Charting NDisc from Sony to Philips to SuperFX to NEC
Even
though this site is supposed to be a 1990s style tribute to StarFox,
we need to get some SNES CD-ROM history sorted out. It’s too
easy to get confused with the stupid Sony debacle, the Philips
drive, the SuperFX, and the “32-bit”ness.
In
facial media youtube-r speak, let’s “EXPLAIN” this
“SHOCKING” and “INSANE” history!
The main objectives of this page are to:
Explain the differences between 3 known Super NES CD-ROM accessories. And a rumored fourth!
Explain why Nintendo dumping Sony was NOT a mistake, and should have happened earlier.
Give context why this was important. The snazzy new CD-ROM hardware could have been a step backward from cartridges.
There's actually a fantastic Kotaku article that documents this saga very well except for a couple of missing elements. And another on nsider.
Back in the 1990s there were no news websites to scour for this information. You had to dig hard between the magazines, writing to Nintendo directly, hunting through patents, and business news electronic periodical searches .
If one pieces together SNES CD-ROM information in sequence, one gets a rare treat in this industry -- outsiders able to witness the labored development of a product as it changes to meet evolving needs of the marketplace.
Products evolve and change frequently before coming to market, and not always in the correct direction. For example, in recent years (circa 2019) Jez San (of Argonaut fame) spoke of his efforts to build a full-color VR system that was canned in favor of Virtual Boy.
So what led up to
the CD-ROM era?
In 1982 Sony and Philips introduced the
"Compact Disc", so termed because it was much smaller than
laserdisc. Although it was envisioned as a general data storage
medium from day one, throughout the 1980s "CD" was more
synonymous with the Red Book audio format (audio CDs) largely due to
the format's cost.
As the
1990s approached the feasibility of marrying optical media and games
seemed imminent. The PC gaming crowd even invented a new marketing
buzzword, "multimedia"! It was a generic word
vaguely meaning "a combination of video, pictures, voice...."
but really just meaning "we now have a bulk storage device and
now we can make video games, too".
Reality was another story. What value
did these expensive add-ons even provide?
1. Cost, although coming down, was still a huge factor.
The 1991 Sears WishBook catalog, page 506, listed the original TurboGrafx-16 CD-ROM2 attachment (a barebones drive with no games and still requires the TurboGrafx-16 to attach it to) for $299.99.
Sega CD (again, an accessory that requires investment in a Sega Genesis) also cost US$300 when it was first introduced.
Nintendo promised a $200 CD-ROM accessory with superior capability. The Electronic Gaming Monthly editorial rabble always laughed and lambasted this as impossible. “A 32-Bit system wih Philips CD-I compatibility for $200 by the Winter of 1993-1994? Come on Nintendo! How in the world are you going to do that?” (Electronic Gaming Monthly, March 1993, Volume 6, Issue 3, page 16). They were unaware the declining cost of these components. That low-power 32-bit chip was already on the market In 1991. Regardless, even a $200 accessory is still twice the cost of a bare-bones Super NES ($99.99) in 1993.
2. When you suddenly have bulk capacity, at costs multiplying that of the original console, what do you do with it?
You can copy lots of small games into the space. This is called "shovelware", which no one wants to spend hundreds of dollars on a dedicated unit for.
You can have a normal size game and tack on some Red Book CD audio tracks. In other words, enhanced audio, but no improvement elsewhere. In fact, depending on the amount of RAM you have, this can actually downgrade the rest of the game versus a huge ROM cartridge.
You can fill the space with video, except:
You can't fit enormous amounts of video even with compression. (Even VCD format Hollywood movie fills 2 discs!)
You only have 150KBytes/second you can pull from the disc, for both audio and video. About 1 megabit/second.
You don't have enough horsepower to run expensive MPEG1 algorithms on video. (Even the Philips CD-I and 3DO consoles required expensive MPEG hardware add-ons to support MPEG video. Again, talk about a segmented market when you need an add-on for an already obscure console.) Most video compression involved some form of lightweight "cinepak" style routines and try to fit inside some small window, often cutting as many frames as possible and as many colors as possible to fit in the confines of the console in question.
