The origins of Sega's Model 2 arcade board, and, an interview with Real3D

[Da_Shocker]

Saturn was basically a system 24/32 evolved into a home console, combined with a fix to all the mistakes they had with the Megadrive. A clean slate. It was also meant to be VERY upgradable, that's why the cart slot was so terrible, it was overcrowded. It also could do model 1-esque graphics. Jupiter was the same thing but based on cartridges, and it was dropped so early that by late 93 / early 94 it was a "never happened" thing when Sega was interviewed about it.

Model 2 was literally Model 1 plus texture mapping from what I understand. Model 3 was a big step up however.

Engineering samples of the Saturn custom hardware are dated 1994 may-june. We also have official documents talking about how to use certain parts of it, like the sound chip, dated late 1993, as well as many flow charts of the system in various detail with similar dates. SH2 silicon came out in early 1994 too.


oh man you sure are telling us things we don't know!

I disagree there Model 2 was a huge leap forward compared to the model 2 to 3 jump. I'm pretty sure both VF and VR were running both at 30FPS with flat shaded polys. Model 2 doubled that and added textured mapping. Model 3 just really added really more polygons and other small graphical upgrades.

[zyrobs]

The core chipset of the Saturn was almost certainly solidified by some time in 1993, and anything going forward in early '94 would have been final bug fixes (in the chips and overall PCB/system design) in preparation for the revision suitable for mass production. Within that though, could include some variables for RAM and CPU(s) being used, since both of those could be somewhat modular (within the address and interface constraints of the chipset).

As such, it's conceivable that Sega had prototypes using the Saturn chipset with varying CPU (and possibly RAM) configurations, like SH1 derivatives (particularly before the SH2 was available, and before it reached commercial production volumes -ie if it hit delays, Sega might use SH1 as a potential fallback), Single SH2, and then the dual SH2 config used in production. (which was probably solidified in early 1994, particularly given the 32x's design config . . . also implying that Hitachi's production capacity for SH2 was reasonably established by early 1994)

It is of course possible that they hooked up prototype versions of the ASICs to different debugger boards, probes, etc, and at one point they weren't driven by SH2s. I believe that is standard procedure at that early point of development when you are just testing if the thing gives picture or not.
Anything meant to be used for developing applications would have to have used something at least close to the final hardware though. Using a single SH2 was a likely option, but a more likely one was having less RAM (the work low ram looked very tucked-on for the final hardware).

As for first silicon:
http://www.hotchips.org/wp-content/u...S4/HC6.4.2.pdf
SH1 September 1992
SH2 October 1993 (so Sega should have been able to get SH2 engineering samples around late 1993)

Notwithstanding the accuracy of that document, "first silicon" does not mean working silicon, or bug-free silicon... or anything worth shipping to third parties. Plus, there were multiple different versions of the SH2. The generic(?) SH7604 had at least 6 samples, and the one in the Saturn was a SH7095, and who knows how that one was different, if different at all.
The earliest actual SH2s I have pictures of date to 1994 late May-June.

Yes, and no . . . and at some point, there's the general impracticality of cost for providing external expansion signals. That' and there's a powerful argument for not going the add-on route at all, at least if you want something that complex:

That and the 32x still would have been in a very problematic position next to the Saturn even if it was more capable and less bottlenecked.

Sega was known for having knee-jerk reactions. Adding so many external inputs for the Saturn was probably a direct consequence of having big issues with expanding the Megadrive. The 32x wasn't a bad idea, but it wasn't powerful enough to make the console truly next-gen, it was frustrating to install, it had competition from the same company, and it was badly supported. If it was more user and developer friendly, and didn't directly compete with another Sega hardware, it could've extended the Megadrives life by a few excellent years.

[zyrobs]

I disagree there Model 2 was a huge leap forward compared to the model 2 to 3 jump. I'm pretty sure both VF and VR were running both at 30FPS with flat shaded polys. Model 2 doubled that and added textured mapping. Model 3 just really added really more polygons and other small graphical upgrades.

I'm not familiar with the architecture, this is just my personal understanding. I may have misheard someone elses opinion about it.

Pushing up polygon count can be done trivially - as long as the graphics processors fillrate can keep up, you can just add in faster 3d math to push more polys. It's entirely possible (and likely) that they touched up the graphics processor to enhance their speed too, on top of adding textures and other features.

[MrSega]

"The core chipset of the Saturn was almost certainly solidified by some time in 1993, and anything going forward in early '94 would have been final bug fixes (in the chips and overall PCB/system design) in preparation for the revision suitable for mass production. Within that though, could include some variables for RAM and CPU(s) being used, since both of those could be somewhat modular (within the address and interface constraints of the chipset).

