http://www.youtube.com/watch?v=BI_puH7gYR8 Not a MODEL 3 demo, but the PRO-1000 image generator was the same base technology.
I'm no expert by any stretch but the PRO-1000 image generator pushed 750,000 textured & lit polygons/sec with all features on. That alone is more than twice the performance of Model 2. Model 3 had twin GPUs (maybe even customized somewhat for Sega) which could sustain north of 1 million such polygons/sec in actual games. Model 2 had Real3D's heritage (GE Aerospace / Martin Marietta) but technically not a Real3D chip. Model 2 was a heavily upgraded Model 1 board with texture mapping but lacked the extra graphic features (starting with gouraud shading) of Pro-1000-based Model 3. But still, the demo is not on par with Model 3 games, as its a single GPU system.
Really cool video! It definitely looks like a step below most Model 3 games, but it looks really cool nonetheless. Maybe they were using a less powerful CPU along with the PRO-1000. By the way, do you know exactly what CPU the Model 3 used? Some sources I've read say it used the PowerPC 603, and others say the 603e. I'm also curious about the clock speeds of the different steps. It was my understanding that the PRO-1000 resulted from a collaboration between Sega and Real3D. It's really unfortunate that Real3D never saw mainstream success outside of the Model 3. It would've been great if the Dreamcast had used a Real3D chipset. Even though Real3D is no longer a company, undoubtably some of the technologies they worked on still exist in current graphics hardware.
IIRC spec-wise this GPU was near M2 levels, the textures alone are better than Model2's or any 32bit console
I am not certain about PowerPC 603 vs 603e, I had thought they were all using the 603e at 66, 100, 166 MHz According to System16.com: STEP 1.0 - 603 66 MHz STEP 1.5 - 603 100 MHz STEP 2.0 - 603ev 166 MHz STEP 2.1 - 603ev 166 MHz http://system16.com/museum.php?id=1 According to Wikipedia they use the 603e http://en.wikipedia.org/wiki/List_of_Sega_arcade_system_boards#Sega_Model_3
Yeah, that's what I was referring to. I wish someone could verify which model it actually uses. I've looked at my VF3 board, but I don't think it says the model number on the chip itself. The CPU is hidden underneath a heatsink.
I'd say that even a single Pro-1000 GPU (MODEL 3 had two of them) was somewhat beyond what the final M2 could handle in practice. Given the Pro-1000's 750,000 pps figure which was not a peak theoretical number but what could be used all the time, independent of a CPU. And that fact that it could be locked to either 30 or 60 fps. Pro-1000 had every graphic feature/effect that M2 had, and then some (better/more lighting, hw AA, etc). The timeframe these chips were developed was, very roughly, the same. Early to mid 1990s. Pro 1000 image generator and Model 3 board were shown in 1996 so hw development was right after Model 2 which appeared in 1993 (Daytona USA's inital released in Japan). Model 3 arcade games released in late 1996 (VF3, Scud Race) and into 1997. M2 development started around 1993 (Bulldog) and shown at E3 1995 (demos, not counting the CGI renders). We saw screens and previews of M2 games during 1995 and 1996. Final hardware with 8 MB RAM and two PowerPC CPUs was somewhere between late 1996 and early 1997, I THINK, before the console version was canceled in early-mid 1997. M2 powered 4 or 5 Konami arcade games and the M2 tech showed up in all kinds of commercial / industrial devices, etc.