Video production is expensive. How does it benefit your gameplay?
Or you can spend lots of money making an even bigger game, breaking it up into small chunks you can load into RAM. This means your customers have to own the base console, an expensive CD attachment, and still be willing to foot the bill for an expensive CD-ROM game. Who cares if the manufacturing cost of the disc is small if you still have many millions of dollars in development cost to recoup? And again, now you have a smaller audience of CD-ROM customers to divide that cost between.
3. What good is a huge bulk storage device that you can only sip data through a straw? To be useful, suddenly you need more RAM. And if you can afford enough RAM, by then you can also afford a bigger cartridge! So then you're chasing your tail again. Is your CD-ROM really an upgrade over the cartridge?
Your game console can only manipulate so much data per frame. There's a limit to how big your game level database will be and how many such levels you can afford to produce.
Cartridge ROM grows exponentially with time, doubling every 18 months with Moore's Law. But once the CD-ROM and its RAM are finalized they are set in stone and do not grow at all unless you left room for a RAM upgrade.
If your new cartridge is sufficiently bigger than your old CD-ROM RAM, eventually you get to a point where the cartridge game can throw more data per level than you can fit in RAM!
We've seen this many times!
One example was Street Fighter 2 on the TurboGrafx-16/PC-Engine. The game and its 12 combatants took 20 megabits of cartridge space. To load up 2 such combatants from CD-ROM would take more RAM than the CD system had!
Electronic Gaming Monthly (April 1993, volume 6, issue 4, page 50) ran an article on this conundrum. Capcom considered releasing the game as a cartridge + CD-ROM combination, where the game ran on cartridge and streamed background music off CD. Obviously this never happened.
This scenario led to the TurboDuo "Arcade Card", which was a RAM size upgrade. Unsurprisingly it was useful for porting other such one-on-one tournament fighting games like SNK’s Art of Fighting.
In later years we saw it again with Mortal Kombat 3 on the Sony Playstation. Shang Tsung's ability to morph into other combatants required the game to pause just to load that new set of combatant data into RAM.
The Neo-Geo CD in 1994 had more RAM than the N64 had in 1996. Its entire purpose was to load a usable amount of CD-ROM game data into a virtual cartridge. It was ample enough to hold the early generation Neo-Geo cartridge games (Magician Lord, Mutation Nation, Last Resort, etc.) entirely in RAM (except for redbook streamed audio remixes). But by this point all new large tournament fighting games were hundreds of megabits.
Was it really worth spending several hundred dollars for a CD-only system just to play older games well? Those same older games that by this point would cost much less to re-manufacture as new cartridges with fewer/denser ROM chips in the first place!
And meanwhile now all of your new games need to be developed with a full cartridge in mind (for MVS arcade and AES system owners) and still be somehow able to fit enough useful chunks into the RAM of the CD system? Congratulations --- you just made game development harder, not easier.
Nintendo Power volume 35 seemed as eager as everyone. "How would you like to go one-on-one with Michael Jordan?" they wrote on page 71. But it wasn't long before Nintendo soured on what they saw. 9 months later, in volume 44, on Super Power Club extra, page 15, they spend paragraphs talking about the need to improve the gaming experience over "cartridge games on a disc".
Unbeknownst to most readers the SNES CD effort had gone through at least 3 major product revisions by that point! It was a major effort not only full of corporate politics (that story is beaten to death now) but also to make a product stand out not only against the competition but also against its cartridge bretheren.
CONCLUSION: An expensive CD-ROM accessory would have to be sufficiently capable to justify its own existence for the duration of the console's lifetime.
In the early 1990s most sources were in printed magazines that ranged from "highly accurate" to "highly speculative" to "highly opinionated". Worse, they gathered juicy bits of news from their own undocumented sources (frequently Japanese magazine correspondents) and presented it as their own article. And while these articles were often accurate (e.g. Electronic Gaming Monthly's final block diagram (March 1993, Volume 6, Issue 3, page 52) matches the 1995 patent), they'd be juxtaposed in the same issue with articles by Quarterman that read like a foreign language rumor mill. Not unlike the Internet of today! Some things never change.
