Jump to content

NEVERMIND


Guest
 Share

You are about to reply to a thread that has been inactive for 3684 days.

Please take a moment to consider if this thread is worth bumping.

Recommended Posts

The whole avatar is only 7,000+ faces. Your necklace is 5+ times that. Have you looked at it's render cost? Have you considered what it does to lag?

And you are skipping LOD's?

The MAV_BLOCK_MISSING is likely a warning that the automated LOD generation is taking out all the faces assigned to a material for one of the LOD models. Make you own LOD's and be certain you have all the materials assigned to at least one face in all the models.

Link to comment
Share on other sites

The SL avatar also looks like **bleep**. It's not a necklace it's chains for a shirt and lowering the poly count on them anymore will make all the chain links look a mess. This worked for me, perhaps it's not politically correct but it works so I'm happy 

Link to comment
Share on other sites


Lexi Zelin wrote:

The SL avatar also looks like **bleep**.

 lowering the poly count on them anymore will make all the chain links look a mess

perhaps it's not politically correct but it works so I'm happy 

Wow. This will end well.

If you need that many triangles for an attachment, you're doin' it wrong.

Link to comment
Share on other sites


Freya Mokusei wrote:


Lexi Zelin wrote:

The SL avatar also looks like **bleep**.

 lowering the poly count on them anymore will make all the chain links look a mess

perhaps it's not politically correct but it works so I'm happy 

Wow. This will end well.

scared.gif

 

( :smileytongue: )

Link to comment
Share on other sites

as a little side note, your MAV_BLOCK issue might also be caused by the number of vertices per texture face. Note that as soon as the number of vertices goes above 21844, the SL Importer will split your texture face into 2 or more with 21844 verts per face at max.

And this almost ever causes problems with MAV_BLOCK when you use the LOD generator.

The reason is simple: the LOD generator reduces the number of verts. When one face that has > 21844 verts in Highest LOD gets reduced to < 21844 in a lower LOD, then the extra texture face is no longer generated and thus there is now one texture face missing in the lower LOD's...

Actualy this sort of issues could also be solved by the importer. However creating meshes with so many faces is rarely a good idea. Unless you make SIM wide builds maybe.

  • Like 1
Link to comment
Share on other sites

You don't have to beleive me. But, Gaia gave you better details for the cause of the problem. She worded it differently but we are saying the same basic thing.

The avatar has problems. But, it is reasonably efficient. Its problems are not going to be fixed by adding tens of thousands of unnecessary faces.

I think most of the modelers reading your post realize that a 23,000 face necklace is a lag maker. Have you even bothered to check the render cost?

While I am glad it works for you and you are free to make whatever you want in SL, we are also free to point out the flaws in your thinking.

There are far more efficient ways to make the necklace and solve the materials error. 

Current viewers are having a problem maintaining frame rate when in a crowd of avatars now. Toss in a couple of your necklaces and I expect everyone will be running at single digit FPS. But, you and unaware buyers are free to do that. But, expect blow-back at some point as users figure out the problem.

Link to comment
Share on other sites


Lexi Zelin wrote:

The SL avatar also looks like **bleep**. It's not a necklace it's chains for a shirt and lowering the poly count on them anymore will make all the chain links look a mess. This worked for me, perhaps it's not politically correct but it works so I'm happy 

Then I hope you'll be happy when I call you out publicly for being a walking lag bomb if I run into you at a club.

I'm getting tired of running into Ava's who's render weights are topping 500,000 units.

A couple of us had to have a talk with one of our friend because everytime she'd show up we'd watch our frame rates drop.  Certainly we didn't like telling her that she couldn't do what she was doing but she was her own walking lag bomb.