As such, it's conceivable that Sega had prototypes using the Saturn chipset with varying CPU (and possibly RAM) configurations, like SH1 derivatives (particularly before the SH2 was available, and before it reached commercial production volumes -ie if it hit delays, Sega might use SH1 as a potential fallback), Single SH2, and then the dual SH2 config used in production. (which was probably solidified in early 1994, particularly given the 32x's design config . . . also implying that Hitachi's production capacity for SH2 was reasonably established by early 1994)"

@koolkitty89. I guess that would mean that SEGA's core focus towards the end of Aurora/Saturn's development, that they possibly just wanted to save face with the Dual SH-2 design. I think they knew that SH-2's dual design had the same bus and 4KB thread configuration, but were afraid of delays. If in October 1993, SH-2 was set for Q1 1994 taping, SEGA could have easily made modifications that made coding it much more simplistic.(by separating the bus and adding extra backup SDRAM giving it an actual true Dual processing mechanism.) In regards to the VDP design, I would say the fact that SEGA didn't separate the DMA bus in both the VDP 1 and VDP 2 really made it all the more difficult to set up a proper middleware tool to help write assembly language much more simpler and made filtering and buffering of its 3D all the more complicated. Could the Dual SH-2 CPU itself also made it more difficult for the VDPs to be coded properly?

I would say that SEGA missed a huge opportunity in not designing Saturn with Middleware tools and yet another missed opportunity came from the lack of developing a useful engine for most of its games.

[Da_Shocker]

I'm not familiar with the architecture, this is just my personal understanding. I may have misheard someone elses opinion about it.

Pushing up polygon count can be done trivially - as long as the graphics processors fillrate can keep up, you can just add in faster 3d math to push more polys. It's entirely possible (and likely) that they touched up the graphics processor to enhance their speed too, on top of adding textures and other features.

I wasn't trying to discount the Model 3 though. The funny thing is how Konami was hyping up there Cobra system one that could pump out 5 million polygons per second. Needless to say it failed miserably.

[Vector2013]

Mr Sega, are you Keith Apicary ?

[MrSega]

Mr Sega, are you Keith Apicary ?

Nope. Not a Engineer or former Sega Amusements USA. Just a loyal SEGA fan who begin to investigate some odd patent and paperwork behavior of SEGA that started in 2009 when they filed 2 Arcade board trademarks,renewed its hardware and home consumer licenses and started issuing hardware patents. SEGA's behavior peaked when they filed the "Kids Pad" Tablet Computer trademark in 2011 and really hit it major early last year with "ORBI" multi-brand the longest,most elaborate and expensive trademark in the 60 year history of Kabushiki Kaisha Sega.

In order to understand all of this, I begin to teach myself how to read patents and also begin to study console schematics(something the average game fan knows nothing about) mainly to understand the capabilities of the Saturn because I became so fascinated with Saturn's complex design and also learned the fundamentals of console design as well as how SEGA converts arcade hardware into home consumer products.

I love SEGA and see them beyond a software publisher. I've seen their finances begin to turn completely around the past 4 years and have noticed that alot of their paperwork has coincided with several different policy changes with the company.

First in 2009, SEGA cut SOA down to size and stopped outsource publishing. Second, shortly after starting "Sega Republic", SEGA's revenue jumped. Third, the Ring series boards were not only successful, they also led to SEGA no longer bleeding operating income.

Also after and since Generations, SEGA has gotten to the point where they were publishing dozens of retail titles to now only a few and mostly only publishing and developing mobile content.

And in 2012 before the Wii U's arrival, SEGA didn't release any major titles for the platform, bailed out of publishing Bayonetta 2, and instead of releasing a major Sonic title, they shipped a Euro racing game late in the year, instead of during the summer. SEGA also didn't reveal or list any new multi-platform games at all(only 2 SEGA titles have been published on the system) and also completely shuttered its major 3rd party division and shut down most of its in house Euro 3rd party development studios.

Its interesting that so-far most of SEGA's major recent in house titles are either, PC,mobile,Digital or portable titles. This strategy is causing them to save significant amounts of budget. Even the PS4 reveal hasn't coughed up ANY SEGA projects of any kind and the Sonic fanatics are going nuts trying to figure out why 2 years after Generations, SEGA is refusing to hint,tease or reveal a new major Sonic title.

Likewise, whatever is going on internally at SEGA(before 2009, there was nothing significant being filed by Sega), it certainly is having an adverse affect on their 3rd party market.

What's interesting is that several upcomming releases like the westernized PSO2 and Company of Heroes 2 have both been delayed.

