Chosen Few Posted November 9, 2011 Share Posted November 9, 2011 Kwakkelde Kwak wrote: The thing that really worries me, isn't land impact though or griefing. The real renderhell always was and always will be in attachments. So with a 255 PE as a limit for a linkset, you won't come close to the horrific 255 x 130k = 33 million faces you can have now for every attachment point. I think it all comes down to that. No idea what is reasonable for one attachment, but let's say 10k faces. That's 33k times lower than what's possible now. So 130k faces / 33k doesn't leave a lot of faces per object to say the least.... Now I am curious how many faces a 255 PE mesh can hold at the small size of an attachment. I'm sure it's a LOT higher than 10k, but closer to that than to 33 million. Maybe the best solution would be simply to limit the maximum total poly count that avatar can be wearing at any one time to a reasonable number. I'd define "reasonable" as something in the tens of thousands. Obviously, the billions that are allowed now are well beyond unreasonable, by any definition. If the limit were, say, 50K (which is still absurdly high for a realtime character model), that number could be divvied up any which way among all the attachments. They could all be in a single mesh, or it could be 1300 in each of 38 separately attached meshes, or whatever other combination the user wants, as long as the total stays under the cap. Ideally, I'd really like to get away from measuring everything in terms of prims. It's arguable that there's still merit to that for builds, since land continues to be the metaphor for resource allocation in SL. But for avatar attachments, which have nothing to do with land, we can easily break away from that mindset. I'd much rather make it so people have to pay attention to the real principles involved with mesh management, rather than the artificial "prim equivalency", which really has nothing to do with anything at all. LL already seems to be thinking along these lines, as evidenced by the fact that there's no land impact stat when examining an attached object. I'd just like to see them complete the picture, by giving avatars a limit on the remaining stats that makes sense. As soon as they do that (and once the bitching and moaning from the usual doomsayers subsides), I think people will automatically start thinking about the right things when making attachment decisions. I suppose attached prims would need to be either exempt from the count (unless linked to a mesh), or otherwise artificially adjusted for, since as we discussed, it doesn't take all that many toruses or sculpties or whatnot to add up to a substantially high poly count, and we're pretty much stuck with that. Like it or not, any new limitation has to leave existing content, and pre-existing methodologies, alone. I stil think 64K verts is ridiculously high even for a static object, though. But for now I'll concede that this is a slightly different topic. Kwakkelde Kwak wrote: For the celtic collar, well I think what I have built back in the day can easily be described as overkill, I wouldn't have brought it up if I thought otherwise. Understood. Kwakkelde Kwak wrote: But this is SL, not a game. You can't compare the two one on one I think. There are a lot of similarities, they use the same kind of software and push the same kind of hardware and building for either one means balancing looks and resources. But as I understand people want 80 fps for a game, where I think the 20-30 fps I normally get in SL are very acceptable for the purpose of SL, which is a social platform rather than a high performing game. 10-15 fps occasionally is no problem for me, I get annoyed when it drops lower. So following that logic I think you are "allowed" to push the sytem a bit more if that's what it takes to obtain a certain look. SL is not a game, but the assets we make for it are certainly game art assets. I try to use the phrase "for realtime" as often as possible, as opposed to "game art", in the hope that it will preempt any tangential philosophical discussions about what constitutes a game. But every so often, I do let the phrases "game art", "game artist", etc. slip out. It doesn't mean I'm trying to redefine SL. I have to say I find your point about FPS and acceptability to be very interesting, culturally. Since the dawn of SL, people have complained literally every day about its low FPS, and all kinds of discussions have ensued over the years about why we can't do better (prim builds inherently have tons of hidden faces, there's no control over UV's so texture inefficiency is par for the course, sculpties have fixed poly counts, etc., and that's before we even get to user mistakes, which are abundant, to say the least). But now that LL has finally given us a means to substantially improve the situation, all of a sudden it's, "SL has always had low FPS, so let's go right on using more resources than necessary." I really don't think anyone should adopt the mindset that they're "allowed to push the system". Even if things didn't get any worse than they are now, the fact remains that SL still can't run reliably on a huge portion of the computers in every day use on this planet, and can't run at all on a great many of them. "Pushing the system" will only preserve that unfortunate status. But the truth is things can get far, far worse than they are now. Poorly made mesh models can make even the most resource-abusive prim builds look like pinnacles of efficiency. I'd much rather spend my time showing people how to do the best possible job on every model, for all our benefit, than I would just saying what might be interpreted as, "Here's what you can probably get away with, so don't bother learning to do better." As I see it, the culture we SHOULD have among creators in SL is one in which people constantly strive to outdo themselves in terms of excellence across all relevant