>>2931053 >it takes some voodoo to make quads do triangles though, that's the real problem
Not really, you just have to make two adjacent vertexes have the same xy positions. If your texturing hardware supports UV points (the Saturn didn't) then you won't even have a problem with textures needing to be adjusted.
The point is moot anyway since with perspective correction, triangles are objectively superior.
>>2928937 all the quads get converted to triangles in GPU so unless you've got a custom-made shader to which you send the extra vertex info for each triangle you won't actually get that result. That isn't the reason quads are used at all.
Quads are favored for wholly different reasons related to topology flow and mesh optimization.
>>2931212 i dont think people care about consumption or noise that much. also of course if you use the newest of the newest you can get the same performance for half the consumption or w/e fanless computers used to be the norm i wonder what was the first computer to have a fan on the cpu
In fact quads are a liability due to implementation dependent behavior when the 4 vertexes of the quad are not co-planar (which, to a small degree, is nearly always due to finite precision math).
The only motivation I can guess for why people may have wanted quads over triangles is that somehow they thought it is more intuitive to map a rectangular textures to rectangular primative. But once you understand the basics of texturing, it is trivial to see how to make quads from 2 triangles.
>Quads are favored for wholly different reasons related to topology flow and mesh optimization. Ok, I could imagine some processor designers trying to optimize hardware for certain data structures, but even so, it takes the same amount of data to specify both triangular and quad meshes.
>>2931212 Part of it is because with the amount of people in PC gaming now compared to then there's a lot more reason to dickwave using silly things like GPU noise to show newcomers that you're a "real gamer" or something like that. Another part of it is that games now are photorealistic and there's a part of our brains that associate noisiness with computing power, and you need a lot of computing power to drive photorealism.
i don't know much about 3d rendering but every developer i've heard from hated working with quads.
the reasons i hear are - vertices defining a triangle can only be co-planar (apparently easier for texturing and lighting angles) - triangles are easier to fill (e.g. with a texture or colour) - easier to interpolate\rasterize - tris are more popular
>>2931880 you can have perspective correct texture mapping with any set of ngons. triangles happen to be the most popular, at least nowadays, because it innately a much more consistent shape than other polygons
>>2934348 Though the GPU has turned into a very different from its introduction. They were initially more simple fixed function devices devices but they've turned into these programmable beasts. It is pretty much a second but slightly different processor now.
>>2934369 a processor that's used exclusively for number crunching (scientific/niche use) and gaming. The primary work horse is weaker than the specialized "support" core. The infrastructure (PSU and busses) is selected for that rare/unusual peak load of the supporting core completely out-juicing the primary work horse. I understand just fine why GPUs are the way they are, but it IS a bonkers situation.
>>2931198 They're actually not that loud, the VSA-100 chip was only on the borderline of needing heatsink/fan. They didn't really push the chips too much. Also the fact that its an unreleased prototype, the consumer version only had 2 GPUs.
>>2934392 the 2GPUs were the Voodoo 5 5500. The 4GPUs were the Voodoo 5 6000. It's not so much consumer version vs. prototype, but subsequent iterations of the line. 3dfx just went under before the 6000 got out.
>>2932149 It was in the early days of 3D accelerators, no one knew how to implement it properly, and quads did have a few quirks that made them desirable. I can see them as a logical successors of 2d accelerators which wrote squares into memory; quad based 3d accelerators did the same but merely allowed you to pick all 4 vertexes freely. So implementation might have been significantly easier.
Using triangles also limited pretty much all games to look the exact damn same for a long time, with only the quality increasing. It wasn't until pixel shaders did we finally get more unique effects.
>>2932174 >I guess no one was certain which way tech would progress.
Bingo. Remember that the NV1 predated DirectX. Also, the NV1 could draw 9-point quads. You had the four corners, plus the four center points between each adjacent vertex, and the middle of the quad, and could specify a coordinate for each one. The hardware than splined the values together to draw curved quads. It could draw perfectly curved surfaces better than any other 3d accelerator until shaders became commonplace.
Unfortunately I don't know of any software that used them.
