Hardware pushed to the limits according to Sega-16 members

[Black_Tiger]

Out of the more "advanced" ones, I think only Fastest 1 scrolls the track at 60 fps (the cars animate at 30 fps IIRC) but it has some noticeable trade offs:
- The cars look pretty small (it looks like an 8-bit game in screenshots).
- Even in split screen mode there's a huge chunk of the screen which is filled with a black border and lap time info.
- Track side is pretty sparse is terms of sprites and details in general.
- There are "hills" in the tracks but the elevations are very slight and the draw distance is somewhat limited.
All in all, I like how Human managed to pull off a game with pretty much no slowdown in split screen mode back in 1991.

That's because the PC Engine version it's a port of has 3-way splitscreen. Hence the title "Triple Battle". I guess they were more concerned with using a pun for the title of the Genesis/MD version.

F1 Triple Battle may not animate the track at 60fps, which seems to be more a result of poor dev kits/skills than a hardware limitation, but it did pull off 3-way racing back in 1989.

[Kamahl]

Speaking of good FM, the default patches on the YM2413 are rather nice, and made it often sound better than the superior YM3812.

Illusion City on the MSX must push the YM2413 to the limit, sounds better than many genesis games.

[Barone]

Raster roads do play to the Amiga strengths since bitmap graphics allow to easily alter the image on a per line basis, but the MegaDrive with its row scroll tables can do the same by simply sliding each line of the desired amount. It also has a very large number of sprites available to display cars and roadside objects and I think that the dual layer architecture could probably be leveraged as well.
The Amiga on the other hand must refresh the moving parts of the bitmap screen every frame. Lotus for example does redraw the road every frame and must then blit all objects on top of it one after another at 50 FPS. The fact that it manages to do so tells me that the MegaDrive should be able to do much better but clearly the proof would be in the pudding.

Some more info about the Lotus games (yes, I'm aware you were partaking in this discussion):

Sout's Amiga Lotus games were something of a masterclass in how to use the blitter incredibly efficiently - the unparalleled smoothness of the first two games came from actually drawing two frames ahead using the blitter*, and copying the pre-drawn frame ahead (in the VBL) into the display buffer. That said, he got it running at a decent enough clip on the ST too, which has no real graphics acceleration hardware. I'd wager than on the MD, drawing the cars and roadside objects was simply a case of using the sprite capabilities as they came, which allowed him time to work out how to get the road to update as smoothly as possible.

[* - And as I understand it, that's drawing pretty much everything with the blitter - player's car, opponent cars, roadside objects, sky, background and the road itself, which to my mind is a pretty good point to bring up when anyone calls blitter objects slow...]

http://www.lemonamiga.com/forum/view...=120920#120920

So, if that info holds any water, it makes sense. 'Cause the MD's CPU is just a bit faster version of Amiga 500's CPU; so things seem to point to the original rendering tricks not being so easy to be translated to the MD hardware in an optimized manner.

The MD version, at least on the surface, seems to do proper use of the DMA transfers; even shutting down the display to speed them up (something that many MD games don't do).
And actually there's very little VRAM updates during the race (Lotus/Top Gear games have CPU cars being clones of the players' cars, which simplifies things a lot in terms of ROM/RAM/VRAM usage; unlike OutRun).

Interesting. This is definitely surprising to me, a cursory reading of the specs of the VDP led me to think that the machine should be able to shine in this domain but I do not exclude the possibility that I am overlooking something.

Being limited to 64KB of video memory (with about usually 25% of that going to tables) with streaming not being particularly fast (there's only a short period of time to stream in new graphics, giving only a few KBs per frame) is probably not helping matters. These games are notoriously memory hungry because of all the prescaled sprites, and on top of that ROM being expensive so you wanted to compress as much as possible.

There's still a lot of details that simply don't add up IMO:
- Super Hang-On: Seems to be a 1:1 port of the arcade in terms of "AI"/traffic waves, course rendering and physics. Lots of multiplication involved in the rendering of the "hills" (something that Lotus' implementation seems to have cleverly simplified a lot), even in the lateral movement of the bikes (super finicky details like you running centiseconds faster when using the insider lines of the track were preserved from the arcade), grip calculation, etc.
Very few and small updates on the VRAM during the race on a same track; only special effects like the skidding one are loaded on-the-fly and they do cause some extra slowdown when you're in heavy traffic (the harder difficulty settings use heavier traffic during the races).
Full use of the 320 x 224 resolution, with the track contours having less jaggies than most of the other games of the same type (also on par with the arcade version); no noticeable flickering and backgrounds with parallax. 20 fps for the most part.
In this one, the main cause of the low frame rate really seems to be the calcs-heavy nature of the original engine which was ported to the Mega Drive.

