A Saturn Programming Challenge

Hi all,

According to internet rumors, Virtua Fighter 3 was able to hit 750,000 polygons onscreen with a trick using the DSP that Suzuki & Co. picked up while creating Digital Dance Mix.

My question to anyone with programming knowledge: Is this even feasible? We know for a fact from interview data that a 3D accelerator cartridge was in development and Shenmue was developed for it, but VF3 is alleged to have been finished on stock hardware.
 
750,000 polygons onscreen is a delightfully abstract number that does not have enough information to extrapolate what was actually happening.
Is it 750,000 per second? Per frame? Per ms? Per kilobyte? The unit of time being absent simply makes this void.

If of course you entertain the concept as simply on-screen, it does not take any extreme engineering to do that. Simply do not clear the framebuffer or VDP2 RAM and draw as many polygons as you like, you definitely won't be able to make out 750,000 unique polygons in 640x480 let alone 320x240. 750,000 polygons is more than the amount of pixels on the screen.
 
750,000 polygons onscreen is a delightfully abstract number that does not have enough information to extrapolate what was actually happening.
Is it 750,000 per second? Per frame? Per ms? Per kilobyte? The unit of time being absent simply makes this void.

If of course you entertain the concept as simply on-screen, it does not take any extreme engineering to do that. Simply do not clear the framebuffer or VDP2 RAM and draw as many polygons as you like, you definitely won't be able to make out 750,000 unique polygons in 640x480 let alone 320x240. 750,000 polygons is more than the amount of pixels on the screen.

Honestly, I have no idea. I assume per second? The above link is posted around places. I've been trying to get in touch with ex-Sega employees with little success. It's not much, but all we have to go on. One way to assess this is the whole Mythbusters approach and see if such a trick is possible, even in theory.
 
Back then it was always "polygons per second" being talked about, so if they said it then, there's no reason to think it would mean anything else.
 
Yes you must divide by 60 but saturn will Nevers reach 750 k poly even with a Magic lib or dsp. The mythical sgl 3.2.0 IS far from allowing that
 
If you do per-second, you have 12,500 polygons per frame at 60fps or 25,000 polygons per frame at 30fps. This is an order of magnitude, somewhere around 20x more than what the system is otherwise established to be able to do.
Of course, I am talking about polygons processed and displayed in a fully real-time manner, compared to this abstract number.
Still, it is absolutely impossible. If we are talking about using VDP1 to draw them, VDP1 cannot even process 12,500 command tables in 60hz. If we are talking about software drawing, it becomes more important to describe, in great detail, what exactly these polygons are. If they are voxels, it becomes more realistic to consider methods of delivering an extremely stretched concept of "polygon" at a rate of 750,000 per second by of course meaning that:
1. most of those polygons are occluded, i.e.,, not actually drawn
2. what is drawn are a multitude of single-color squares
At that point, your closest comparison point is Amok, which can often fill the screen with voxels which by some translation between Japanese may be able to be misconstrued as "polygon".

VF3, in no form, used voxels though. The Saturn also does not have enough memory to describe animated characters drawn by voxels. I don't discount that they may have been able to do that with a dev board with a massive RAM extension (something on the order of hundreds of megabytes) because voxlap, a renderer in use around the same time, could be used to animate voxel characters. It still did so using heavily compressed data, but the idea is extendable.
 
Ponut, thanks for your take. The number is indeed well above the theoretical maximum I've seen elsewhere of 500,000 flat-shaded polygons per second (200k textured).

As far as interactions with the DSP, would it be possible to incorporate that in graphics processing? I've also read that something similar happened with Burning Rangers where the sound suite was used for graphical purposes. But again, it's internet information, so...
 
OK, so I looked and I didn't find a link about that Burning Rangers claim, but there is this nugget from Sega Saturn Shiro:

Not only are the character models themselves colorful and detailed, but load times have been cleverly masked by utilizing the Saturn’s Yamaha sound chip as a pre-loading processor (wow!) to hide actual loading. As a character moves towards their target in battle sequences, the Saturn is busy loading 3D model data into the sound processor whilst VDP 1 and 2 continue to render the sprite characters and 3D battlefield. When the gamer selects their target and takes action (attack, cast spell, etc.), the 3D character models are instantly rendered, and the sequence plays out. Consequently, the game gives the Saturn’s laser track a heck of a workout, but the trade-off is necessary to keep the game moving at a brisk pace whilst rendering any of a dozen characters in quick succession without any perceptible breaks in the action.


If true, this proves that later era first-party games used the sound suite for graphical purposes. They tend to be a reputable source, although they do not cite their own for this information.
 
OK, so I looked and I didn't find a link about that Burning Rangers claim, but there is this nugget from Sega Saturn Shiro:




If true, this proves that later era first-party games used the sound suite for graphical purposes. They tend to be a reputable source, although they do not cite their own for this information.
This is not true......


The models are loaded into 2 sections of LWRAM, and decompressed on the fly. It uses standard SGL, so unless SGL does this, (which I have not encountered with my reverse engineering tasks on SF3) this is a bit misinformed. I have opened a dialog with that article's author to find out where that source came from.
 
KoD, thanks for your input. Isn't it a bit unusual for something released that late (Dec. '97 JPN according to Wikipedia) to use SGL 1? Shouldn't it use 2.0?