Link to comment
Share on other sites

I've been 3D Modeling since 2005, that's when I first saw how optimized models where in games like "FarCry" and "Doom 3". To my absolute amazement they solved their visual problems by using Normal Maps. I understand Normal Maps can be pretty intimidating to try and learn, but unfortunately this is the way things are properly done in an effort to not use so many triangles on a model. And in 3D Modeling for Games all objects have proper LODs created. The Second Life uploader is a decent repalcement for creating Mesh LOD's in my opinion, for those new to making 3D Models.

There is a discipline in 3D Modeling that I've strived to live by ever since 2005, and because ot that discipline I begin to sweat whenever I am building anything over 7000 triangles, but that depends on the size. But for a Chain 37,824 is an extreme amount of triangles. There have been games in the past that used that as their triangle budget for an entire game level. 

This triangle amount would be pretty bad if it was a large object, it would affect people's GPU's at a far distance, but since its a Chain that size it only affects people who will be upclose to your Chain. And when I say affect, I mean 1 - 2 frames per second, because their is still a high percantage of people whos computers can't handle that many amount of triangles - this Chain and its triangle count + Avatars + Enviroment and Props.

 

Here are a few threads that need to be read and considered by anyone who is interested in making meshes for Second Life. Another thing that should be kept in mind in reading these threads is the polycount for the type of object and how big is that object... Character, Weapon, Prop, etc...

 

Historical Poly Counts

http://www.gameartisans.org/forums/threads/23520-Historical-Poly-Counts

 

Mesh Polygon Count for PC Game

http://forum.unity3d.com/threads/140675-Mesh-Polygon-Count-for-PC-Game

 

The Theory of LOD – Understanding LOD in Mesh can help reduce land impact #SL 

 

http://www.lokieliot.com/blog/?p=670

 

Link to comment
Share on other sites


Perrie Juran wrote:

I'm getting tired of running into Ava's who's render weights are topping 500,000 units.

The problem with these kind of setups (very detailed high LoDs and minimal low LoDs) is that by the SL calculation the render cost will probably be not all that high. From a distance you won't have to render the object at all, which seems to justify the relative low render weight. However, in SL people usually flock together, so especially for attachments it's the higher LoDs that determine the rendering performance (lag).

Anyway..that triangle count is completely unjustified, regardless of the weight SL puts on it. My experience is that SL runs well on my old computer (which is probably about as powerful as the average user's computer) when the total triangle count on screen is under 700,000. 5% of that for just a small attachment doesn't fit that picture very well.

Link to comment
Share on other sites


Kwakkelde Kwak wrote:


Perrie Juran wrote:

I'm getting tired of running into Ava's who's render weights are topping 500,000 units.

The problem with these kind of setups (very detailed high LoDs and minimal low LoDs) is that by the SL calculation the render cost will probably be not all that high. From a distance you won't have to render the object at all, which seems to justify the relative low render weight. However, in SL people usually flock together, so especially for attachments it's the higher LoDs that determine the rendering performance (lag).

Anyway..that triangle count is completely unjustified, regardless of the weight SL puts on it. My experience is that SL runs well on my old computer (which is probably about as powerful as the average user's computer) when the total triangle count on screen is under 700,000. 5% of that for just a small attachment doesn't fit that picture very well.

I know I am a little sensitive about this subject and I also know I am far, far, far from being an expert on it.

I also know that correlation does not equal causation and I can't from a technical point of view proove my case.

Also the last thing I want to be is an ARW Nazi.

And I certainly don't go around all day long checking ARW's.  I've got better things to do with my time.

But what I do know is this, that as you said, when a number of Ava's congregate it begins to affect everyone, and the higher the average render weights are the more lag I see and also the more at risk I am from crashing from the texture memory discard problem.

I also see that the OP has exited this thread.  Rather than asking for help on how perhaps he could make these chains without possibly causing a negative impact.  Nice.  :(

Link to comment
Share on other sites

Whether your customers recognize the cause of the problem or not, you should be well enough educated to realize there is a problem.

Since you didn't answer my question on having looked at render cost, I'll asume you haven't. That is sad. In the SL Viewer Top menu Advanced->Performance Tools->Show Draw Weight for Avatars.