- OutRun: Seems to use simplified course rendering and physics when compared to the arcade. I'm not sure about the AI/Traffic faithfulness to the arcade version but it seems a bit simplified too.
There's a good amount of VRAM updates during the race (using DMA), mostly related to the CPU vehicles. Sik already pointed it has a measurable performance cost to load those sprites and there's noticeable slowdown especially when you're overtaking cars and they are about to vanish from the screen; and IDK what's the exact cause of that: collision detection routines being triggered for no good reason or new sprites being loaded/decompressed to the VRAM/decompressed to the RAM or some other cause. Full use of the 320 x 224 resolution, noticeable flickering with several vehicles on screen; some backgrounds have parallax, some others don't. 15-20 fps for the most part.
Here things are a bit fuzzier to me, but we could still suppose that the RAM/VRAM updates and possibly on-the-fly decompression are causing the low frame rate.

- Lotus ports: 256x224 resolution, with noticeable black borders, very little VRAM updates during the race (using DMA); the engine used seems to be a very faithful port of the Amiga one, but running round 20 fps most of the time.
This one is hard to understand why it can't even reach 30 fps, for an example.

- RoadBlasters: There's a TON of VRAM updates going on during the races and it seems to make NO use of DMA (by the way, it seems to be a trademark of Sterling Silver Software), the game scrolls at 60 fps with no noticeable slowdown.
This one makes it hard to comprehend how it do NOT struggle with so much VRAM updates going on (and, apparently, making no use of DMA on top of that), even considering its big ass HUD and NES bad-looking backgrounds.
In terms of pure calculation, though, it seems to be the most simple of these games: collision causes explosion of the player's car (the other vehicles are not moved either, they explode or stay intact), traffic is really just streaming lines of vehicles come at your statically, no hills or chicanes, very simplified representation of grip, no speed variation according to the terrain type or elevation (since there's no elevation changes whatsoever), etc.

That's because the PC Engine version it's a port of has 3-way splitscreen. Hence the title "Triple Battle". I guess they were more concerned with using a pun for the title of the Genesis/MD version.
F1 Triple Battle may not animate the track at 60fps, which seems to be more a result of poor dev kits/skills than a hardware limitation, but it did pull off 3-way racing back in 1989.

I was aware of that, but it's still a letdown compared to the 3P option of the PCE version or full/bigger screen modes.
And, yes, both F1 Triple Battle and Fastest 1 are amazing IMO. They're soooooooo far ahead of MD's Super Monaco GP (I & II) it's kinda sad.

[gamevet]

http://www.lemonamiga.com/forum/view...=120920#120920

So, if that info holds any water, it makes sense. 'Cause the MD's CPU is just a bit faster version of Amiga 500's CPU; so things seem to point to the original rendering tricks not being so easy to be translated to the MD hardware in an optimized manner.

The difference is like 500Khz, which is pretty much less than 5%.

[Flygon]

Hell, that's an error of margin, akin to PAL vs NTSC.

[Barone]

That's why I said "a bit". But in reality, I think the CPU efficiency in the Amiga 500 is hurt by the lack of fast RAM.
Anyway, the point is: CPU-wise, the difference is either negligible or in favor of the MD while the performance of the game is much better on the Amiga.

[Kamahl]

Not sure this could be called pushed to the limit, but this game is probably the best example of what the amiga should look like graphically all the time.
The forest stage (19:00) is particularly pretty.

[Barone]

You must spread some Reputation around before giving it to Kamahl again.

Really nice looking game, Kamahl. And I've never heard about it before.

IMO this one good example of art style and color use which takes advantage of the 12-bit palette of the Amiga and which wouldn't look that good on the Mega Drive mainly because of its 9-bit palette (despite the higher number of simultaneous on screen colors you can usually push).
The Mansion stage (around 42:00), for an example, it doesn't look anything special with all those grey-ish tones but good luck trying to port it to the Mega Drive without completely breaking the intended atmosphere, lighting and depth perception.


Talking about how much limitation the MD's 9-bit palette brings to the table... The other day I realized that several stages of Super Hand-On are "hiding" the striped pavement they had in the arcade due to repeated colors in their palette setup.
So I tried to toy with the road pavement and track side sand a little bit. I ended up getting depressed in the process:


Mega Drive (Original)


Arcade


Mega Drive (Alt. #1)


Mega Drive (Alt #2)



Of course, I didn't stress all the possibilities and the fact that I didn't even touch the rest of the scene contributes for a drab looking image overall; BUT those big tone steps between the pavement stripes are actually the least abysmal ones you can find in the MD's 9-bit palette without trying to combine completely different tones (which results in an even shittier looking pavement).