One of the greatest sources in recent years was the discovery and teardown of the prototype Sony unit.
Electronic patent searches by website were a
novel thing by the mid 1990s, long after the CD unit was shelved. And
while patents can be controversial (or sometimes intentionally vague)
in their own right, generally a patent on a physical product requires
some sort of physical product or implementation before it can be
granted.
The patents for the Super FX (#5,357,604) and the final CD-ROM
add-on are both public.
In recent years Nintendo's CD-ROM developer documentation was discovered. It confirmed the EGM block diagram and details the XA format and its ADPCM compression. The most juicy details (how to control the data flow from CD-ROM to HANDS) are still missing.
There is a supreme irony that while the video game industry struggled with CD-ROM, libraries around the country used them very effectively as a text-only data archive called InfoTrac. One of the greatest sources on the Super NES CD-ROM product was itself archived on this CD-ROM system. Think of InfoTrac as a self-contained world-wide-web and search engine all in one, except the searches were periodicals listings. Most of the search results were abstract summaries of the text (occasionally whole articles). By academic standards that technically means relying on InfoTrac without finding the original article is a dreaded Secondary Source, but it's about as close to authoritative as you can get these days.
The other irony is that our big source here is from Lehman Brothers, a once large company that imploded spectacularly in the 2008 financial meltdown. So good luck finding the original sources.
A financial analyst named K. Segawa authored a Nintendo company report on Feb 24, 1993 and it was captured in the InfoTrac General BusinessFile October 1990 - October 1993. This one details the 32-bit upgrade.
The same author wrote an earlier company report on September 14, 1992 with a paragraph mentioning launch dates for the previous (pre-32-bit) unit.
I have scans of these InfoTrac summaries here. Click for larger sizes.
There are at least 2 CD-ROM base units (the Sony unit and the newer Philips unit). Strictly speaking, there might be 2 Philips drives. We know average access time improved from 750ms to 700ms, per the Lehman Brothers report.
At the end of the day, the CD-ROM drive is just a CD-ROM drive, with the Philips unit being superior both in speed (it supports 2x speed) and disc format (it also supports XA format, if desired). The Philips unit also had those cute CD caddies with the built-in save RAM, which is smart for a number of reasons. But again, it's all bulk storage. It's important to the success of the product, but alone it doesn't elevate any of the machine's capabilities.
There are at least 3 system cartridges that hold RAM, and sometimes other processors. Under this model, the CD-ROM fills the cartridge up with data, and the SNES sees it as a normal cartridge. It also means you can (theoretically) upgrade the system by upgrading the system cartridge, much as what NEC/TurboTech did for the PC-Engine/TurboGrafx-16 Super CD-ROM2 and Arcade Cartridge upgrades, and much as Nintendo did for the 64DD (which is just another cartridge underneath the Nintendo64).
Most of the system's magic is in the system cartridge. This, above all else, makes or breaks the product. This article is going to demonstrate how pathetic the Sony system cartridge was, and why it had no future.