factors. SL builders have already proven quite well that they do think exactly that way, when presented with limitations that they understand. "Hey, check out my new shiny rocket ship. It looks kick-ass, and it's only 25 prims." We hear those kinds of excited comments all the time, with heavy emphasis on the "it's only ____ prims" part. There's absolutely no reason the same enthusiasm couldn't be developed for "it's only _____ polygons." Among non-SL realtime artists (game artists, VR artists, etc.), it absolutely has always been, "My new rocket ship is awesome because it looks killer, and it's only 200 polys." Again, there's no reason the same can't become true of SL artists. We just need to allow for people to expand their awareness beyond simply prims, prims, prims. I believe that paradigm shift will come in time, and the cultural consciousness will right itself. It has to. But habits are hard to break, and there's going to be natural resistance to change. So it will likely take a while to permeate. But that's OK. It's already begun. The mere fact that we're having this discussion is proof enough of that. Further, this notion of, "We all hated our low FPS yesterday, but some of us are now defending it today," is evidence of that both processes (the change itself, and the absolutely necessary accompanying resistance to it) are in full effect, I think. All that said, you did include the very important qualifier, "if that's what it takes to obtain a certain look". I'm certainly in full agreement with you that if there's truly no other way to get something to look the way you want it, then expending extra resources is the thing to do. However, that too has to include a qualifier: if the look you had in mind is truly the only way to go. Often, as artists, we need to take a step back from ourselves, and assess our work from a fresh point of view. We get so focused on the close-ups that we often just don't see the forest through the trees. I can't tell you how many times I've been faced with a resource restriction on a project (usually poly count) that I was not allowed to breach, and I've ended up having to start over once it became clear that the strategy I'd first employed simply wasn't going to work. In nearly every case, my redesign has been just as good looking as my original intent, even if not quite the same. SL artists are already well accustomed to this when it comes to working around alpha sorting problems. "Take a look at this new house I made. At first, I wanted to design the windows this way, but that couldn't work, so I came up with that way instead. I think it worked out very well in the end." Once again, there's no reason that kind of thinking can't be applied towards all aspects of mesh modeling. "I had originally wanted to do it this way, but the display cost was super high. So, I did it that way instead. It worked out great." Kwakkelde Kwak wrote: Alpha layers can't come close to actual prims with metal shine. Maybe that's because alpha layers don't exist. Alpha CHANNEL! Anyway, your point is certainly debatable. For certain things, I'd have to agree with you, but for an awful lot of things, SL's shine shader just looks really terrible. A well textured image of a piece of metallic lattice can often look a thousand times better than a bunch of shiny prims. On a side note, it is unfortunate that SL forces users to make a choice between the two. There's no technical reason why an object can't be both shiny and transparent. But I'd be willing to bet there's a whole army of SL artists out there who think there's some kind of rule against it, across the whole of 3D modeling. If so, it's just one more example of how the SL-centric experience gives people an inaccurate impression of what 3D modeling really is. (I've always assumed this particular restriction was in place in order to negate the need for shininess maps in SL, which if present, would increase the texture load per object.) Kwakkelde Kwak wrote: I do agree that when you can build an object at the fraction of the weight (and this is ofcourse measurable) and at 90% of the looks, you should go for that option. But the 90% is very subjective in two ways. The number itself is subjective, some might say 40% of the rendering cost and 60% of the looks should be the turning point, others 5% and 95%. Next to that it's very subjective what "90% of the looks" means. Yes, it's totally subjective. There's obviously no way to quantify aesthetics. I was merely speaking figuratively (no pun intended on the word "figure" there, by the way, either). My point was simply that if something can be obviously recognizable as what it's supposed to be, with just a low expendature of resources, there's no reason to sink massively more resources into it, just because you might think it'll end up looking slightly better. And, of course, the verse is also equally true. If something can be made to look massively better with only a slight increase in resource usage, then the increase is probably justified. The governing principle in both cases is this. Make the best looking object you can, utlizing the fewest resources possible, always. The precise optimum balance point is going to be a matter of opinion in every single case, of course. But we can usually all agree on what is obviously wasteful, and what is obviously not, when we see it. In RL, these same kinds of decisions are made all the time. A structural engineer might change an architect's blueprint if the available building materials simply won't support the design safely. An auto parts manufacturer might alter the shape or material of a particular part if the original spec is too expensive to produce. A dieter might give up his or favorite food if other foods are markedly healthier. Even non-humans know how to allocate resources. A predator will let the biggest meatiest prey go if a perhaps less nutricious, but easier-to-catch, straggler is