I know the 4 palettes limitation is also huge and makes tons of games harder to port and make good use of color of their original art, but, seriously, if I had to choose between an improvement to 8 palettes or an improvement to 12-bit palette I'd go with the latter in a heartbeat.

[MasterGenesis]

I finally found a game that looks just as good on the Genesis as it does in the arcade (not counting games like Pong or Frogger).

[Kamahl]

9bit to 12bit is a really big jump in colour palette. Much bigger than from 12bit to 15bit. I've argued many times that an improvement to the Master Palette would have helped the Mega Drive a lot, even with it sticking to just 4 palettes. You'd be able to select colours that not only mix better, but can be more effectively reused.

You can do a lot with minimal colours as long as they are good, compare this hagane pic vs the same one using only dawnbringer's fixed 16 colour palette (which is better for colourful games, and I didn't even use any dithering).

[Barone]

9bit to 12bit is a really big jump in colour palette. Much bigger than from 12bit to 15bit. I've argued many times that an improvement to the Master Palette would have helped the Mega Drive a lot, even with it sticking to just 4 palettes. You'd be able to select colours that not only mix better, but can be more effectively reused.

Yep.
There's also the thing about how much it limits your artistic possibilities. You can see that it's pretty much impossible to convert that dust-filled art from the original Golden Axe to the MD's master palette; there's really no workaround for that.

In general, games which use color brightness to create depth sensation are very, very difficult to accommodate. And that's one of the key elements that defined 4th gen 2D when compared to 3rd gen stuff. QuackShot is one of the very few games which managed to pull it off on the Mega Drive; artists might have worked like crazy to come up with background designs which would "click" with such limited resources.

Even the 32X games were hurt by the 9-bit palette IMO, 'cause the backgrounds generated by the MD's VDP simply don't have the needed colors to match the foreground tiles; the gradients are too rough, they simply don't fit.
32X's Virtua Fighter and Virtua Racing (and those are NOT colorful games) backgrounds always looked broken to me, even on real hardware and using CRT TVs; they were too saturated and rough, they would stand out when they shouldn't. They only made "sense" when I upgraded my setup to use a Trinitron Wega that has that built-in filter which smooths out tone transitions; then the sky got proper bright, the clouds had proper blue-ish gradients, etc.

32X's MKII is another one plagued by the roughness of 9-bit Master Palette backgrounds. Compare SNES' Forest stage to 32X's; you'll ask for someone to shoot you in the face.

You can do a lot with minimal colours as long as they are good, compare this hagane pic vs the same one using only dawnbringer's fixed 16 colour palette (which is better for colourful games, and I didn't even use any dithering).

It's really impressive.

I think even dithering would look far less nasty if you had better tone choices to mix up.
Proper levels of grey and brown would do wonders for many, many games.

[Black_Tiger]

The better I get at pixelart, the fewer unique colors I end up using, while packing in lots of detail and shading using a 9-bit palette. 12 and 15-bit palettes are more useful for lazy conversions from hardware using overly subtle shading, but 9-bit is perfect for the resolution and overall detail possible within the combined restrictions of 16-bit consoles. Whether you have a 9, 12 or 15-bit palette to choose from, the sub-palettes are still the most important factor.

All of the amazing color hacks for Genesis games show how much better a game can look simply by swapping around colors within an existing set of palettes. Imagine what those games would look like if the color hacking artists were in charge of the coloring while the games were being built from the ground up.

Even without re-working the pixelart to line up shading for a 9-bit palette and cater to the strengths of the target hardware (increasing rersolution, detail, shading, etc), what is possible in a lazy straight-conversion is dramatically more than what people believe.













The problem is that everyone's perspective of what 9-bit color is capable of is skewed by Mega Drive and PC Engine games not even touching their potential.

There's also the thing about how much it limits your artistic possibilities. You can see that it's pretty much impossible to convert that dust-filled art from the original Golden Axe to the MD's master palette; there's really no workaround for that.

Golden Axe/System-16 hardware is one of the best examples of how sub-palettes are more important than a huge master palette. The arcade's backgrounds are full of fugly color clashing combos (looks like the hardware can only do 8-color tiles), with things like browns in the middle of a blue gradient. The extremes in high-number monocromaticly colored elements mixed together with multiple clashing palettes for other elements, lots of chunks of mismatched alternately colored tiles and minimal color variety and then contrasted by 16-color sprites (seemingly from different hardware) slapped on top... creates a dull and bland looking picture color-wise.

