color signal, which used a 3.579545 MHz carrier wave, the highest frequency generated in the TV became much higher. To keep the hardware simple, all lower frequencies were still divisors of the highest frequency, now color carrier wave. For the frame rate, it came out to 59.94 fields per second.

masfoobar3• 2 years ago

Trivia: Besides the annoying black band, the game code was also rarely revised to account for the VSYNC which occurred at 50.00697891Hz instead of 60.098Hz. This resulted in game running 17% slower than intended. European gaming was a real dumpster fire. But luckily without the internet we did not know about it.

This one hits home. Although my examples are not specific to the Super Nintendo, it reminded me of the first time I played/watch Sonic the Hedgehog on the Mega Drive (Genesis)

I wasnt impressed with the game. It looked clunky and just felt slower compared to the Master System version. It wasn't until the rise of youtube I realised the difference in speed between the NTSC and PAL is huge. Its not just the speed of the game, but the Music. It sounds horrible on PAL!

Don't get me wrong - I knew about the PAL during the 16-bit, and the need for the "black box" but I didn't realise how much of a difference it was. I am sure the console magazines at the time would say the difference is minor in most games. One of the exception (honesty) was DooM on the SNES. The NTSC version had a bigger screen.

I remember being good at Punch-Out when I was a kid on the NES. I could beat Mr. Dream (or Mike Tyson) in the first round. Of course, I was playing the PAL version. If there was some kind of competition in the USA, I would have been destroyed in the first round! I would have been convinced I was framed!

Past times, right?

badgersnake• 2 years ago

Wow, that sounds like it might explain why my timing is always off playing SNES games on emulators. I grew up playing the PAL versions. I always put it down to lag on modern TVs. Be

bobim2• 2 years ago

This is raising the question if emulators are using PAL timings for EU roms. And do we know if the snes mini is giving the NTSC or PAL experience?

giantrobot• 2 years ago

I would have been convinced I was framerated!

Fixed that for you.

masfoobar3• 2 years ago

You legend.

rob74• 2 years ago

Its not just the speed of the game, but the Music.

I get that the game speed depended on the framerate, but playing music at a frequency reduced by 17% would have sounded really horrible, I don't think they would have gotten away with it. But then again, what do I know... The only system I know a bit about is the Amiga, which had dedicated sound hardware, so I'm pretty sure it was not tied to the video frequency, no idea about other systems.

masfoobar3• 2 years ago

At the time, I would not have figured anything different.. other than I just didn't understand the hype that is the Mega Drive (Genesis)

- The 3-button game controller is horrible, and the D-PAD is worse! - The main mascot game (Sonic the Hedgehog) was clunky with poor music

Of course, if I had experienced the Genesis on an NTSC then my initial view of the machine might have been totally different!

Overall many games I played in the UK - no matter the console... I was happy because I had nothing to compare it to. It's just Sonic the Hedgehog, from memory, was the main one that just felt off. I would not have thought it was a PAL vs NTSC thing.

MBCook4• 2 years ago

I remember being floored when I first heard about speed differences between US (where I grew up) and PAL once I got older.

Seems crazy games would be sold while so different, although it makes perfect sense why the choices were made.

As a kid I had just always assumed Mario was Mario and Sonic was Sonic everywhere. Why would they be different?

Was it the 3D consoles when this finally ended, since rendering and logic were no longer in lockstep most of the time?

giobox• 2 years ago

No, it was a common issue on PS1/PS2 era 3d consoles as well. It wasn't until we got digital connections like HDMI etc that the PAL/NTSC timing issues went away completely. The disconnect between rendering and logic was not so common back then even in 3d titles, there are many games that just run slower in PAL etc.

NobodyNada5• 2 years ago

Super Metroid is a game where the developers did make revisions so that the game would play at the correct speed on PAL consoles. But those tweaks still end up making material differences in gameplay, especially in a speedrunning context, because the slight differences in physics constants and animation timings can make a difference when exploting glitches or race conditions. For example, the slower framerate makes it easier to clip through things, because Samus and her projectiles move more pixels in one frame, so this gate glitch can only be done on PAL: https://www.youtube.com/watch?v=RvyIwtO_qgM

Also, the developers properly adjusted Samus's physics constants and animation timings for the new framerate, but they didn't adjust enemies, cutscenes, or other aspects of the game environment. So Samus moves at the same speed as on NTSC, but the rest of the world moves slower. This means that on PAL you can grab Bombs and escape the room just in time before the door locks, skipping the miniboss fight: https://www.youtube.com/watch?v=R3t8TIIj7IM On the NTSC version, that same skip requires a complicated setup and several dozen frame perfect inputs in a row, and only one person has ever managed to pull it off: https://www.youtube.com/watch?v=jcKUMk5g8Wk

Here's a comparison of the fastest tool-assisted speedruns between NTSC (left) and PAL (right): https://www.youtube.com/watch?v=KD_-thqcB5s Both runs take the same route up until the very end; the NTSC version is faster in almost every single room, but PAL ends up finishing first because the arbitrary-code-execution setups are very different. The NTSC run has to do a very slow sequence of pausing and unpausing to move through a door without activating it, in order to get out of bounds and trigger memory corruption. Whereas on the PAL version, we're able to exploit a race condition in the game's animation system to achive ACE fully inbounds. The race is between a spike's knockback timer and Samus's landing animation; because Samus's timings were revised for PAL but the spike's were not, the timing works out a little differently and the race ends up being exploitable in this context on PAL but not on NTSC.

progmetaldev2• 2 years ago

I absolutely love Super Metroid, but only played it briefly on a Super Nintendo in the 1990's. I've played the NTSC version on emulators for years, and definitely came across timing issues with memory corruption when pausing through a door, but it wasn't even intentional. I always thought developing for different architectures was the only issue developers had to deal with (beyond faster or slower CPU/GPU/RAM), but had no idea that even something like the SNES had issues between NTSC and PAL. This whole comment section has been amazing as someone who loves these classic games, and also enjoys watching speedruns. Gaming got me into software development, even though I never got into that side of development, but your comment and articles like this remind me why development is so interesting (and difficult for the strangest reasons!).