avaialable. It's no different in a simulated world. Resources are still finite, even if imagination is infinite. The more of resources are used up in any one item, the worse off we all are, for all kinds of reasons. Therefore, we should all strive to use as few resources as is possible and practical, at all times, in all things. I think we're all in agreement on the principle itself. We'll have no end of disagreements on what is the best balance for any particular model. That's fine. That's what discussion is for. In every case, either one point of view will win out of the others, or we can all just agree to disagree agreeably. Either way, informed decisions are being made, and that's what's most important. Kwakkelde Kwak wrote: As for your example, first of all, looks good and sounds very reasonable to render. But it's nothing like the tubular celtic thing I have made. Mine is a real mesh (yes let's confuse everyone by using this term..I mean in a chickenwire way of mesh, not a 3D program mesh), so two polys per section won't cut it, maybe 6 would be enough rather than eight , but two certainly won't. Ah, it would seem I misunderstood what you meant by "Celtic collar". I had though you meant something like this, a leather collar with Celtic knot medallions and such attached to it. I do have to say yours is pretty cool looking. As for the model itself, you could certainly do it with just two or three quads per section. Keep in mind that the avatar's neck will cover the inward facing surfaces, so those faces don't need to be there. The larger gauge upper and lower rings will do just fine with 6 quads per section, and they'd look great. Here's one I just made now, based very loosely on your picture. It's not nearly as well made as yours, since I only spent a few minutes on it, but that's obviously not the point. The point is it's just 1242 polygons. Had I been willing to put more time into it, it could be just as nicely built as yours, and still have the poly count of just two or three toruses, tops. With the exception of the odd misplaced ring here and there, which only exist because I did such a rush job on the thing, you'd never know that the inner faces are absent, or that the small gauge wire is only two quads per section. This would be even more true with proper texturing, rather than just the base material in the screenshot. Admittedly the large gauge parts could look better. I used four quads per section instead of six. They really need six. Realistically, it would probably come out to an average of four, anyway, once all the faces obscured by the avatar body are removed. I didn't remove any faces from those parts in this version. Kwakkelde Kwak wrote: The doorway (5.0x0.9x7.0m which is rather large for a RL door, don't blame me, blame the person who wants it:)) The arch has 16 sections I think 24 sections, which look like the above in the picture below... I could make the section slightly different, using 20 instead of 24 faces...or use a box with textures, which would still be 8, again this is very subjective. All I ment when I said the doorway doesn't have sections like you descibed, was to show out of the 800 faces, none are useless. I never thought you ment me, afterall you've never seen the thing:) 24 sections is a lot for a half arch. All the objects people are used to defining as "round" in SL have 24 sections for a full circle. So, for a half circle, you only really need 12, and everyone looking at it will interpret it as perfectly rounded. That would save you nearly half your poly count right there. Since you say there will be a lot of these on the building, it's well worth doing. If you then make each section look like the lower of your two sketches instead of the upper, you'd save an additonal 96 polys. It would be slightly less texture efficient, but nothing that would necessarily cause you to have to jump to a larger texure size, so no real loss there. Also, the sections of the vertical pillars can obviously just be two tris per side. If you make them by extruding, you'll end up with eight tris on the underside of each section of the upper modling, and another eight on the top side of the base molding. Again, there's a slight loss of texture efficiency in using only two tris for each, but not enough to gripe about. Those faces are so small, they're not going to get many pixels, regardless. Here's how I'd do it: Those are 422 polys each, including the wall sections. The arches themselves are 404 polys each. So, going this route would save you around 400 per doorway. Multiply that by the 15 doorways you said would be present, and that's 6000 polys, hardly an insubstantial amount. If your 40 windows are of similar design, that's another 16,000 polys saved, wich is huge. Now, let's weight the aesthetics against the resource usage. Does the 12-section arch look as smooth as your 24-section one? No, it doesn't. But it still looks very good, certainly as good as the average game asset model, and definitely no worse than a prim arch would look. The poly count was cut in half, while the visual quality only dropped ever so slightly. Therefore, the prim low prim count wins out. When you put it context of the entire building, the aesthetic difference betwen 24 and 12 sections per arch becomes even less noticeable. It's not like anyone's going to arrive at the parcel, take a look around, and go, "That would be such a nice building if only the archways were just a hair smoother." No one's ever going to notice at all, unless you park the two versions right next to each other, and ask them to make a conscious effort to tell you the difference between them. Add all that up, and it gets really hard to justify 800 polys per arch. This is why I said an arch with that many aught to be pretty elaborate. For such a simple design, a nice low-poly approach works perfectly well. Link to comment Share on other sites More sharing options...