>>2933284 Sega and Nvidia had a deal at the time, but it fell through when they couldn't provide a chip for a new console / upgrade for the Saturn (this was back in late 1995!).
>>2934378 >I understand just fine why GPUs are the way they are, but it IS a bonkers situation.
Not really if you consider that they pretty much have supercomputer levels of power.
Also a lot of the power draw comes from the insane memory speed the cards require. Look at the AMD Fury Nano, it ditched GDDR5 and got power usage cut in half (they then used the surplus to pack in 1/3rd more number crunching power to it, and it was still possible to cool the card with a single fan).
>>2934401 It is not clear whether they even wanted to release the 6000 as I recall, due to it being problematic and requiring external power (that they had to supply via an extra power brick, since PCs at the time had small 150-200W power supplies with low efficiency).
>>2934410 >Not really if you consider that they pretty much have supercomputer levels of power. Again, the bonkers situation is that you HAVE a supercomputer level of power SUPPORT CPU in your machine, that's only used in a FRACTION of the usage scenarios, but requires practically the MAJORITY of the PSUs power.
>Also a lot of the power draw comes from the insane memory speed the cards require And yet we decided it's an ok thing to have an EXTENSION CARD, dominate the machine like that. What anon in >>2931212 originally pointed out is that users and reviewers at the time were shocked about the outrageous requirements in terms of physical size, power supply and bus support. While nowadays these things are shrugged off as normal and desirable, as witnessed by your posts. Nobody, not even that anon, questions why this hardware has these properties. The insanity is merely that we accept these former exceptions to be the norm, and even required. In a way it's a hardware bubble. We bought advances in GPU capabilities by scaling the cards up in terms of size, power, heat handling, etc. It would be sane to scale the hardware back down to normal, but that would require a standstill in terms of power and capabilities, and we can't have that. So we're stuck with the cards growing, their requirements dominating the computers. In the not so distant future the inside of a desktop will be a GPU with a motherboard attached to it, not the other way round, and everybody will nod and smile, because they're dulled towards this development that looks absolutely insane to anybody not "in the scene" of desktop gaming.
>>2934568 Look at the heat sink and fan of an Atom. People might have adopted the 6000 if it came as an evolution of the 5000 and not a simultaneous development. The 6000 was just too much at once, with more time they could have gotten the PSU or Mobo manufacturers into the game as well.
>>2934610 >it was a more natural progression The Voodoo 5 6000 is essentially a single card SLI system. Considering 3dfx did SLI before, this was a very natural step.
>Compare the Titans to the preceding and competing models Why? Anon's point was precisely that we lost sense a sense of reasonable. Just because many competitive models are also unreasonable, this is suddently fine? I'd say it's just confirming the original point made in >>2931212 >The fuck happened to people over the last 15 years? Holy fuck, everyone got stupid.
>>2934405 >Also, the NV1 could draw 9-point quads. You had the four corners, plus the four center points between each adjacent vertex, and the middle of the quad, and could specify a coordinate for each one Isn't this too much resource consuming operation? I mean, I heard that even modern PCs are slow in calculating accurate curves, but in 1995?
>>2934481 >Yeah, but only a year later after NV1 you had the N64's GPU RCP which had a programmable vertex shader.
Could the RCP draw something like this? Note that this isn't a sprite but a 6 poly cube drawn in real time. It could also have phong shading applied to it (according to the devkit, anyway).
>>2934775 >Isn't this too much resource consuming operation?
The card was specialized for it so I guess on the CPU side it wasn't resource consuming. Hit detection might have been more difficult, but I haven't ever checked how you do that in 3d, there might be a simple solution to it.
I look at this effect like the early attempts at tessellation - and even those came by in like 2003 or so. This card was in 1995.
>>2935164 >Take a look a page 34 of this magazine. It explains how curved surfaces can be generated using RSP's vertex shader engine.
That one just tessellates a simpler figure into a more detailed one along Bezier curves. But at the end of the day it was still rasterizing triangles. Meanwhile the NV1 was rasterizing NURBS. Not the same thing.