Lots of arcade games using various hardware have these and other problems, but at the the time our minds filled in the gaps and we all remember them with rose-colored glasses. A quality conversion of the Golden Axe arcade to any hardware should have first addressed all of the problems of the source material and then reworked things to suite the target hardware. The Genesis/Mega Drive could do a much better looking take of Golden Axe than the arcade, if it was done right with a large rom or on Sega-CD.

[TrekkiesUnite118]

If I remember correctly though, didn't tomaitheois mention once that the 9-bit RGB palettes that the PC-Engine and Genesis use are not standard 9-bit RGB and therefore are even more limited?

[Barone]

Whether you have a 9, 12 or 15-bit palette to choose from, the sub-palettes are still the most important factor.

It depends on what you're trying to do.
Like I said above, sometimes you simply need the colors you don't have and no number of sub-palettes will change that.
With only 4 sub-palettes, what makes MD's Super Hang-On looks different from the arcade original is really the Master Palette 'cause for the most part there you still have free slots in the sub-palettes.

All of the amazing color hacks for Genesis games show how much better a game can look simply by swapping around colors within an existing set of palettes. Imagine what those games would look like if the color hacking artists were in charge of the coloring while the games were being built from the ground up.

I've talked to Pyron a lot about this matter and he has pretty much the same opinion I have.
Most of the time, the most frustrating results you get in those hacks is thanks to the 9-bit Master Palette and not to the also important lack of sub-palettes. Most of the time you can figure out a re-arrangement of the sub-palettes which would works great and, at the same time, you'll fail to find a set of unique colors to reproduce what you need.
Many Mega Drive games (I can't talk about the PCE because I've never hacked any of its games) have the sub-palettes problem *extremely* exacerbated by the re-use of palettes across different levels or screens where they could have been easily exchanged by more fitting ones. Games like Midnight Resistance use the same palette for most backgrounds in the game and don't even make use of the whole four sub-palettes.

Even without re-working the pixelart to line up shading for a 9-bit palette and cater to the strengths of the target hardware (increasing rersolution, detail, shading, etc), what is possible in a lazy straight-conversion is dramatically more than what people believe.

I'm aware of that.
But the point I was discussing with Kamahl was really more about preserving the original art direction of the games; not about doing things form scratch or deeply re-working them to play the hardware strengths.

The problem is that everyone's perspective of what 9-bit color is capable of is skewed by Mega Drive and PC Engine games not even touching their potential.

I think you're exaggerating and generalizing.

The art you provided as example looks great but it's more of the same in the context of the 9-bit color issue we're discussing: high contrast and darker artwork.

Golden Axe/System-16 hardware is one of the best examples of how sub-palettes are more important than a huge master palette. The arcade's backgrounds are full of fugly color clashing combos (looks like the hardware can only do 8-color tiles), with things like browns in the middle of a blue gradient. The extremes in high-number monocromaticly colored elements mixed together with multiple clashing palettes for other elements, lots of chunks of mismatched alternately colored tiles and minimal color variety and then contrasted by 16-color sprites (seemingly from different hardware) slapped on top... creates a dull and bland looking picture color-wise.

Lots of arcade games using various hardware have these and other problems, but at the the time our minds filled in the gaps and we all remember them with rose-colored glasses. A quality conversion of the Golden Axe arcade to any hardware should have first addressed all of the problems of the source material and then reworked things to suite the target hardware. The Genesis/Mega Drive could do a much better looking take of Golden Axe than the arcade, if it was done right with a large rom or on Sega-CD.

Of course arcade games usually have a lot of waste in terms of sub-palettes use, as they also have in terms of processing and memory usage.
But then again, we were discussing what could be done with 9-bit and 12-bit coloring in terms of porting and preserving the art of those games.



My main point here was: given an screenshot like the one below (or the SHO one), I think you can reproduce it far more faithfully using 4 sub-palettes of a 12-bit master palette than 8 sub-palettes of a 9-bit master palette.

but 9-bit is perfect for the resolution and overall detail possible within the combined restrictions of 16-bit consoles.

I wouldn't say "perfect" but I surely agree that subtle gradients wouldn't even be noticeable in the most common CRT TVs back in the days, especially considering composhit output.

[Kamahl]

About the PCE. Tomaitheous has a blog post about the PCE master palette. Its 9bit but not the rgb values you'd expect, the actual colors are softer. It makes one of the rpgs on the system look substantially better than what it looks on emulators. I cant find it now on my phone but i'll look it up later.

Edit: found it https://pcedev.wordpress.com/2010/12...e-color-space/