Kwakkelde Kwak Posted November 10, 2011 Share Posted November 10, 2011 Chosen Few wrote: Maybe the best solution would be simply to limit the maximum total poly count that avatar can be wearing at any one time to a reasonable number. I'd define "reasonable" as something in the tens of thousands. Obviously, the billions that are allowed now are well beyond unreasonable, by any definition. If the limit were, say, 50K (which is still absurdly high for a realtime character model), that number could be divvied up any which way among all the attachments. They could all be in a single mesh, or it could be 1300 in each of 38 separately attached meshes, or whatever other combination the user wants, as long as the total stays under the cap. I agree, I really do, a limit would be good for performance. But I don't see from the top of my hat how one would implement that, given the fact we agree we are stuck with prims and their use won't be changed. Another simple calculation: the maximum number of attachment points is 38, with 255 prims each. The worst prim I can think of is a hollowed torus with a very small cut, within the first loop of faces. So we have 38 x 255 x (2 x 24 x 24 + 2 x 2 x 24) = 23 256 000 faces as a maximum pre-mesh, exluding the avatar. I'm not saying this is any good, but again, we're stuck with these numbers. There will be hell for LL if they'd ever dial this back. I'll assume people aren't wearing anything like this, but with a detailed pair of boots, some funky hair and a rediculous celtic collar worn, you'd be well over the 10s of 1000s of faces already. You wouldn't be able to add any mesh object to your av, but you could happily keep adding prim gloves, weapons, bracelets etc. This doesn't sound very fair to me. Personally I wouldn't mind if they did it, but lots of builders and buyers will be stuck with useless items. It's the same with scripts, I don't hear anything about the memory cap after all the moaning. (Not that I am looking for any news on that btw, so I could be wrong). Chosen Few wrote: Ideally, I'd really like to get away from measuring everything in terms of prims. It's arguable that there's still merit to that for builds, since land continues to be the metaphor for resource allocation in SL. But for avatar attachments, which have nothing to do with land, we can easily break away from that mindset. I'd much rather make it so people have to pay attention to the real principles involved with mesh management, rather than the artificial "prim equivalency", which really has nothing to do with anything at all. I agree on the first comment, although a "prim" is a concept people can understand, since they are familiar with it. Maybe "prim box equivalent" is a better term... or maybe the community needs to be educated and start thinking in polygons, quads, triangles, vertices etc. I don't know. I however don't see why you could compare meshes with prims when they are rezzed on land and not when worn. The prim equivalency is connected to the rendering cost (and other costs) afterall. Chosen Few wrote: I really don't think anyone should adopt the mindset that they're "allowed to push the system". Even if things didn't get any worse than they are now, the fact remains that SL still can't run reliably on a huge portion of the computers in every day use on this planet, and can't run at all on a great many of them. "Pushing the system" will only preserve that unfortunate status. Heyhey! I said push it "a bit". (This is wide open to interpretation ofcourse) All I ment was I don't think as a builder you should restrict yourself in such a way you would when building for a high end action game, where you really need to maximize performance to make the program function. Also in a game all objects need to match eachother. I think SL objects can have a high "look at me" factor, rather than blend in with the scenery. In SL there are lots of different "languagues" of building...mix 'n match. Everything within reason ofcourse (again very subjective), I didn't mean stress the servers and processors and graphic cards even more than is happening right now. The arrival of sculpts gave us the opportunity to make things look nicer (and save some "prims" here and there). The mesh gives us the ability to work more efficiently, but ALSO to make things look even better than sculpts. The way you are describing how you want mesh to work is one I can't share. It sounds like you want to make SL looking slightly worse with a huge performance gain. I'd rather see the mesh push the looks of SL to a new level with similair, yet slightly improved performance. We disagree, that's fine. (And no I am not on a high end gaming computer by any means, my buzzing black box is over 4 years old with a graphic card upgrade about a year ago) There are two things people can't stop complaining about: the first you already mentioned, that's bad performance, the other one is outdated looks. Both will never match up with professionally made programs, but both may be lifted a bit if you ask me. Chosen Few wrote: It's no different in a simulated world. Resources are still finite, even if imagination is infinite. The more of resources are used up in any one item, the worse off we all are, for all kinds of reasons. Therefore, we should all strive to use as few resources as is possible and practical, at all times, in all things. Hmmm, in most cases..yes. Don't forget in RL you can't put your car in your backpack and in SL you can. When resources are scarse people find ways..like in RL So if you only have 10 prims left on your parcel, you can sit on one chair on monday and replace it with a new one on tuesday. The same way of reasoning can be applied to performance resources. Anyway