Dreamcast used quads as well? Forgive me I don't follow this stuff as close as others. I thought back then SEGA was pushing that the DC could push 3 million Polygons per second for their marketing hyped to developers and Games Journalists.
Dreamcast used quads as well? Forgive me I don't follow this stuff as close as others. I thought back then SEGA was pushing that the DC could push 3 million Polygons per second for their marketing hyped to developers and Games Journalists.
Dreamcast used triangles, but it had a tile based deferred rendering setup. It broke up the screen to a number of small tiles which could fit on fast internal cache, wrote the polygons to that first, and then flushed the tile to framebuffer. This was a more efficient setup (phones today all use tiled rendering, and desktop videocards started doing it too), but it made it difficult to do certain effects because you couldn't see past the edge of each tile. There were some other differences compared to most common hardware at the time (different floating point standard, different z buffering, order-independent transparency, I think), but I was told that Sega's devkits hidden away all of that and basically allowed you to work the machine as a traditional setup, up to I think DX6 feature level.Quote:
Originally Posted by SegataS
Also IIRC Sega asked NEC to customize the PowerVR2 to have things like palette support, but I don't know the exact details. It wouldn't surprize me if it could also do quads, at least without deformation (even if it was internally breaking them up to two triangles).
But Yu Suzuki designed the Model 1/2 boards that also used quads.Quote:
Originally Posted by zyrobs
From what I can piece together, quads are also used in 3D modeling for film (and still are). Apparently meshing for lighting is one of the things that makes more sense with quads than triangles (at least when modeling directly). That said, quads are less efficient, which was critical for the Saturn as it was developed in an era where every polygon counted.
Another bizarre "quirk" of the Saturn rendering pipeline I've heard about is that it uses "forward rendering," instead of the industry standard "backward rendering." This means that instead of rendering by scanline, the Saturn renders everything by polygon. This means it overdraws a bunch of polygons you never even see, eroding it's polygon-pushing count even further.
Yu Suzuki was too busy with the arcade division, which was too important to pull off from that. And from what I understand he didn't design the Model 1/2 boards themselves but miniaturized the Lockheed Martin military grade renderers to fit into an arcade board. That's not to say that he couldn't have done better on the Saturn, but we don't know how much he could and could not do. The strength of the Model 1/2 lied in how much sheer bandwidth it had - which was the weakness of the Saturn, it had so little bandwidth for polygons.Quote:
Originally Posted by Blades
The Hideki Sato interviews confirmed that they went with sprites because they did not know how to do 3d (perhaps did not even have time to get into it), so they modified the sprite display to be deformable so it could draw 3d.
That's what forward rendering means. It means you take a texture and draw it at a given size, which is inefficient if the target size is smaller. If you had a 32x32 texture shrank down to 16x16, you were still processing 32x32 pixels (or maybe 16x32 - looking at the transparency overdraw artifacts, I think it may have skipped unused lines, but I'm not sure), and that was a waste of time.Quote:
Originally Posted by Blades
What everything else did was UV mapping aka backwards rendering, which means you first calculate the size of the target polygon, and then from the U and V texture coordinates, you calculate the texture pixels that go at a given pixel. The advantage is that you do not waste pixels when drawing. It also means you can animate textures by only modifying the UV coordinates, this was most useful for reflections. Not that the VDP1 had the speed to do reflections.
Additionally the quad rendering meant that if the two sides of the quad were not equal, then some pixels would get overdrawn, as the textures were processed line by line, and if you for example collapsed a quad to a triangle, then the collapsed corner gets every pixel of that corner drawn into it multiple times. Although I think you could potentially get around that issue with clever use of end codes, but that would open up holes at the seams.
There is that interview with Hedkia Sato where he said that originally he thought 4000 hardware sprites (funnily enough the same as the PS 2D) was enough for the Saturn 3D, but upon seeing the PS spec's he had to change that, and also added a 2nd CPU and where he also said that SH-2 had a function to support such a featureQuote:
Originally Posted by zyrobs
All the early spec leaks said the Saturn was to have a Hitachi CPU running at 27 Mhz, no talk whatsoever of dual CPU's. That spec's changed from just the timeline of EDGE issue 1 to EDGE issue 4.Quote:
based on interviews it seems like they decided on dual SH-2s fairly early in the process.
