Can The Amiga CD32 Run Perfect Versions of Capcom CPS1 Games?

[Thorham]

Originally Posted by DamienD

No way can this be done on an A500

But is arcade perfect game play possible?

[Retro-Nerd]

On an A500? Hard to tell. Not even sure if full 50/60fps would be possible. Even with limited colors, reduced resolution and smaller sprites. Maybe, but it wouldn't have the Arcade "feeling".

[touko]

The CPS1 is a more powerfull system than any of these era, except the NG .
Even the base X68k needs to be upgraded to have a chance of running an almost 1:1 version .
So i don't see how an amiga CD32 can do it, you need a least, a lot of RAM,of course with a lot of RAM you can probably do a close version of ghouls'n ghost for exemple, but for FF, it would be more difficult.

[dlfrsilver]

Originally Posted by touko

The CPS1 is a more powerfull system than any of these era, except the NG .
Even the base X68k needs to be upgraded to have a chance of running an almost 1:1 version .
So i don't see how an amiga CD32 can do it, you need a least, a lot of RAM,of course with a lot of RAM you can probably do a close version of ghouls'n ghost for exemple, but for FF, it would be more difficult.

Lately, Ghouls'n'Ghosts has been dismantled by a french guy called Upsilandre.

The game use 512 colors, 1 layer of 256 1 layer of sprite 256 colors.

It requires a 68000@16mhz to run adequately, and a 68030 with 4mb of ram to be exactly like the Arcade (screen & objects refresh).

well the problem that the X68000 overcomes is that the CPS1 or CPS2 games are in files. Much more easier then to use 2mb of ram.

The Arcade version is unified, everything is linked together.

[dlfrsilver]

Originally Posted by Gilbert

OK guys I am not trolling but just challenging and developing some ideas . What do people think about these thoughts?

I would disagree trying to copy the code as a 1 to 1 match. I think it would need to be re-written from he ground up. So that's more what I'm talking about.

The CD32 has some very large sprites and maybe the Copper can help (e.g. multiplexing etc) and then it has the blitter too? It must be able to get very close when pushed to the limits. (yes as people have said a very close replica is possible)

I understand that the CPS1 system at its most powerful would be tough to emulate - but arcade developers don't write optimal code. Some of their code is pretty bad in fact (in terms of efficiency and the CPS1 is a powerful system at moving sprites and so their 68k code wouldn't need to be very efficient at all)) and games like Strider, Ghouls n Ghosts were the very first games made for the system. Also lets be honest the 16 bit Megadrive has great versions of both. G n G was made on a very low capacity cartridge for that system (768k!) so I'm sure it could have been a lot better. Supergrafx version = 1 Megabyte. The CD32 has 2Mb of RAM and a whole CD's worth of capacity - and it can stream graphics data in and out during gameplay

The graphics data is available on the net and there aren't a huge number of animation frames in Ghouls n Ghosts https://www.spriters-resource.com/arcade/ghoulsnghosts/

According to my (legal) ROM image in MAME, the whole arcade game of Ghouls n Ghosts is 4.06 Megabytes. So that means one level's data can fit in memory easily. Much of this will be music data too maybe? (Edit: Strider arcade is 5.31 MB)

Also lets say that the A500 at it's best can do a 50% accurate arcade conversion of Strider. People always say here that the the A1200,CD32 are 2 or 3 times as powerful (without additional ram).... So.....

Something else I wonder - is that some arcade games run at half frame rate (at least the display code, I think the logic code runs at the full frame rate) So Metal Slug and Outrun are both 30fps. And there isn't a huge difference between that and 60fps. Most casual gamers wouldn't notice if you didn't tell them. So if you could recreate a CPS1 game at 30fps - it would still seem a pretty perfect conversion. Is this possible?



Please try. I actually thought (once I learn hardware scrolling) I could easily better the Amiga version of G n G using AMOS or Blitz on my A500. Maybe I couldn't (in reality) but it doesn't seem like a hugely demanding game in terms of resources. It would be very time consuming to do admittedly.

Just for the record, Capcom developers were incredibly talented. They did clean and tidy code for their arcade games.