So let's start with the embarassment of the Sony system cartridge:
Sony Play Station System Cartridge, 1992 |
|
PROOF: |
We have the teardown video showing the contents |
ROM: |
There is a boot ROM (think of it as a "BIOS", or in console terms "IPL") and it holds all the pretty pictures for the splash screen and initializing the system. There's no magic here. Every system cartridge will have some flavor of this. |
RAM: |
There are 2 SNES "WRAM" custom DRAM chips, which means 2 megabits (or 256 KBytes) of RAM to hold data from the CD. (Unclear how this is mapped and whether or not the SNES can read these at 3.57 MHz or 2.68 MHz like its internal RAM is.) This is the first huge disappointment. It proves the Sony unit is more or less just a TurboDuo, scarcely able to buffer up 1 second of read data from the drive, let alone do much with it. |
Backup RAM: |
All game saves (for your entire CD game library) must fit on the battery-backed 64 kilobit (8 KBytes) SRAM inside the system cartridge. This may seem small for RPGs, but it's identical to how the later Neo-Geo CD works and those small save states do fit fine. I've never run out of space on my Neo-Geo CD, but I've also never had to bring it over to a friend's place to share my game save state. To bring a saved game to another location, you need to bring the CD disc and your Sony system cartridge. |
Coprocessors? |
NONE! This is the second big disappointment. Did they really expect the SNES CPU to decode video and run a game at the same time? You could, but then it'd be no different from existing cartridge games like Out of This World. Where's the upgrade? What did you spend all of that money for if you got nothing new in return? |
Audio? |
Red Book CD-DA audio on the side. |
Now we move on to the first generation Philips model, promised to be $200 with serious performance upgrades. This is actually the system cartridge we know the least about because our sources are more indirect. We have to read a little between the lines. You also wouldn't believe the incredulity in Electronic Gaming Monthly editorials (EGM editors always hated on Nintendo back then), but they were publishing specs as they received them. Nintendo was always vague about the "coprocessor" in the system cartridge, but it's painfully obvious that it was indeed a Super FX. What else did they have at their ready disposal that ran at that speed, did 3D, already worked with the system, and had that 3-way bus design? All they had to do is replace the ROM with a RAM that could be filled by what is also obviously an ancestor of the "H.A.N.D.S." chip.
It's worth noting we have some conflicts between these sources:
2 magazine articles (Electronic Gaming Monthly and Nintendo Power) hint at 8+1 == 9 megabits of total RAM
One other Electronic Gaming Monthly article says 4+4+0.25 megabits total RAM.
The Lehman Brothers article indicates a total of 10 megabits of RAM, with no mention of how it was distributed. A logical method would be 4 (HANDS) + 4 (SuperFX "ROM" bus) + 1 (SuperFX "RAM" bus) + 1 (SNES exclusive). A similar split-model SuperFX memory map is well-documented elsewhere, although never really employed in practice due to cost. (Nintendo’s own SNES development manual documents a 16 megabit (SuperFX ROM) + 1 megabit (SuperFX RAM) + 48 megabit (SNES ROM) + 1 megabit (SNES RAM) configuration that was obviously never used.)
Any of these is feasible. The SuperFX will happily take up to 1 megabit of its own work RAM, and could easily take 4, 8, or later on 16 megabits on its ROM bus. It's evident from the EGM schematic that we probably ended up with 4 megabits of HANDS memory (used for audio, CD buffering work, CD packet error correction, or whatever else needed), 4 megabits of memory mapped into the SuperFX's ROM space, and 1 megabit of SRAM mapped into the SuperFX's RAM space instead of 256 kilobits (FYI StarFox has 256 kilobits.). That would match the "9 megabit" articles and make the most sense for decompressing larger 8-bit images. All of this is visible to the SNES as well, provided everything is set up correctly.
Regardless of the exact RAM configuration, this system cartridge is quite nice! In fact, it soundly kicks Sega CD tail. Remember the Sega CD was often called a "downgrade" for the Genesis because it gave the Genesis only 2 megabits of RAM plus 4 megabits hidden behind a second 68000 CPU (so, 6 megabits total, but not all directly visible). Firstly, we instantly benefit from a 2x CD-ROM drive versus the 1x Sega CD. Also, not only is the total RAM quantity bigger than the Sega CD, but the SNES can access all of it with less fuss. We also get a faster coprocessor (SuperFX over 68000).......at the expense of wide development tool support. I must concede it's easier to get tools for a popular Motorola chip than a custom chip. Because the Philips XA format requires ADPCM audio support, undoubtedly this HANDS chip ancestor also had the 4 extra ADPCM sound channels the final version had (hence the 4 megabits local RAM), so we also get a clear sound upgrade beyond just streaming Red Book CD audio.
In fact, the only down side of this configuration was imminent obsolesence. StarFox itself was on the horizon. If SuperFX cartridges were already in the wild, what more could CD-ROM offer beyond bulk storage and extra audio? In fact, I've always suspected this was why Nintendo was so quiet about this system cartridge. The marketing department wouldn't want to detract from either product. StarFox is both a teaser product to whet your appetite for more, and also a competitor.