Possible resolution: It's been literal decades since I've used SSF, but it used to have options to turn specific processors on and off. Could we yank the sound suite and see if that affects the display? Or is it not that simple?
 
who said it uses 1.0? We also do not need to yank out the sound suite. The music in SF3 is all done via the SCSP with no CDDA audio. It is filled with PCM sounds to create the synthesized music. The author of that article has not gotten back to me yet, but TrekkiesUnite118 speculates the source of this rumor confused the SCU DSP with the SCSP DSP, and thus thought the sound processor was somehow linked to the loading of models.
 
I have opened a dialog with that article's author to find out where that source came from.
I have no idea of the origin, but e.g. here is a post from 2003 making the same claim. You're better off just ignoring any forum or blog post making technical claims unless they provide some evidence to back them up, especially anything from 10+ years ago. I can remember more than one instance of people just making shit up.
 
I have no idea of the origin, but e.g. here is a post from 2003 making the same claim. You're better off just ignoring any forum or blog post making technical claims unless they provide some evidence to back them up, especially anything from 10+ years ago. I can remember more than one instance of people just making shit up.
oh yeah, that is what is so wonderful about the Saturn, so many rumors to debunk now they we have better tools to play with.
 
Those claims about VF3 are probably all BS. Even Yu Suzuki didn't remember anything about VF3 on Saturn. And while an accelerator might have been considered at some points, it makes no sense from a commercial point of view. If anything a VF3 game would have most likely used an improved version of the Fighters Megamix or VF2 engines.
Fighters Megamix even uses a lot of the VF3 moves, so you can imagine how part of the work was done, all they needed to do was to work on the backgrounds, finish implementing the movesets and improve the 3D model a bit to call it a day.
Games like Last Bronx or Dead or Alive show that good looking backgrounds are possible with the VDP2, the only thing missing would have been the arenas having different heights, but it's not that hard to do on the VDP1 if you do it smart.
All that to say that pushing 750 000 polygons per second is impossible on the Saturn, any add-ons for the system wouldn't have been greenlit in 1997 and Sega didn't need to push an insane amount of polygons to do VF3, so those internet claims are BS.
 
Those claims about VF3 are probably all BS. Even Yu Suzuki didn't remember anything about VF3 on Saturn. And while an accelerator might have been considered at some points, it makes no sense from a commercial point of view. If anything a VF3 game would have most likely used an improved version of the Fighters Megamix or VF2 engines.
Fighters Megamix even uses a lot of the VF3 moves, so you can imagine how part of the work was done, all they needed to do was to work on the backgrounds, finish implementing the movesets and improve the 3D model a bit to call it a day.
Games like Last Bronx or Dead or Alive show that good looking backgrounds are possible with the VDP2, the only thing missing would have been the arenas having different heights, but it's not that hard to do on the VDP1 if you do it smart.
All that to say that pushing 750 000 polygons per second is impossible on the Saturn, any add-ons for the system wouldn't have been greenlit in 1997 and Sega didn't need to push an insane amount of polygons to do VF3, so those internet claims are BS.

Actually, the evidence for the 3D accelerator cartridge is pretty solid:

[Part 1] Shenmue Discussion with Yu Suzuki & Developers | IGN Japan

Above link has the devs on tape talking about having to re-develop Shenmue twice: Once for the accelerator, and again for the Dreamcast.

I know the Yu interview you are talking about. There is another time when he went off-record in 2011. I've tried to contact him directly and intermediary guy said he is contractually unable to talk about anything related to his time at Sega. So, I don't "trust" that per se.
 
Actually, the evidence for the 3D accelerator cartridge is pretty solid:

[Part 1] Shenmue Discussion with Yu Suzuki & Developers | IGN Japan

Above link has the devs on tape talking about having to re-develop Shenmue twice: Once for the accelerator, and again for the Dreamcast.

I know the Yu interview you are talking about. There is another time when he went off-record in 2011. I've tried to contact him directly and intermediary guy said he is contractually unable to talk about anything related to his time at Sega. So, I don't "trust" that per se.
an "actually" wasn't warranted. XL2 acknowledged the acceleration card, he just explained that it was too late in the Saturn's life to become a commercial product, and that VF3 was not the game that would require it.
 
an "actually" wasn't warranted. XL2 acknowledged the acceleration card, he just explained that it was too late in the Saturn's life to become a commercial product, and that VF3 was not the game that would require it.

No offense was intended by that comment. He mentioned no add-ons would have been greenlit in 1997. I'm a bit fuzzy on the timeline of the thing, but it has been verified that it was in active development at some point, and is not internet BS. In the quest for the truth on VF3, that's an important consideration. That's all I meant.
 
By mid-1997 the Saturn was already killed in the west and a lot of games were moved to the Dreamcast, which was already being worked on. So while a VF3 game was certainly being considered or worked on, I think something like Fighters Megamix would have been the most likely route they would have taken. I know the accelerator prototype has been mentioned and probably exists, but Sega did what had to be done and moved Shenmue on the Dreamcast as in 1997 it just wasn't viable. The Shenmue prototype we saw in Shenmue 2 was beautiful, but it didn't push an insane amount of polygons compared to stock Saturn, so it seems to me that at best it was to improve cpu calculations instead of GPU - if that video was really running with the accelerator board.
 
Back
Top