Just to illustrate, here is the routine used by the X68000 version of Daimakaimura to update on screen just one tile :

https://i.postimg.cc/7w5mLz93/X68000...Subroutine.png

They made use of Move.m and Move.p instructions.

basically, adapting such games would be quite demanding, with recoding of the tile drawing system.

And no way using a CD32 without Fast RAM. the minimum thing is to use a hard drive for dynamic loading.

[Gilbert]

Originally Posted by mcgeezer

You really are an idiot - you do know that don't you?
How's that for trolling?

I am just answering this today because don't have time to answer more but I will be civil to this guy

Just play any arcade game with slowdown - should never happen on an arcade system (Double Dragon, Metal Slug 2 are more extreme examples) but almost every arcade game has slowdown at some point.

Marble Madness arcade game was written in C. And many arcade games after. Because when you have the number of sprites/playfields you have in arcade hardware - then the CPU is mostly just used for game logic, collision detection. Imagine an an A1200 that had 224 sprites, 3 character(tile)-based playfields (or however many CPS1 has of these things). Would you need to write as efficient CPU code to get the same result on screen?

Strider is a good example of a game released with poor QC (wrong music, extra pixels on character animation, messed up transition screens). I think multiple versions were released. The game code it's self is pretty solid tho. But they would have had no need to optimise it to the same extent you would have to on an Amiga to get in running smoothly.

[touko]

Originally Posted by dlfrsilver

Lately, Ghouls'n'Ghosts has been dismantled by a french guy called Upsilandre.

The game use 512 colors, 1 layer of 256 1 layer of sprite 256 colors.

It requires a 68000@16mhz to run adequately, and a 68030 with 4mb of ram to be exactly like the Arcade (screen & objects refresh).

well the problem that the X68000 overcomes is that the CPS1 or CPS2 games are in files. Much more easier then to use 2mb of ram.

The Arcade version is unified, everything is linked together.

It require a faster CPU just because some parts are coded with the ass .
I don't see why it would require a 68030 for GnG, this game is not that demanding in CPU power,a NG could run a 1:1 version of GnG with his 68k @10mhz just perfectly .

[roondar]

Originally Posted by dlfrsilver

Just for the record, Capcom developers were incredibly talented. They did clean and tidy code for their arcade games.

Just to illustrate, here is the routine used by the X68000 version of Daimakaimura to update on screen just one tile :

https://i.postimg.cc/7w5mLz93/X68000...Subroutine.png

They made use of Move.m and Move.p instructions.

basically, adapting such games would be quite demanding, with recoding of the tile drawing system.

And no way using a CD32 without Fast RAM. the minimum thing is to use a hard drive for dynamic loading.

I sincerely hope that routine is not for updating just a single tile, unless you mean updating the GFX content of a tile in RAM rather than just replacing one on the screen with another. If that really is the amount of code needed to change a tile on the screen to a different tile, then the CPS1 is a horribly inefficient design to code for (though obviously still much more powerful than either OCS or AGA).

--
Anyway, I don't think talking about coder talent will do us any good here. There's a world of difference between coding for an Arcade system and coding for a home computer/console and this will lead to different coding styles by default. The simple truth is that home systems tend to have bottlenecks that Arcade systems didn't have, which means coders often had no choice but to either accept poor performance or use all kinds of tricks to get the desired results.

The Amiga in particular is quite hard to code for well. At first glance it appears like a programmers dream with all the custom hardware that is not just powerful for its time but also super flexible. But to get the most out of it is not easy - using the hardware well can be quite arcane and is a rather different way of doing things from most systems out there. The flexibility can become you enemy as there are so many ways to try and get the result you need, but usually only a few that'll work in your particular case. It doesn't help that the system has struggled with chip memory bandwidth problems from day one and doesn't have a tile based graphics mode so it actually needs to update all the pixels on the screen for many of the effects that only require changing an index in memory on systems such as the Sega MD, C64 (or the Arcade hardware).