[stu]

Mr Sega, are you Keith Apicary ?

http://images.search.yahoo.com/searc...=Keith+Apicary

[MrSega]

I'm not familiar with the architecture, this is just my personal understanding. I may have misheard someone elses opinion about it.

Pushing up polygon count can be done trivially - as long as the graphics processors fillrate can keep up, you can just add in faster 3d math to push more polys. It's entirely possible (and likely) that they touched up the graphics processor to enhance their speed too, on top of adding textures and other features.

I do appreciate you having actual documented beta and prototype Saturn components.

[TrekkiesUnite118]

Mr Sega, are you Keith Apicary ?

No, just a crackpot conspiracy theorist.

[zyrobs]

I do appreciate you having actual documented beta and prototype Saturn components.

I'm just keeping a bunch of folders of photographed sega hardware. Thank the guys who made the photos instead, like Retrojunkie. I only told him to align the camera for better pictures. Guy buys 10 kilo devkits down to Australia, I don't even want to know how much money he has invested.

[MrSega]

I'm just keeping a bunch of folders of photographed sega hardware. Thank the guys who made the photos instead, like Retrojunkie. I only told him to align the camera for better pictures. Guy buys 10 kilo devkits down to Australia, I don't even want to know how much money he has invested.

Yeah, many thanks to Retrojunkie. BTW would it be appropiate to show the photos here? I'm a bit skeptical because SEGA is STILL protecting the Saturn's patents.

[TrekkiesUnite118]

Yeah, many thanks to Retrojunkie. BTW would it be appropiate to show the photos here? I'm a bit skeptical because SEGA is STILL protecting the Saturn's patents.

Why wouldn't they protect the patents? Look at what happened with all the random Famiclones years ago when Nintendo let the NES patents expire.

[The Jackal]

Nope. Not a Engineer

Of course you're not. There was and will only ever be the ONE.

[kool kitty89]

It is of course possible that they hooked up prototype versions of the ASICs to different debugger boards, probes, etc, and at one point they weren't driven by SH2s. I believe that is standard procedure at that early point of development when you are just testing if the thing gives picture or not.
Anything meant to be used for developing applications would have to have used something at least close to the final hardware though. Using a single SH2 was a likely option, but a more likely one was having less RAM (the work low ram looked very tucked-on for the final hardware).

I was thinking either of both of those things . . . a single SH2 (or SH1) as well as just the 32-bit SDRAM hi-RAM might have been considered or even implemented in prototypes. Using an SH1 as a stop-gap for prototyping and testing/trouble shooting the chipset (and possibly as a back-up option) would have made a lot of sense: the performance would be different, but the bus and instruction architectures would be the same.

EDGE magazine made mention of a SH7032 in pre-release specs for Sega's 32-bit next-gen system (what would be the Saturn) in 1993
Edge issue 4, Novermber 1993 listing this:

Hitachi: SH7032 CPU RISC CHIP Running at 27 Mhz
Memory: 3MB Ram
Sound: 32 channels, support for PCM and FM
Relase: November 94

While I'm certainly not one to use typical magazine "leaked" specs as hard evidence, but the specific mention of the SH7032 is rather intriguing. Not just because it's odd for a magazine to be that specific (actual CPU model number, not just a vague and typical magazine speculative listing), but also since that SH1 model is also the most useful in the role of a general purpose CPU. (it's got an 8 kB scratchpad and no embedded ROM; no SH1s have cache, so a large scratchpad would be the next best option for a CPU . . . the SH1 used for the Saturn CD-ROM subsystem has no scratchpad RAM and only ROM on-chip, so definitely MCU-like)
If Sega had used an SH1 as a stop-gap (or even back-up option) prior to SH2s being available, the SH7032 would have been the one to go for.

And, honestly, with decent optimization using an SH1 and using the SCU DSP could have worked decently well for the Saturn for 3D, still much better than the 3DO at least (no CPU/GPU bus contention, faster CPU, faster GPU, etc). And for 2D stuff it should have been perfectly fine. But, overall it obviously would have been tougher to program for (lack of hardware cache) and generally weaker in CPU resource (limited game logic performance and pretty much writing off CPU driven 3D geometry engines). Then again, the very presence of the DSP in the SCU kind of implies they were considering a weaker CPU solution at some point, and a single SH2 could apply there, but a single SH1 would really have demanded that. (SH2 having faster multiply/divide performance . . . and actually faster overall than the SCU DSP, I think -but with a single SH2, there's the issue of game logic overhead eating up CPU time and making the DSP more useful)

Notwithstanding the accuracy of that document, "first silicon" does not mean working silicon, or bug-free silicon... or anything worth shipping to third parties. Plus, there were multiple different versions of the SH2. The generic(?) SH7604 had at least 6 samples, and the one in the Saturn was a SH7095, and who knows how that one was different, if different at all.
The earliest actual SH2s I have pictures of date to 1994 late May-June.