Philips Unit "SuperFX" System Cartridge, 1992 (anticipated launch: January 1993, pushed back to August 1993) |
|
EVIDENCE 1: |
We have the SuperFX patent itself. #5,357,604, granted finally October 18, 1994. This is evidence, not proof, as patents are intentionally written as vaguely as possible to cover broad use cases. Usage in a CD-ROM environment is specifically detailed in columns 7 and 8. |
EVIDENCE 2: |
We have the first Electronic Gaming Monthly article, from June 1992, issue 35, page 48. This issue is available from Retromags. |
EVIDENCE 3: |
We have the second Electronic Gaming Monthly article. Electronic Games, April 1993, which discusses the upgrade from "16-bit" to "32-bit". This can be downloaded from Retromags. |
EVIDENCE 4: |
We have the Nintendo Power article, volume 35, pages 70-71. This makes no mention of processors, and in one sentence mentions "8 megabits" and in another mentions "9 megabits". Either this was a misprint or they are factoring in the SNES’s own 1 megabit internal WRAM, or they are factoring in an extra megabit for the SNES, or they are factoring in an extra megabit for the Super FX's working RAM space. |
EVIDENCE 5: |
We have the Nintendo Power article, volume 43, page 6. The sudden announcement of a 32-bit system cartridge, claiming that "16-bit" wasn't good enough. That's reverse evidence that we did, in fact, have a middle of the road system cartridge like this one. |
EVIDENCE 6: |
We have the Feb 24, 1993 Lehman Brothers financial analyst article by K. Segawa that specifically mentions the upgrade from "16-bit RISC chips, 8Mbs of DRAM, 2 Mbs of SRAM". So that implies we had something like 10 megabits in this era. The distribution is unclear. 4+4+1+1 would be sensible. (4 for HANDS, 4 for SuperFX, 1 for SuperFX, 1 for SNES) |
ROM: |
There is a boot ROM (think of it as a "BIOS", or in console terms "IPL") and it holds all the pretty pictures for the splash screen and initializing the system. There's no magic here. Every system cartridge will have some flavor of this. |
RAM: |
There is a 4 megabit RAM for each processor, with 4 megabits taking the place of the Super FX ROM, and most likely 1 megabit of RAM for the Super FX to work with. The types of these memories (DRAM, SRAM, PSRAM) are unstated. Normally Super FX RAM is SRAM but it can accept DRAM if needed. |
Coprocessor 1 |
Super FX According to the articles, apparently the full-speed 21.477 MHz flavor as well. |
Coprocessor 2 |
Some early version of HANDS Something has to manage the CD-ROM drive and its XA filesystem, fill up the audio RAM, fill up the Super FX “ROM” bus, and play sounds from audio RAM. Note, however, that in this era HANDS would not need to convert bitmap graphics to bitplanes yet, as the SuperFX already has hardware to do that inside. |
Audio? |
Judging from the EGM schematic, this looks conspicuously like the later HANDS chip (described below), so most likely it had the same 4 on-board ADPCM channels and the 65C02 CPU to control them, sent through the cartridge analog input ports to be mixed with the SNES audio. |
At last, what we presume to be the final system cartridge! And what a beauty!
In total we would have doubled the combined memory of the Sega CD.
We would have a clear improvement in audio over the Sony drive, as now we have the XA filesystem and 4 extra ADPCM channels in addition to the CD audio.
The V810 would have far eclipsed the Sega CD’s Motorola 68000, probably about an order of magnitude on a good day depending on how we hammer the memory bandwidth. It had standard assemblers and debuggers, and was also a compiler-friendly CPU that for the first time would have enabled C language. (No one used C in consoles then, even on the 68000 which also had C compilers.)
In fact, the V810 even had an FPU (floating point math support) which even the later PlayStation CPU did not have, although realistically this is academic as most games were not big floating point math users and the V810's FPU wasn't particularly fast.