My understanding is that Sega intended to release three arcade titles to showcase the Model 3 board in the middle of 1995, but were forced to significantly delay their plans because of Lockheed Martin having problems mass producing enough stable Real3D chips. I've never seen any confirmation of what these were, but Virtua Fighter 3 is a definite - Yu Suzuki commented on how development of VF3 had to begin using a modified version of its predecessor's engine on the Model 2 board for a while, plus it's well known that Indy 500 was meant to have been a Model 3 game. As for the other, I reckon it was probably SCUD Race, since an interview with staff at AM Annex confirms that production on Sega Rally 2 did not begin until long after the Model 3 platform had been finalised. On the subject of SCUD Race and its creation, I seem to recall once reading somewhere that AM2 had started work on the original Model 3 specification, urging Lockheed to push ahead with a "Step 1.5" upgrade sooner than they had wanted after R&D tests revealed that VF3 was maxing out the original design's capabilities. Always considered an interim solution, only a few titles ever made it to 1.5 before the Step 2.0 revision was produced. I'm not sure exactly when the final 2.1 version came into existence, but it's clear Sega's relationship with Lockheed continued in some capacity long after Real3D had been dropped (in favour of 3dfx providing the graphics solution for the ill-fated Eclipse upgrade cartridge project, produced internally within Sega of America without any initial knowledge from their Japanese counterparts). By the middle of '97, both the Eclipse and NetLink-incorporating standalone Pluto unit had been abandoned, with the existence of competing proposals for a standalone next generation system - Black Belt and Dural/Katana - already being reported in videogaming journalism circles. Then again, when did management finally choose NEC's PowerVR2 chipset over the US-backed 3dfx alternative? The dates are probably listed in the lawsuit made against Sega, so this timeline may not be entirely correct. After all, what makes it to print especially is usually several weeks out of date before readers get to digest such information. Two details I'd love to know more about are whether the Eclipse would use the Saturn's cartridge port or the same connector as the Video CD Card, as I've heard conflicting rumours on this point, and just how much on-board RAM this device would contain, if any. Also, does the name Guppy seem familiar to anyone? This was recently mentioned in a slide during the GDC 2014 postmortem on Shenmue as a working title for what appears to be the game itself, though I'm fairly sure I saw this listed on various news sites back in early '98 as a possible further codename for the Japanese team's efforts, later officially unveiled as the Dreamcast. P.S. Weren't there brief negotiations between Sega and the M2 partnership to make VF3 an exclusive home conversion for this console? From what I've seen of the hardware in terms of actual performance as opposed to theoretical figures, I doubt it would have been anything like arcade perfect!
I recall reading in Next Generation that Matsushita claimed M2 had similar capabilities to the Model 3 board. I think it was worded carefully to impy M2 had similar graphic functions, not performance. NG later asked 3rd party devs what they were getting out of M2. The consensus was 2 to 3 times the polygon performance of Nintendo 64. Nintendo 64 = 160,000 triangles/sec in games with everything on M2 = 300,000 ~ 500,000 triangles/sec depending on the shading, lighting etc. Model 3 Step 1.0 = over 1,000,000 rectangle polys/sec with everything on (lighting, AA, etc) in game, with framerate locked at 60fps no matter what. Of course a graphically arcade perfect port of VF3 would be impossible on M2. That said, in 1997, there's little doubt M2 would've been the only consumer-level 3D hardware capable of handling a reasonably impressive VF3 conversion in terms of polygon count, textures, lighting and other graphic functions. According to devs M2 was somewhat more powerful / better than first gen 3Dfx Voodoo Graphics and PowerVR PCX2, which were the most powerful PC 3D accelerators of 1996-1997. It's understandable why Matsushita didn't want to risk 1+ Billion dollars on M2 in 1997. Dreamcast was in development and right around the corner, would launch in 1998, 1999 with much more powerful hardware. Nevermind PS2 and Dolphin.
The Model 3 used quads? I didn't know that. I guess I'd always assumed that it used triangles. I think the Model 2 used quads.
Model 3's R3D Pro-1000s could do both square polys and triangle polys. http://system16.com/hardware.php?id=717
On the subject of Sega's preference for quadrilaterals around Saturn period, does anyone know if there would have been any compromise from 3dfx as chipset provider for the Eclipse upgrade cartridge? In other words, would it have been possible for such an add-on device to turn a stock console from natively generating quads to the more conventional triangular polygons used in 3dfx graphics cards, or would they have created a solution to simply increase the vanilla system's existing capabilities? Apart from boosting its host's geometry performance, I suspect the Eclipse would have contained additional memory, though probably not other features such as hardware anti-aliasing, alpha blending or z-buffer processing (which only made their debut with the Dreamcast once Sega turned away from more complicated architecture designs). P.S. Isn't the Saturn's use of quads a leftover from when it was still intended as a mainly 2D powerhouse directly following on from the "Super Scaler" games and System 32 board, where distorted sprites were favoured over polygons in the established sense most think of them now?