Interesting, so engineering samples may not have appeared until 1994, or are you talking about production chips?

In any case, that would leave an even wider gap where using an SH7032 would have been critical for testing the Saturn chipset, and a longer period where Sega may have been concerned that the SH2 wouldn't hit production deadlines. (making the idea of settling for using an SH1 as the main CPU a real possibility) Then-again, any such concerns should have meant the 32x design team would have focused less on an SH2-specific design. (much less than those SH2s and software rendering would have taken a huge dive)

Sega was known for having knee-jerk reactions. Adding so many external inputs for the Saturn was probably a direct consequence of having big issues with expanding the Megadrive. The 32x wasn't a bad idea, but it wasn't powerful enough to make the console truly next-gen, it was frustrating to install, it had competition from the same company, and it was badly supported. If it was more user and developer friendly, and didn't directly compete with another Sega hardware, it could've extended the Megadrives life by a few excellent years.

A more powerful 32x could have been even worse . . . assuming it WAS still like the 32x (a MD add-on), and would have conflicted worse with the Saturn itself moving forward. Having both didn't make sense . . . having a low-end expansion module for 3D and some other coprocessing (especially cheaper than 32x) made some sense to cater to a very different market segment than next-gen platforms, but the 32x didn't fall into that category completely, and a more complex/powerful/costly one would do so even less.

OTOH, if Sega wanted a cheaper, "more immediately accessible to mass market" next-gen console, that's another issue entirely (and the one major facet of the 32x's design goals that was impractical to accomplish as an add-on of any sort, at least with the Sega CD already on the market to further complicate hardware design logistics). If they wanted something cheaper than Saturn in general, then the Saturn should have been designed differently (and perhaps with provisions for modular expansion too), but from a 1994 perspective, the best option I can see for a "last minute" design to complement but not replace Saturn would have been the so-called "Jupiter." (cutting out the CD-ROM interface and drive should have cut a huge chunk of initial retail price and made it much closer to mainstream pricing in North America or Europe in 1994 and/or '95), and such a design would have been fully compatible with Saturn too. (CD-drive add-on could upgrade it, and all Jupiter carts could play on Saturn, not to mention potential games specifically adding more content to cart games with expansion CDs)

That's a big part of the context of this thread:
http://www.sega-16.com/forum/showthr...32x-Make-Sense


Also, the MD already had tons of expansion support in hardware, it just lacked a handful of things . . . the most specific one often listed being the VDP expansion lines. Compared to the Saturn, it's expansion support was definitely on par with it, if not MORE comprehensive considering the relative complexity. Saturn has no provisions for adding more full-speed 32-bit SDRAM, MD cart bus can and does add RAM that's as fast or faster than onboard work RAM. Saturn can only expand bus A or B, at 16-bits wide on the cart bus, and 32-bits to bus A (I think) from the MPEG expansion port, and while memory can be relatively fast, actual RAM carts were slow.
The MD has expansion directly to the main CPU bus, and particularly wide on the cart slot (the entire 24-bit address range), though more limited on the side port.
There's no connection for the VDP pixel bus (nothing as such on the Saturn either -PCE does though, but never used it), and no expansion lines connected for VRAM, but since the Z80 can directly access the 68k bus (in 32 kB windows -or smaller segments with external mappers), it does mean the Z80 can make use of expansion too. (especially through the cart slot)

Both the MD and Saturn were good for modest expansion support, the PCE arguably more so, but in all cases, it gets to a point where it's so much more efficient to do a standalone new system (even with compatibility) that add-on upgrades don't make sense. The MCD itself was complex and costly enough that it really would have made much more sense to opt for a standalone upgraded MD derived CD system, cutting out some redundancy and unnecessary cost while opening up certain bottlenecks that aren't practical to do via add-on, all at similar cost to the MD add-on unit alone. (or maybe even cheaper if they compromised on some things)
Ironically, the SuperGrafx could have been done relatively cost effectively as an add-on with basically no waste or over-engineering . . . and arguably even better if integrated into an upgraded CD standard. (ie in place of Super CD and part of the Duo hardware) They actually could have done better than Super Grafix by not only adding the 2nd VDP, but an improved RAMDAC using higher color depth. (12-bit RGB should have been totally reasonable, and perhaps 15-bit or even 18-bit, particularly in the 1991 context of the Super CD release date)