The V810, which is a low-cost 32-bit RISC design, did not have a huge instruction cache, but it was still double the size of the Super FX instruction cache, and far better than none on the 68000. We don't know the RAM bandwidth, but it's most likely larger than the Super FX in aggregate.
Versus the Super FX, we do lose a few specialty functions. We lose the ability to read from one bus and write to another in parallel, and we lose the dedicated pixel "PLOT" function that hid complex address calculations from the programmer. On the Super FX you can just think in terms of (X,Y) coordinates and it can figure out what data you mean regardless of 2-bit, 4-bit, or 8-bit pixels, but on the V810 it's all up to you to figure it out.
At this point the programmer is also now faced with the bewildering task of managing the symphony of 4 very unique CPUs (the 65C816, the SPC-700, the HANDS chip, and the NEC V810) and continuous data flow between them. Certainly no less herculean a task as coordinating two Motorola 68000s, a Z80, and 2 SH-2s with a 32X and Sega CD. But this was indeed the era (e.g. the twin SH-2 Sega Saturn) where John Carmack lamented about wishing for just one good CPU.
So now let's put all of this in context with the competitive landscape:
CD-ROM ADD-ONS FOR |
NEC TurboGrafx-16/PC-Engine |
Nintendo Super NES (Sony CD-ROM) (1991-1992 vintage) |
Nintendo Super NES (Philips CD-ROM + SuperFX) (1992-1993 vintage) |
Sega Genesis/MegaDrive |
Nintendo Super NES (Philips CD-ROM + 32-bit V810) (1993 vintage) |
SNK Neo-Geo |
Vintage |
1988-1994 (various) |
1991-1992 |
1992-1993 |
1991-1992 |
1993 |
1994 |
Additional RAM (not including backup) |
512
kbit (64 KByte) for CD-ROM2 (1988/1990) |
2 megabits (256 KBytes) |
8 megabits (1 MByte) + either 256 kbit or 1 megabit SuperFX RAM |
6 megabits (0.75 MByte) + 512 kbits (64 KB) in total; 2 for Genesis, 4 for coprocessor only, rest for ADPCM sound decoder |
13 megabits (1.625 MByte) in total; 1 for SNES, 4 for HANDS, 8 for V810 |
56 megabits (7 MBytes) in total across all cartridge buses |
Coprocessors? |
None |
None |
Yes. 2: HANDS + SuperFX |
Yes. Supplemental Motorola 68000 and sound decoder |
Yes. 2: HANDS + V810 |
None |
CD-ROM speed |
1x |
1x |
2x or 1x |
1x |
2x or 1x |
1x until Japan-only CD-Z |
Audio? |
Disc Streaming + 1 ADPCM channel |
Disc Streaming |
Disc Streaming + supplemental 4-channel HANDS ADPCM decoder |
Disc Streaming + Supplemental 8-channel RF5C164 ADPCM decoder |
Disc Streaming + supplemental 4-channel HANDS ADPCM decoder |
Disc Streaming |
Backup memory? |
Built in to TurboDuo (not transportable) |
Built in to cartridge (awkward but portable) |
Built into disc caddy (portable) |
Built in and optional memory cartridge. |
Built into disc caddy (portable) |
Built into console (not transportable) |
In this era, any Nintendo CD-ROM accessory has 2 chief competitors to fend off: 1. the outside competition and 2. the ever-growing cartridge itself.
The original Sony CD-ROM accessory stood no chance of lasting in the market. It's just a TurboDuo! By 1993 we had 16 megabit cartridges with eyes toward 24 and 32 by the end of 1994. Did we really need to limit game developers to loading no more than 2 megabits at a time off a disc?
The original Sony CD-ROM accessory featured no coprocessors of any kind at a time when Nintendo was already working with Argonaut on what became the SuperFX. That meant that cartridges were growing in both size and capability while the CD-ROM just had (slow) bulk storage behind it.
This was the same problem that killed the Famicom Disk Drive! Cartridges eventually outpaced the disk in capability (thanks to the MMC chips), speed, and even total capacity.