I think we agree on this, it's comparable to your "1300 faces per attachment point on average" suggestion. When you run low on resources in SL, you replace rather than add. Chosen Few wrote: As for the model itself, you could certainly do it with just two or three quads per section. Keep in mind that the avatar's neck will cover the inward facing surfaces, so those faces don't need to be there. The larger gauge upper and lower rings will do just fine with 6 quads per section, and they'd look great. Here's one I just made now, based very loosely on your picture. It's not nearly as well made as yours, since I only spent a few minutes on it, but that's obviously not the point. The point is it's just 1242 polygons. Had I been willing to put more time into it, it could be just as nicely built as yours, and still have the poly count of just two or three toruses, tops. Seeing it, three polys per section might work, although I'd add a backface. I think the 16 sections for a ring don't cut it though. Also You've skipped quite a lot of rings it seems... Anyway, your models proves what we both already knew. Mesh is far far more efficient than prims. (Your model is about 60 times lower in faces than the prim one?) I might redo this one at a certain point, but it's not on my todo list right now.... Chosen Few wrote: 24 sections is a lot for a half arch. All the objects people are used to defining as "round" in SL have 24 sections for a full circle. So, for a half circle, you only really need 12, and everyone looking at it will interpret it as perfectly rounded. That would save you nearly half your poly count right there. Since you say there will be a lot of these on the building, it's well worth doing. If you then make each section look like the lower of your two sketches instead of the upper, you'd save an additonal 96 polys. It would be slightly less texture efficient, but nothing that would necessarily cause you to have to jump to a larger texure size, so no real loss there. Also, the sections of the vertical pillars can obviously just be two tris per side. If you make them by extruding, you'll end up with eight tris on the underside of each section of the upper modling, and another eight on the top side of the base molding. Again, there's a slight loss of texture efficiency in using only two tris for each, but not enough to gripe about. Those faces are so small, they're not going to get many pixels, regardless. Here's how I'd do it: [PIC] Those are 422 polys each, including the wall sections. The arches themselves are 404 polys each. I didn't add the size just for the fun of it and the arch isn't half a circle. Where a circle has one focal point and an ellipse two, this one has three (I think the English term is surbased arch). That means there's a lot more curve in the ends. On the other hand a lot less in the centre. The fact the prim ring or cylinder has 24 sides doesn't mean this is what one should aim for I think. At the current size this would mean .5 meters per section, which in my opinion is too little for this particular arch, in this particular build, for its particular purpose. If this was a building block to go on sale as building block I might lower the number though. I used the 2 faces instead of 8 where possible, I think my second of the two sketches shows that I already grasped that idea. (Always a good thing to post these things, since everyone should be aware of these "tricks") One bottom is sloped, but it's 4 faces instead of 8 by lenghtening it into a pyramid. I don't think there's any room for improvement without changing the shape. I'm a bit confused on the polycount though (where I suspect your poly is a triangle, more confusion, people should talk in quads and tri's since they can both be polys, this includes me:)) My arch "as is" has 24x24=576 triangles (hey half a torus!) I don't see how yours has 404. I also don't see how you managed to build the rest out of 16 triangles. I guess by arch you mean everything except the walls..duh, nvm. If I replace all the sloped pieces by flat ones I would come up with 116 faces less, maybe that is a justified loss of detail or change of shape for that matter. my model wouldn't be 766, but 650 faces. I don't agree on the arch itself, not in this particular case:) Link to comment Share on other sites More sharing options...
Chosen Few Posted November 10, 2011 Share Posted November 10, 2011 Kwak, I just spent the better part of the last couple hours, writing a really good post, in response to the points you rasied in your last post. Unfortunately, a browser hiccup just ate it. :matte-motes-angry: I'll rewrite when I find some time later. For now, I'll just say I'm more in agreement with you than you seem to think, on a lot of the most important points. It appears there are a few points of mine that you might have misinterpreted. I'll explain more later. Thanks for such a great discussion, by the way. Technical difficulties notwithstanding, this has been highly enjoyable so far. Link to comment Share on other sites More sharing options...
Kwakkelde Kwak Posted November 10, 2011 Share Posted November 10, 2011 Oh as I said earlier we're on the same page alright, different paragraph maybe.... But one has to play the devil's advocate here and there to make a point or to keep things enjoyable hehe... very annoying your post was eaten, that happened to the first half of mine aswell yesterday...looking forward to your reply Link to comment Share on other sites More sharing options...
Facebook
Instagram
Twitter
YouTube
Flickr
Recommended Posts
Please take a moment to consider if this thread is worth bumping.
Please sign in to comment
You will be able to leave a comment after signing in
Sign In Now