When I buy mesh I look at the demo to see how dense the mesh is. If it is too dense, I won't buy it. I am not alone in this checking. And there are those that review mesh clothes and point these types of problems out.

My belief is those that make good products consider the impact their items have on overall render time. Over time they get recommended by various bloggers as being a good designer. that word spreads through the fashion community. Those that don't often get pointed out too. On occassion someone gets upset and starts a campaign to shame a designer into better behavior and design practices. Some have even been known to show up carrying signs to protest a designer. Others have review bombed designers to shame them. All these things will affect sales. 

Some ARC Nazis run around with Avatar Render Cost turned on. Expect them to rag on your customers and spread the word about your products. While the little bit of stuff I checked in your store, has a reasonable render cost (3k units for a top) the item shown in you orignal OP would way exceed what you typically market. So, not having compaints from those things, suggests you haven't built this way before. 

How you handle all the information people have given you is up to you. People here have tried to help you. Do what you will with the information. But if you really went to the trouble to trademark AngelRED Couture™, I would think you would want to protect its reputation.

You are free to make whatever you want at whatever render cost will get by the Linden limit filters. You have already run into one problem from high poly count. Expect more if you continue thinking small items with tens of thousands of polygons aren't necessary.

Link to comment
Share on other sites

I imagine some can argue that Computer Hardware gets faster, cheaper, and more capable every year, that soon using tens of thousands of triangles on a mesh won't be a problem. That may have been true at some point but now people are moving on to Tablets and portable devices and the computer industry is suffering because of it.

A counter argument I would expect at this point is "Well the Nvidia Tegra K1 should take care of that problem." This mobile processor from Nvidia called the Tegra K1 is capable of rendering PC quality graphics on a mobile device, I didn't for the life of me think we would be this close to having capable hardware of this quality on mobile devices this soon.

The problem with this hardware is that it still relies on portable power supply solutions. A mobile device dosent have the privilige of a huge power supply that you see in PC's to render lots of triangles on screen. A portable device will be able to render PC quality graphics for a several hours before its battery runs out. Sound building practices started years ago in an effort to relieve the burden that the GPU has when rendering too many triangles. Normal Maps aren't just a nice graphical feature introduced by Linden Lab, they have been around for Ten Years and they are the Holy Grail of building resposibly.

Now that people are moving onto less capable portable devices resposible building is just as important as it has always been. If Second Life residents ever hope to see SL on a mobile platform for free, and not limited in graphics, people need to consider building responsibly with mobile devices in mind.

Link to comment
Share on other sites


RipleyVonD wrote:

I imagine some can argue that Computer Hardware gets faster, cheaper, and more capable every year, that soon using tens of thousands of triangles on a mesh won't be a problem.

A question here is do we actually know what the limits are here?  Eventually this 'faster, cheaper, etc, is going to hit  a Brick Wall. 

Consider Moore's Law.

On 13 April 2005, Gordon Moore stated in an interview that the law cannot be sustained indefinitely: "It can't continue forever. The nature of exponentials is that you push them out and eventually disaster happens".

I'm hoping we can go a lot further than we have but eventually we are going to hit that Brick Wall.

At some point in time the atoms are going to say you can't push us any harder and start to push back.

 

Link to comment
Share on other sites

There is something that keeps running through my head as I read about the controversy of large numbers of vertices making up a mesh.

I know they are not treated the same and I am sure there is a technical reason why one is not nearly as taxing on the computers processors but I can't help but wonder.  If vertices are really just points in 3D space in XYZ how are they really so much worse than pixels on a texture?  I mean doesn't the computer have to calculate the position of each pixel on every texture?  Basically isn't each pixel just a colored point with an XYZ coordinate?