The Philips CD-ROM + Super FX combination was pretty solid! This would fare well against the Sega CD.
The Motorola 68000 is far easier to develop for, with more tools behind it. However, the SuperFX is considerably faster when treated right. It would be a great tool for the same uses as that 68000 --- software graphics decompression, sprite scaling, sprite rotation, etc.
Nintendo's unit would have far more RAM available for sound than Sega CD. Great for beefy, bulkier sound effects.
Nintendo's unit would share 4 megabits of RAM directly (and transparently) between the Super NES itself and the SuperFX, so you could use it however you needed to.
Nintendo's drive was faster when called to be so (i.e. not when streaming audio).
The Philips CD-ROM + V810 combination was really solid! This would surpass the Sega CD. At last we might have a unit with a few years of life in it.
The V810 is faster than the 68000, compiler-friendly, and had official toolchains at the time.
The even larger RAM could buffer up more data. The V810 could be used as a graphics decompressor or even run game logic if desired. These are things that weren't quite cheap enough to build onto a cartridge (yet).
Short answer: NO!
If you read
Diehard Gamefan, August 1994, Volume 2, Issue 9, page 157, you
came across this perplexing rumor.
As tantalizing as the idea
may be (having a beefy CPU to run a game engine and a “Super FX
3” to draw the graphics), at that point you might as well just
make a new system from scratch or else you just have another Sega
32X.
Imagine the chaos of programming the SNES (2 CPUs) and an
R3000A and a “Super FX 3” and coordinating it all. And
fitting it physically into a cartridge. What a mess. And in the end
you’re still limited by the VRAM and DMA performance of the
SNES anyway.
Indeed, I think this author just got some wires
crossed and is confusing the Super NES CD-ROM effort with the
aforementioned Argonaut stereoscopic display system. That fits all
of the other details of the story: “new market”,
“projection system”, “VR effect”, “VR
headset”.
Ultimately, this Argonaut VR project was
scrapped for the Virtual Boy instead. My suspicion is political
expediency. When the CD-ROM project was cancelled, NEC lost
potential sales on its V-810. And yet this was the same NEC
that was contracted to build chips for N64. Better to keep NEC
happy and sacrifice Argonaut instead? Not to mention a
project to keep your in-house engineers occupied while their other
foreign-designed (Silicon Graphics, Inc.) console made
progress.
The final patent for Nintendo’s
CD-ROM never makes any mention of such things either.
For this example let’s use Rare’s
1995 Super NES port of Killer
Instinct. It was a 32-megabit
cartridge with no special hardware at all. We all knew it was cut
down from the arcade original, and yet it was very well-received in
spite of this. It made a great tide-me-over while we waited for the
Nintendo 64 and it played well despite its obvious limitations.
The
game features 11 combatants.
We can’t dive too deeply into
the guts of this game here. For certain there is plenty of generic
stuff inside: the main program, the sound driver to play back music
scores, common title screen and character select screen graphics, and
of course all of the narrator combo announcements like “Super
Combo!” or “Monster Combo!”. I don’t know
how much that all takes.
So let’s for now assume 30
megabits divided by 11 or just under 3 megabits per character.
This
is all armchair back-of-the-napkin arithmetic.
And that 3
megabits includes: one or two background levels, music score,
character sound effects, character data and graphics (undoubtedly
compressed somehow), and each character’s victory pose and
game-ending story scenes.
If we ported Killer Instinct onto the Sony
CD-ROM, the only tangible gains would be:
1) We can stream the
original arcade audio off the CD-ROM. This is good for quality (but
adds lag seeking tracks). That saves the SNES audio memory to hold
the sound effects.
2) We can leave the victory and game endings
on the disc (maybe even as short videos or simple animations!) as
they can be loaded as-needed. The result is they can be bigger and
take more space. Again, disc seeking lag is still a factor.
We can load the background graphics and
sound effects to the SNES’s own audio RAM and video RAM
directly. Maybe we can load the main program to the SNES’s
internal WRAM and have some room to spare if we are lucky. But since
that system cartridge just has 2 megabits of RAM, we have maybe only
1 megabit per combatant left to fit all their moves (graphics and
data).