This degree of specialization needed to get the best results is partly why the demo scene can sometimes do stuff that is so amazing compared to the games - they can control exactly what is needed where and that is usually not the case in games where those pesky players and their self-determination get in the way of your perfect-performance dream

--
By the way, I have no doubts that the CD32 won't be able to do 1:1 ports of most CPS1 games. The CPS1 has some killer sprite and tile layer hardware. If a game uses that stuff well then you can just forget it. Note that this is a separate question for whether or not you can make a reasonably similar version (one that plays similar and looks "close enough"). That question is more complicated and has been discussed before

[Gilbert]

Yes that is too much code for changing a ROM reference to a graphics tile. But I agree Capcom did generally make very well coded (slick) games compared to other arcade developers

I wasn't saying that arcade coders weren't talented - I was just pointing out that arcade games don't need very efficient CPU code to run smoothly and the fact that so many have slowdown - speaks volumes about the algorithms/approach used by the coders. Not sure why that other guy got so angry at me. Is just logical

My point was if you optimise for every cycle. Unroll loops, whatever 68k coders do etc - on the Amiga you can probably gain a lot of free time for drawing stuff. Not to mention the CD32 has a faster and full 32bit processor.

Capcom released a leaflet in the 80s about the CPS1 that implied it had 3 68000 CPUs when it fact it was talking about it's suite of chips. I think some people still think it has 3 68k CPUs when it just has one 68000 chip running at 10Mhz. But yes of course the sprite/playfield hardware is massively powerful.

Also as good as Ghouls n Ghosts is - the SNES version has more colours and playfields. So let's not pretend that more modern hardware doesn't have some advantages that hardware designed in the mid 80s might not have. A SNES is very cheap compared to the X68000

I do agree it would be very hard to make a copy on the CD32 but I'd like to see someone try and use all the tricks of the Amiga in doing so. The big sprites on the CD32 are ideal for the bosses like Shielder. And the long pause before he appears - for loading in graphics data from CD

What do you think about my idea of running the game logic at 60fps and display code at 30fps? (like some arcade games do) That would give a lot more time to draw objects to the screen. Maybe almost twice as much time

[str0m]

Yes the CPS1 was only 10mhz but the CPU is only one part. The Neo Geo was only a 12mhz 68000 but look at what that could throw around ie the gfx hardware makes so much difference. (NG treats everything as sprites I think)

[PortuguesePilot]

Originally Posted by Gilbert

(...) Not to mention the CD32 has a faster and full 32bit processor.

You ought to know this, surely. The Motorola 68EC020 - which is what came inside the CD32 - is NOT a full-32bit CPU. It's a cost-reduced version of the 68020 and "only" has a 24-bit address bus. This may be sort-of irrelevant for how AGA and Akiko were structured and given the system's RAM (2MB), but even so, it's not a full-32bit CPU.

[Thorham]

Instead of going for a 100% perfect conversion, why not make the game look nice and make the game play 100% arcade perfect? Seems a bit more reasonable.

[PortuguesePilot]

Originally Posted by Thorham

Instead of going for a 100% perfect conversion, why not make the game look nice and make the game play 100% arcade perfect? Seems a bit more reasonable.

It is. And I whish this logic was used more often. It's the single reason why some games on the Spectrum and C64 are more enjoyable than their ST or Amiga's counterparts. They focused on gameplay and not on looks. But we are very visual animals. We care a lot about graphics, and surely you know how it was back in the day when we counted how many colours were on screen and the sheer number was a selling point. On an attempt to give impressive graphics that the 16bit systems made possible, some games became almost unplayable. ZZKJ, usually such a talented programmer, was guilty of that mistake on the Amiga's version of Power Drift, for instance, which looks very nice but runs like a dead horse. But hey! these pretty but unplayable games looked damn nice on the magazine photos or on the game's box cover, and on a time period that preceded the internet, those used to sell a lot. As an ex-Spectrum owner, I still remember that some later games used the images of the Amiga version on their boxes, with tiny lettering saying "not the actual Spectrum version" below, precisely because salesmen knew that humans are very visual animals...

[touko]

Originally Posted by PortuguesePilot

You ought to know this, surely. The Motorola 68EC020 - which is what came inside the CD32 - is NOT a full-32bit CPU. It's a cost-reduced version of the 68020 and "only" has a 24-bit address bus. This may be sort-of irrelevant for how AGA and Akiko were structured and given the system's RAM (2MB), but even so, it's not a full-32bit CPU.