A 1024 by 1024 has 1,048,576 individual pixels that the processor has to calculate their position in XYZ coordinates.  Multiply that by a possible 8 textures per mesh object and that is 8,388,608 individual points the computer has to deal with not to mention if each of those 8 textures have Nomral Maps and Specular Maps that were all 1024 by 1024 and we are talking about 25,165,824 bits of information all that has to be calculated and XYZ coordinates determined.

My gut feeling is LL has put way to much emphases on vertex counts and all along it's been the sheer number of big textures and scripts that avatar's are wearing that is causing the biggest amount of lag.  When LL allowed basically unlimited number of objects per attachment point that was the biggest mistake they could make.  I know it started in a third party viewer and LL seemed to have no choice but to let it continue.  I bet you that is why LL reined in the their party viewers with the shared experience rule.

Link to comment
Share on other sites


Cathy Foil wrote:

I know they are not treated the same and I am sure there is a technical reason why one is not nearly as taxing on the computers processors but I can't help but wonder.  If vertices are really just points in 3D space in XYZ how are they really so much worse than pixels on a texture?  I mean doesn't the computer have to calculate the position of each pixel on every texture?  Basically isn't each pixel just a colored point with an XYZ coordinate?

The short and easy answer to this question is 'no'. I'll see how well I can explain using general terms, not linked to specific technologies or tools.

While the quality issue is mostly a strawman in this instance (1024x1024 textures are horrendous lagmonsters, LL just missed the boat on taxing them), it's not quite as simple as pixels being easier than vertices. The first difference is that textures can typically be stored as lossy - not all colour information needs to be retained, the remainder can be determined using clever maths. This provides compression. The second is that all images in SL are stored in an RGB format (this might've changed, but alpha used to be stored as a separate B/W mask), which - while having a diverse colour palette - only occupies 24-bits per pixel, and an integer (whole) number. Vertices are stored as floating points in X, Y and Z dimensions - three values - a each floating point might be as many as 24 or 32 bits.

All of this is further complicated by topology, rigging, normals, lighting, and probably a bajillion other things. To correct your post a little, pixels have known locations which don't need to be stored (especially not in a 'Z' axis) - the element about them that changes is the colour - almost certainly using fewer bits per point.

Anyway, the real weight of mesh doesn't come from storing mesh. Rendering a texture is incredibly easy, graphics cards are designed to display many textures simultaneously (they have special routes to make this as fast as possible). There's a few ways to do it, but the easiest to explain is 'wrapping' or 'baking', where it works like textures in SL - a flat image is wrapped around a shape. In this model, rendering the deformation of the texture (as it disappears around the side of a cylinder, for example), it takes substantially more computing power to calculate the angle of the cylinder and the distortion of the texture being applied by this shape, than the texture itself. Rendering in 3D is slightly more annoying in SL than in other places, since OpenGL places its own significant performance overhead on users.

Blaming people for making use of multiple attachment points is a mistake - it's always been possible to attach 35 * 255 prims to an avatar (this didn't change). The only way things will ever get less laggy in SL is for creators to take some responsiblity for the things they build.

Understanding is the first step, and the OP is wasting an opportunity here. A respectable creator would be more interested in serving their '7000 customers' well and more efficiently. Once mesh is integrated into an attachment, it's difficult for non-technical consumers to analyse individual influences of visual lag - especially in a complicated scene. The OP, as a creator, has control over how complicated their products are - and should be using sensible scales of optimisation so that their customers aren't being weighed down by things they may not realise were poorly built.

Using the maths in this thread already (which is being under-valued), imagine that in order for the OP's outfit to be worn fully, 5 avatars had to be derendered. On any sim where you wanted to wear it, you would have to drop 5 avatars from view - otherwise your FPS will drop. Is this single object worth 5 people?

Intentionally releasing high-lag items - in my personal view - is almost analogus to having intent to defraud. The creator is laying a 'trap', and when it is discovered by savvy consumers they will detach these incredibly heavy objects and never wear them in public again.