It is hard to see this as a real win over the
cartridge! The in-game play is probably already compromised.
It’s
not even clear that all of the combo announcements would fit anywhere
in any of the RAMs to load into the SNES audio memory when needed.
(I don’t know how much space those take.)
The overall footprint of the game would be
much bigger than the SNES cartridge (because the audio tracks and
maybe the game endings could take up lots of space) but the in-game
action feels worse -- like maybe 1 or 2 megabits per
character times 11 characters, or about 22 megabits in total.
It
feels more like a DOWNGRADE.
So would this $200 machine really be an
“upgrade” to an SNES in 1992 or 1993? Only to be outdone
by a cartridge in 1995?
That’s fundamentally why this was
such a dead-end design.
The Philips CD-ROM + Super FX system cart
combination already helps us.
Again, we gain:
1) We can
stream the original arcade audio off the CD-ROM. This is good for
quality (but adds lag seeking tracks). That saves the SNES audio
memory to hold the sound effects.
2) We can leave the victory
and game endings on the disc (maybe even as short videos or simple
animations!) as they can be loaded as-needed. The result is they can
be bigger and take more space. Again, disc seeking lag is still a
factor.
3) And now we also have roughly 4 megabits of
audio memory, assuming a little overhead for other activities that
HANDS might perform. (HANDS has supplemental ADPCM decoder channels,
which are required by the XA format, so you might as well use them
while streaming the background audio off the disc, right?) With any
luck, hopefully this can hold the combo announcements and leave room
for better quality character sound effects. And you can still play
hits and smacks on the SNES audio itself.
4) And now we have
roughly 4 megabits of storage on the Super FX (minus a little program
code overhead). We can also use the Super FX to do whatever graphics
decompression duties might be useful.
So now we can bat somewhere in the range of
3-4 megabits per combatant for the in-game action, or about 33-44
megabits total. Again, not counting the audio or game endings or
victory animations. The in-game action should be about equal to the
cartridge, and maybe a little faster with the FX doing decompression
for us.
Just about everything (audio, character graphics, combo
announcements) ought to be as good or better.
And that’s
why this unit starts to shine.
Furthermore, data
loading off the disc COULD be double-speed, although audio streaming
still has to be 1x. And skipping around to load up characters can
still be laggy.
Still, I hope it’s clear that this unit
just beat the pants off the Sony drive.
Virtually the same scenario as
above.
Admittedly now we’ve traded out the Super FX for
the NEC V-810. We lose dedicated pixel plotting hardware for a more
general-purpose CPU, but it’s beefy enough. It could probably
run the whole game itself if we wanted to, meaning faster action
everywhere.
But now we’ve also got even more local RAM
hooked up to it. A portion of it is required to hold its program
code and work space. So, just guessing if we have 7 megabits left
over….
That becomes about 3.5 megabits of graphics + 1-2
megabits audio per combatant or about 4.5-5.5 megabits each.
Or
the equivalent of a 49-60 megabit cartridge plus room for audio and
ending animations.
So here, if you are willing to put up
with CD-ROM lag, you can get a genuine upgrade over the cartridge.
It would never match the arcade original, but it’d be nice!
We should have seen 64-megabit Super NES
cartridges. There are a number of ways to map that much space on the
console. Even the SA-1 can do it.
However, we never got
any Super NES games that big because:
1) 64-megabit ROMs were late to market.
Moore’s Law is not perfect and there are occasional hiccups.
This hiccup hurt us.
2) The capacity was being hoarded for the
first round of Nintendo64 cartridges, which ended up not arriving to
market before late 1996 anyway.
3) Developers were already
moving to the new consoles.
But if we DID have a
64-megabit cartridge back then it would have given fresh competition
to the above examples. Especially if bolted to the various special
chips Nintendo had at the time.
With Rare porting Killer
Instinct 2 for the Super NES (which was in development and
cancelled), one could have dreamed for such a configuration. The
cartridge they used was never confirmed.