No, the CPU is a true 32 bits, the address bus only determines how many quantity of RAM you can address, what matters are the ALU and the data bus, which are 32 bits.

[PortuguesePilot]

Originally Posted by touko

No, the CPU is a true 32 bits, the address bus only determines how many quantity of RAM you can address, what matters are the ALU and the data bus, which are 32 bits.

The matter was not if it was a true or false 32bit but rather a FULL 32bit and that it is not. There's a reason why Motorola made two chips with different naming. The 68020 is a FULL 32bit with a 32bit address bus and the 68EC020 is a crippled, mangled, cost-reduced version with a 24bit address bus and - thus - NOT a full 32bit chip. I said in my previous post that it may be irrelevant for the Amiga itself, but people shouldn't go on spreading false information in spite of that.

[roondar]

It's obviously correct to say that the 68EC020 has a 24 bit address bus and not a 32 bit one. However, that doesn't make it a "crippled, mangled" version though. Cost reduced, yes. But crippled and mangled makes it sound like a bad CPU and that's not a fair assessment IMHO, a 24 bit address bus allows for 16MB.

Remember that in 1992, 16MB of RAM was still a huge amount. Many PC motherboards only supported about that much and almost no one actually had that much memory in the consumer or even business space. Only highly specialized professional use required that much (or more). It wasn't until around 1995/1996 that system requirements started creeping up towards 16MB required (Windows 95 for instance recommends 8MB of RAM be available, Windows 98 requires 16MB).

[nobody]

The CD32 could get close to Megadrive. Not arcade perfect but close. Having 2 layers with 16 colors is identical to Megadrive, probably a bit better colors too due to larget palette. But it also requires a big team, having arcade assets etc. The Megadrive didn't have bedroom coders doing mega games. Talking about some large teams, 10 to 20 persons. And money, they wouldn't do it for free.

[roondar]

The number of colours is a bit of a hard one to compare actually. While it's true that the Mega Drive supported 64 colours (in 4 palettes of 16 each)* and the CD32's Dual Playfield limited it to 48 (2 layers of 16, plus 16 for sprites)* the real difference is often somewhat smaller than it might appear for dual layer screens. The basic reason for this being that the Mega Drive's palette are a per-tile limit so they can't be fully freely used, while the CD32 has no such limitation.

Overall, you'd expect the Mega Drive to have somewhat more colourful graphics (though with a far more limited palette) for dual layer games if they're converted to Dual Playfield. For games that only use a single layer the CD32 should be more capable as it's not limited to 4x16 colours, but anywhere from 2 to 256.

Fun fact, a 64 colour single layer Amiga game can actually be impossible to correctly represent on the Mega Drive if the combination of colours used in any 8x8 tile requires more than 31 colours (the MD can use it's two layers on top of each other to get to 31 colours per tile).

*) Both systems lose 1 colour per layer/palette for transparency

[nobody]

Originally Posted by PortuguesePilot

It is. And I whish this logic was used more often. It's the single reason why some games on the Spectrum and C64 are more enjoyable than their ST or Amiga's counterparts. They focused on gameplay and not on looks.

The spectrum had an advantage that used only tiny memory for screen like 1k or 2k (don't remember) so it could do a very fast game (z80 is fast) and the developers focus on gameplay mechanics. Yeah primitive graphics but really fast. Faster than Atari ST 32k screen.

[touko]

Originally Posted by PortuguesePilot

The matter was not if it was a true or false 32bit but rather a FULL 32bit and that it is not. There's a reason why Motorola made two chips with different naming. The 68020 is a FULL 32bit with a 32bit address bus and the 68EC020 is a crippled, mangled, cost-reduced version with a 24bit address bus and - thus - NOT a full 32bit chip. I said in my previous post that it may be irrelevant for the Amiga itself, but people shouldn't go on spreading false information in spite of that.

In a coder POV, it's a 32 bits CPU which can only address 24 bits of memory, his CPU power is exactly the same than a 68020 with a 32bits adresses bus .The 68k also have a 24bits address bus, but still a 16 bits CPU
And like roondar said, 16 Mb was more than enought for a home computer.