Polygon amount, Land Impact and Lag

[gustav2005]

It may sound bit ambiguous, but I have a question.

 

There's a 1LI object.

Both of them are using 512x512 single texture.

One is 400 polygons. Higher LoDs settings. Low-polygon looking model but durable from a long distance view.

Another one is 1600 polygons. Extremely lower LoDs settings. It looks fantastic but dissapear in a very short distance.

Both of them are simply 1LI, but I feel my viewer gets slower and heavier when I put 1600 one around.

 

Is this truly happening? Or am I just hullucinating?
If 1600 object is heavier than 400 in some ways, where can I see the numeric defference in the mesh uploader when I upload them to SL?

[Ivanova Shostakovich]

   While your viewer might slow a little with that 1600 poly object, I think it's 1LI can be explained by its aggressively efficient LOD settings which prevent potential nearby viewers from rendering it.

   When uploading a mesh, a very well optimized object, with aggressive LODs and a simple physics shape, can easily have an LI of less than one. Two different meshes, when uploading, can each have LIs of less than one. For instance, one could be 0.75 and the other could be 0.23. They would both end up being 1LI objects rezzed in-world (depending on size) because 1LI is the lowest it can be for a single object.

   The LI of any object reflects the rendering cost of not just your viewer, but of any other potential viewers which could also need to download and render that object. I know there are other factors, but I don't have the technical acumen to give you 42 on this topic.

[arton Rotaru]

It's unlikely that 1200 triangles more or less have any significant impact in a scene. However, you can check your Frames per second in the Statistics display (ctrl+shift+1), at the very top.

In the viewers Build floater, under the More Info link, you can check the Display weight of your objects.

The Uploader will just tell you the triangle count for each LOD, and the in-game vertices count. The Download weight, the Convex Hull Physics weight, Server weight, and the Land impact.

Depending on the size of the object, it's not really necessary to have poor lower LOD models to keep the object at 1 LI. I have meshes which have more than 1600 Tris, in the High LOD, at 1 LI, but still recognisable shapes down to the lowest LOD.

And if it counts 2, or 3 LI in the end, with proper LODs, it's still good IMO.

[gustav2005]

Thank you very much for your replies, Ivanova and Arton.

 

Yes, mesh objects can go down below 1LI and all of 1LI objects are not the same actually.
However.

Let's think, 400 and 1600, both of them are exactly 1.0LI. LoDs is the only difference.
They are physically just the same size. 50 cm diameter spheres.
Looking at them from the same distance (this is important). 1 m away from them.

 

In my idea, 400 has less points to render. On the other hand, 1600 has more points to render. Even both of them are 1LI. If so, does it mean 1600 gives more work to do to our PC or server or whatever? In that case, can't we say 1600 is much “heavier” than 400? Can we treat them equally?

 

Or SL is not working in that way and dosen't it matter because they're simply just 1LI objects?

 

[arton Rotaru]

Well, the Land impact calculation takes all 4 levels of detail into account, as well as the size of the object. The bigger the object, the less impact have the lower LODs in the calculation.

So it's quite simple. You would have to look at the triangle, vertices numbers for the lower LODs of the objects. To have these meshes have an equal download  weight, the 1600 mesh will have less polys to display on the lower LODs than the 400 mesh will.

Which makes the 400 mesh heavier than the 1600 mesh when viewed from a distance where the LODs kick in.

Lets say the 400 mesh has 200 tris at the medium LOD, but the 1600 drops to 20 tris already. 200 is much heavier than 20 obviously.

I'm still  not sure what you're exactly after though?

[gustav2005]

Hmmmmmm, I try to make my question clear. It's mostly because I'm not a tech expart (and that's why I'm asking this though). It might take a while, please forgive.

 

400 and 1600 polygon spheres.
Both 1LI.
I set "Medium" LoDs of both spheres as "Use LoD above". 
Only the difference between them are  "Low" and "Lowest" LoDs settings to achieve 1LI.
Physically they're the same size.

Now, looking at 400 and 1600 spheres from 1 m away from them.
Both of them are appearing perfectly in my display. 
No distortion is appearing because they're close to the camera and "Low" and "Lowest" LoDs settings are not affecting them.

At this moment, does 1600 requires more rendering work than 400 because it has to show more details?

 

[arton Rotaru]

---

gustav2005 wrote:

At this moment, does 1600 requires more rendering work than 400 because it has to show more details?

 

---

Certainly. This will be more clear if you would rezz, lets say, 1000 of the 1600 spheres. Which would be 1.6 million triangles, compared to 400k tris for the 400 spheres. Which is quite a difference. and certainly will have an impact on the rendering performance.

But that's all theoretical. In reality, you hardly would place 1000 spheres all at the same spot, where all of them would show the high LOD. They'ed rather be in different locations where some show the high LOD, and others the lower LODs.

That's what the Land impact calculation is based of. The practical example. That's also why it's a not so good idea to set  the RenderVolumeLODFactor to high, because you would have to render many more 1600 spheres at it's high LOD, even in the practical example, than with a default RVLODF setting.

Personally, I spend quite a few polys on small objects in the high LOD, because due to the small size, they LOD down very quickly, so their impact on a whole scene isn't too bad anyway. Given the LOD factor is set to a reasonable number.

The key is in making pretty lower LOD models, which will still represent the object somehow, but at very low polycounts.

[ChinRey]

---

Ivanova Shostakovich wrote:

   While your viewer might slow a little with that 1600 poly object, I think it's 1LI can be explained by its aggressively efficient LOD settings which prevent potential nearby viewers from rendering it.

---

Umm, I really, really wouldn't use the word "efficient" in that context. Efficient LOD settings is when you manage to reduce the LI without sacrificing too much quality.

 

---

arton Rotaru wrote:

It's unlikely that 1200 triangles more or less have any significant impact in a scene.

---

That's right in theory but there are several indiciations that mesh (and sculpts) can add quite a bit more lag than it should - compared to prims of equal complexity.

It's also important to realize that land impact is not a very good indication of lag. For a start it can only be a fairly rough estimate of the load, there are way to many unkown and ever changing factors to make the calculation exact. BUt there's also the matter of what the LI calculation is based on.

To put it simply, for an object to be rendered, it needs to go through five stages:

• First it has to be retrieved from the asset server. That is the server weight.
• Second the sim server has to do some calculations, mainly related to the physics. That's the physics weight.
• Third stage, it's sent from the server to the client (that's a fancy name for your and mine computers). That's download weight.
• Fourth the client cpu has to do some calculation. That's the .... ummm no wait, that's not included in the weights system at all.
• Finally the client's gpu has to actually draw the object on the screen. That's the render weight, aka draw weight.'

Now, obviously it's the weakest link in this chain that really matters. It doesn't matter if four of the stages run smoothly and efficiently if the fifth keeps holding everything up. Land impact is based on the weakest of the first three links. It does not take into account the client side cpu and gpu loads which are nearly always the real limiting factors in today's Second Life. So usually there's little or no connection between LI and lag. (That is: items with higher download weight tend to be heavier on the client too but that's a rule with many big exceptions.)

The reason why even the best made mesh is a bit heavier than one should expect, has to lie with the client cpu; by the time the gpu gets to work, it's all vertices and tris and pixels anyway, regardless of whether the object started off as prims or sculpts or mesh. There seems to be something in the viewer software that makes it harder for the client to deal with mesh, or perhaps we should say makes it easier for the client to deal with prim.