Link to comment
Share on other sites

If i understand your remark correctly, then i think that you overlook one important aspect. Whatever is in the scene is not rendered just once and shown as a static 2D image. Your camera moves, the objects in your scene move, the environment (sky, lights, wind...) changes all the time. Depending on your graphic card you get your entire scene recalculated all so often per second (the frame rate, or fps).

Really fast graphic cards can get up to about 100 fps, maybe even more i don't know. You start feeling lag when your framerate drops below 15. You get nervous when it drops below 10, and further down ... well not good at all.

Now the more data you have in the scene -> the more calculations have to be done -> the lower the frame rate will be -> the sooner lag will be -> the more frustration will be.

The fatal thing is that you personally might not suffer from too many vertices on screen because you might have a Graphic card monster in your computer which performs very well. But your neighbor down the street might only use a notebook with less capabilities and suffer a lot more from vertex monsters.

So at the end when nobody cares about efficiency, then more people will have a less adorable second life. And many might not even know that it could look all so much better when done right (and when wearing more efficient clothes so to speak)

[edit] one more essential difference between vertices and pixel pon screen:

 

  • Make a simple flat face with 4 vertices
  • Upload that to SL
  • put a 1024*1024 image texture on it.
  • rezz it and move your camera to it until it covers the entire screen

Now you have 4 vertices and 1 million pixels

 

Link to comment
Share on other sites


Gaia Clary wrote:

 

[edit] one more essential difference between vertices and pixel pon screen:

 
  • Make a simple flat face with 4 vertices
  • Upload that to SL
  • put a 1024*1024 image texture on it.
  • rezz it and move your camera to it until it covers the entire screen

Now you have 4 vertices and 1 million pixels

 

You now have about 1 million texels :)

The amount of pixels is the same as your screen resolution or it can be more when you throw in some anti aliasing. (Or it can be less when you run SL in a window)

Link to comment
Share on other sites

WOW thanks for all the great responses Gaia, Freya & Kwakkelde! :D

Judging by how hot this topic is sounds like LL should definitely set Torley the task of making a series of videos about best practices for mesh making.  One that goes into detail why and what causes lag.

Personally I love to see a graph comparing how much lag or stress is put on a computer processor between textures, vertices and scripts.

I was thinking that perhaps a new slider should be added to the graphics preferences.  One that you could set the maximum number of vertices a single mesh could have and if a mesh had more vertices than that number it would automatically be rendered the next lowest LOD level unless that LOD also had exceeded the number of vertices and so the next one under that LOD would be used and so on.

I know there is already Object Detail slider that does but it changes everything on your screen to a lower LOD basically.

I have a friend who designs mesh clothing and was told that they tried selling efficient low vertex count mesh but no one would buy it and was basically forced to use higher poly counts in order to get decent sales.  We talked about normal maps that should have been able to make lower poly meshes look a lot better but Normal Maps in SL seem muted.

I generated normal maps from both Maya and 3D Coat which both looked great in those programs but once I uploaded them to SL they had become blurry and the depth just wasn't there.  Best I could do was to open the normal maps in Photoshop and play around with the brightness and contrast of the Red and Green channels separately.  I also sharpened the whole image just a bit.  All this compensated a bit but to really get any detail I had to use 512 x 512 or 1024 x 1024.  Isn't that kinda defeating the whole purpose of normal maps?

Perhaps LL should have a separate upload window for Normal Maps where maps can be uploaded Lossless.  Just have textures uploaded as Normal Maps can not be applied as Diffuse textures.  I am sure the problem I am seeing with normal maps is because when the normal map texture is being uploaded it is being compressed blurring the image making it lose detail and depth.  128 x 128 is just not enough to get any real sort of detail for most situations.

PS is anyone else "Spell Check" not working?

Link to comment
Share on other sites

You are about to reply to a thread that has been inactive for 3684 days.

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
 Share

×
×
  • Create New...