DaoVeles• 2 years ago

My memory might be a little off but in Australia, because of magazines like 'Hyper' we were made aware of the timing differences but there wasn't anything we could do about it. Thanks for the heads up guys! We could have lived in ignorant bliss!

So when the Dreamcast came along, it was the first to offer games that you could switch between 50Hz and 60Hz but only if your TV could handle it. It also meant that with a lot of games that didn't account for this, you could make things easier by switching back to 50Hz. I recall Crazy taxi being much easier at 50Hz.

roflchoppa2• 2 years ago

I wonder how long it takes Fabien to write these up. So many details, so clean.

fabiensanglard5• 2 years ago

It took roughly a whole Sunday from 9am to 9pm.

roflchoppa2• 2 years ago

Damn dude you didn’t have to speed run it. :)

butz• 2 years ago

Hires mode games list is incomplete. I wonder, if it would be possible to quickly detect if game has a CPU instruction to toggle hires mode on?

NobodyNada5• 2 years ago

It's not a dedicated CPU instruction, but rather a write to an MMIO register with a certain bit set. And many games will follow patterns like computing the register value during game logic, saving it in RAM, and copying the value from RAM to the register later during vertical blanking, or they might have code path to enable hires mode that is present but not used if the code came from a library or engine. So in general, the problem is undecidable; and limited to SNES games, probably not very realistic.

The most reliable thing to do is probably to play through the entire game in an emulator (maybe from a TAS movie) and record whether it actually enables high-res mode or not.

Masterjun• 2 years ago

Doesn't this miss the part where the 256x224 (8:7) output resolution gets stretched into a ~4:3 (actually 64:49) image?

The SNES has a dot rate of ~5.37 MHz which is slower than the square pixel rate defined by the ATSC standards of ~6.13 MHz. It's exactly 8/7 slower, so pixels are stretched horizontally by 8/7, causing the 8:7 resolution to be stretched to (8/7)*(8/7)=64/49, which is close to 64:48 = 4:3.

Result in an aspect ratio close to 4:3. This would mean 224*(4/3) = 298 visible dots.

If you consider what I mentioned, the factor would be (4/3)/(8/7) = 7/6, so they would have to choose something closer to 224*(7/6) = 261.33... visible dots. Which is much closer to what they chose with 256.

JoshTriplett2• 2 years ago

How much of the SNES resolution is hard-coded in the console hardware, versus being something the cartridge could drive? Could a cartridge that didn't need to load sprites (e.g. because it had its own coprocessor), and had its own onboard clock, theoretically drive more than 256 horizontal pixels per line?

MBCook4• 2 years ago

It’s all fixed in the PPU, I think.

So with a coprocessor you can render your own frames and put them in memory where the next line’s tiles are going to be pulled from. That is what the SuperFX did, I think.

But in the end you’re still stuck with the limitations of the PPU actually drawing it in pixels and number of colors and such.

JoshTriplett2• 2 years ago

So there's no bypass giving the cartridge access to the video out lines?

Dwedit• 2 years ago

Also check out Rodrigo Copetti's article about SNES architecture. https://www.copetti.org/writings/consoles/super-nintendo/

gary_02• 2 years ago

Are there any emulators that accurately simulate the CRT appearance instead of just drawing the pixels straight to the window? This could be done performantly with a GPU shader. (I recall some emulators having an aesthetic scanline effect you can enable, but that's not the same thing.)

rimunroe• 2 years ago

The last time I looked there were several shaders available for this. CRT-Royale[1] was excellent when I tried it.

[1] https://emulation.gametechwiki.com/index.php/CRT-Royale

gary_02• 2 years ago

Yeah, that's it! And wow, that photo-real screenshot of Super Metroid briefly teleported me back in time 27 years.

It was a painful experience! I don't recommend it.

rminla• 2 years ago

I was hoping there'd be some insight on how this led to the design of such amazingly iconic games :)

I LOVED SNES and was unfortunately never able to convince my immigrant parents to buy me one

Thank you for the writeup. Super interesting

Lammy• 2 years ago

The 8:7 artwork aspect ratio is visible in a few SFC/SNES ports to other platforms, like ROCKMANX3 / Mega Man X3. The PSX/Saturn/PC versions of that game retain the original art unstretched and instead add stage-appropriate pillarbars to pad the 8:7 out to 4:3. Very distracting to play since I'm used to the original ver. Check out some screenshots of the Saturn ver here and you can see it — everything is slightly too skinny: https://segaretro.org/Mega_Man_X3

Waterluvian• 2 years ago

Wow. I had no clue that an input device actually had any sort of control over the drawing of a CRT. I thought you just throw out signal at a certain timing and that’s it.

Also this article was wonderful in the way that it didn’t waste a word. Very concise.

jzer0cool• 2 years ago

"Don't sit too close, move back a bit". Any truth whether sitting too close makes for bad eye sight? Wondering of any effects (e.g. leakage) that goes past the target screen area.

Understanding the design of the the Super Nintendo video system

Related Stories