But to be fair SEGA Kazuhiro Hamanda (Cheif section design chief of the Saturn) told EDGE that early tech demo showed a single CPU wasn't enough to handle a 3D world and SEGA felt it had to go with 2 CPU's. So who knows, what SEGA would have done differently without SONY joining in the console race.
SEGA developed Model 1 their self's (and it kind of showed with Saturn like levels of multi processors joined together lol) it was with Model 2 and 3 that Lockheed Martin came onboard. I also remember that some mags were saying that because Model 3 could also use quads that gave it a performance gain over the Dreamcast, given they thought you need double the triangle polygon to equal a quad.Quote:
Originally Posted by zyrobs
Mad days lol. Btw Did Namco's System 21 and 22 used quads? and many mags, even some developers like CORE said that the PS also used distorted sprites (but in a different way) but had the 66 Mips of the Geometry engine to give more than enough grunt to push more polygons.
Well here's what Hideki Sato saidQuote:
Originally Posted by axel
Quote:
it seemed like we were finally nearing completion. Then, the final PlayStation was revealed. It supported 300,000 polygons. Well, that was ultimately a bunch of lies, but… When you compared the Saturn with the PlayStation, we were completely missing something. The response that I chose was to add another SH processor, so we ended up with two SH-2s. By chance, the SH supported two-way cascaded data transfer. You could add a second processor and connect them in a cascade and get multi-CPU performance. When you get to about the PlayStation 3, multi-processors had become common, but the Saturn was the first home console to use multi-processors. So I added a second SH-2, but I felt that the ‘impact’ was still weak. Well, the SH-2 is a 32-bit processor, and we had two of them, so we could call the Saturn a 64-bit machine. It’s a dirty way of getting to 64-bits. But we revealed the CD-ROM-based Saturn using 64-bits as our sales point.
It makes sense. I've seen some people with engineering expertise suggest that the second pool of slow ram was added late based on the motherboard design.Quote:
Originally Posted by zyrobs
From my understanding, those quads used in 3D modeling are broken into triangle on a hardware level.Quote:
Originally Posted by Blades
As for why the arcade boards used quads, my completely uneducated speculation is that because 3D games at that time had low polygon budgets, so they thought making box shaped cars and houses would be easier with quads?
Quads have some real world uses. NVidia's first card (NV1) used quads, but that's not what made it great: it allowed for, I think, 9-point deformable quads. Basically you have the 4 corners, plus one point between each corner, plus one in the middle. You could deform all of those points and the chip would render these distortions fine. So you could make more smoothly curved surfaces with far less polygons than with triangles. Of course, that never really caught up, they ditched their half-complete second card, and then made a fully DirectX/OpenGL compatible card with the NV3, and eventually ended up dominating the market with the Geforce.
The extra ram is connected to an interface chip that controls the SH2s in dual chip mode, plus they have some of the last IC numbers on the board. So it would make sense that the two chips were added as an after thought.Quote:
Originally Posted by bpguimaraes23
I think it was a bad choice, they should've gone with adding 2MB fast ram instead. But maybe it made sense from a price point of view.
I think the interview with Sato-san makes it clear it was added after learning of the SONY spec. I really don't think Sato-san is a man to point score and given he worked on every SEGA console ever made, I think he's the highest authority on what happened about all aspects of the Saturn design (and also to blame for some of its shortcomings) . BTW does anyone know if Konami Cobra Board and Namco's system 21 and 22 used Quads or triangles, they all had some nice 3D graphics for the timeQuote:
Originally Posted by zyrobs
No idea what the Konami board actually uses but the IBM system it is based on was advertised as being able to do points, lines, triangles, rectangles and quads: https://www-01.ibm.com/common/ssi/cg...7/ENUS7043-140Quote:
Originally Posted by Team Andromeda
As for Namco see https://www.youtube.com/watch?v=Bp6aHkZD2Wg
So not just SEGA then, thanks for the infoQuote:
Originally Posted by axel
Nintendo DS can also do quads natively. But it isn't obligatory and I think most (if not all) games use triangles. Or at least, I can't think of a DS game that is quads only.Quote:
Originally Posted by Team Andromeda
Never knew that, thanks very much for the infoQuote:
Originally Posted by axel