>>2936184 >So what happens if you use a quad where all 4 points aren't coplanar?
It depends on the hardware implementation. In the case of the Saturn, the concave part of the quad got curved. You could draw bowties or curved outer corners in 1 poly. Gouraud shading got screwed up a bit, but at so low resolution it wasn't a serious deal (getting gouraud shading working on the Saturn was problem enough).
I don't know how any other hardware would've handled it. The Saturn was perhaps the most prolific machine that used quads. Some sources claim the Model 1 and 2 used quads but I consider that hogwash.
>>2931212 >>The fuck happened to people over the last 15 years? Holy fuck, everyone got stupid. It's hardly surprising, look how everyone threw away their CRTs in favor of LCDs that even today are barely on par with the stuff that was coming out in the early 2000s.
Easily the biggest travesty in the history of hardware, but retards needed the screens they moved maybe twice in their lifespan to weigh less.
When I think about the kinds of displays we'd have now if consumers actually gave a shit about quality over minor convenience, it just depresses me.
>>2936582 Weight and size are major improvements, so flat panels are a good idea. Size even more so than weight. You can complain about the lack of SED panel, if you want. Though it seems OLEDs are a better option in the long term. Frankly, as much as you complain about people not giving a shit about quality, I'm not convinced that making phosphorus glow was the way to go. It's a really messy way to produce a "controlled" frequency.
>>2936879 >>Weight and size are major improvements, so flat panels are a good idea Why? Why do you need an extra 2 feet of room for a screen? What the fuck are you using the extra space for in the corner of a room?
The value was ENTIRELY on the side of manufacturers. Lower weights and sizes made transporting screens far cheaper.
LCDs can only dream of hitting "messy" phosphor refresh rates.
>>2936954 and nobody gives a shit, because the original post was about someone mentioning odd hardware that uses something that is not plain quads, and someone else had to go all technically correct on them, producing workarounds that contribute fuck all to the actual statement. For the record, if triangles could do what nurbs do, CAD and 3D modelling wouldn't use nurbs. They do, massively. The rasterizer for them just approximates them using triangles, at some fairly harsh expenses
>>2936964 >For the record, if triangles could do what nurbs do, CAD and 3D modelling wouldn't use nurbs. They do, massively. The rasterizer for them just approximates them using triangles, at some fairly harsh expenses
That's still using a triangle rasterizer. CAD and 3D modelling don't demand GPUs that use quad based rasterizing.
Of course faking nurbs with triangles will be less efficient than just doing them with quads, but the post was about why quads are better for 3D, and that as a statement is obviously false.
That doesn't mean that quads don't have native advantages in a few areas, but that overall they are less versatile and efficient.
>>2936984 >That's still using a triangle rasterizer. CAD and 3D modelling don't demand GPUs that use quad based rasterizing. Thanks for repeating exactly what I said.
>Of course faking nurbs with triangles will be less efficient than just doing them with quads That is not at all what I said
>but the post was about why quads are better for 3D No, >>2934405 said >the NV1 could draw 9-point quads. You had the four corners, plus the four center points between each adjacent vertex, and the middle of the quad, and could specify a coordinate for each one. The hardware than splined the values together to draw curved quads. It could draw perfectly curved surfaces better than any other 3d accelerator until shaders became commonplace. It was a side point, not about quads, but quirky 3D hardware. In fact, that anon clearly pointed out that quads are not the best general solution (just like triangles aren't), just that quads have some desirable properties, and at the time, some hardware manufacturers figured it's worth it to support these properties in hardware.
>That doesn't mean that quads don't have native advantages in a few areas, but that overall they are less versatile and efficient. That's what >>2934405 said, and here you are, loudly and proudly disagreeing with the point you think they made.
All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.
This is a 4chan archive - all of the shown content originated from that site. This means that 4Archive shows their content, archived. If you need information for a Poster - contact them.
If a post contains personal/copyrighted/illegal content, then use the post's [Report] link! If a post is not removed within 24h contact me at firstname.lastname@example.org with the post's information.