Jump to content

SL Performance


You are about to reply to a thread that has been inactive for 992 days.

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

Recommended Posts

LL cannot continue to look the other way, they cannot [continue to] sweep this under the rug, nor can they deny the obvious. When is LL going to address, let alone rectify, the intensely & woefully degraded performance of SL since its migration to the cloud?

And no, this is not a function of my 'puter, my connection or anything else having to do with my system.

  • Like 2
  • Haha 1
  • Confused 1
Link to comment
Share on other sites

 

44 minutes ago, Rolig Loon said:

Same here.  I haven't seen any change in performance, up or down, since the migration.  I'm doing fine, and I am in world a lot.

There was the expectation (oft alluded to by Oz) that SL would perform better in the cloud. I'm not saying it doesn't, but it's entirely subjective and what statistics we get in the viewer aren't a huge help beyond being able to declare a region "fine" or "not fine". What would "fine +10%" even look like? Once client side factors are mitigated even super-busy regions seem to be holding up their end. Region crossings are what they are .. as always.

On the whole .. I would say SL feels more "fine" that it did before the uplift. 

 

From my experiences with performance testing, client side it's a trashfire and only getting worse. We're (Catznip) spending more and more time each release cycle optimizing and testing just to stay still. EEP came with a massive hit that meant we missed our release window by almost a year going down rabbit holes trying to make up the difference, in the mean time content in SL hasn't stood still.

 

  • Like 5
Link to comment
Share on other sites

Same for me, its been as good if not better since the uplift. Except in certain places with too many objects and oversized textures around (*cough*Bellisseria*cough). But that's the fault of users and creators, not Linden Lab.

  • Like 3
  • Thanks 1
Link to comment
Share on other sites

I have a decent computer and fast Internet.
SL loads fast and quickly most of the time with high frame rates. It is all a tat slower since the uplift but not worth talking about.
Two big exceptions:
1. Arriving at Bellisseria coming from other non Belli parcels is a lag crime. Gray, gray and more gray, sometimes for minutes.
We pay the price for the Moles extravaganza with beautiful scenery but sadly also with extreme LI heavy houses, scenery and 1024x1024 textures.
The log home sims and the camping sims are the worst IMHO.

2. When people gather at events, clubs, beaches etc. Things slow down as well.
People want to look too good and sharp.  Result: all the efforts with nice jewelry, extra layers etc. are mostly for nothing because others see you as blurry clouds because they want halfway decent frame rates and a less grayness.

 

Edited by Sid Nagy
A few adjustments
  • Like 4
Link to comment
Share on other sites

9 minutes ago, Coffee Pancake said:

it's entirely subjective and what statistics we get in the viewer aren't a huge help beyond being able to declare a region "fine" or "not fine". What would "fine +10%" even look like? Once client side factors are mitigated even super-busy regions seem to be holding up their end. Region crossings are what they are .. as always.

Exactly right.  Everyone is going to have a different definition of "fine", and we're all going to use different benchmarks. Our computers age, our Internet connections get up/downgrades, and we deal with CoVid madness. The only definition that matters is your own.  On my personal scale from "Gone to Hell In a Handbasket" to "Cue the Choir, it's Hallelujah Time", I'd say things are pretty much as they have been for at least the past year: "Meh, Could Be Better But I'm Not Complaining".  So yeah, "Fine".

  • Like 3
Link to comment
Share on other sites

26 minutes ago, Maitimo said:

Same for me, its been as good if not better since the uplift. Except in certain places with too many objects and oversized textures around (*cough*Bellisseria*cough). But that's the fault of users and creators, not Linden Lab.

I blame LL mostly  for the Bellisseria lag festival.
They manage what the Moles build. And the Moles build very LI heavy and with a lot of 1024x1024 textrures where there is no need for it.

LL decides the LI allowance people get in Belli. And of course they use them.
Bad non LL designs from creators are to be seen on almost every sim, but they don't lag the place up as much as in Belli.
LL provides too much lag heavy scenery in their attempts to make it look beautiful
 

Edited by Sid Nagy
  • Like 3
Link to comment
Share on other sites

4 minutes ago, Maitimo said:

Except in certain places with too many objects and oversized textures around 

 

3 minutes ago, Sid Nagy said:

LL decides the LI allowance people get in Belli. And of course they use them.

 

It's always the decorative clutter long before it's the textures. Meshes consume the bulk of VRAM to the point that they can squeeze textures into a thrash-reload loop.

Li isn't a measure of complexity, it's more accurately described as measure of scale. Almost anything, regardless of geometric complexity can be 1 Li if made from a single object and smaller than the size of a default prim.

Gazebo I made .. normal avatar scale it's 14 Li, if it wasn't for the complex physics (that could be omitted if this was just decorative desk trash) It would be 1 Li with a decent portion of the geometry sub pixel in size. 

(I did go to town hand making LOD meshes for this, but that only affects the full size version's Li .. at this size it's always going to be 1 - 2 Li regardless)

It's trivial to fill a house with little things like this .. that the viewer will likely render even if you can't see them.

YRbmpoQ.png

  • Like 5
  • Thanks 2
  • Sad 1
Link to comment
Share on other sites

27 minutes ago, Coffee Pancake said:

It's always the decorative clutter long before it's the textures. Meshes consume the bulk of VRAM to the point that they can squeeze textures into a thrash-reload loop.

True, however that should have been taken into consideration in the initial designing of the buildings by the Moles and the landscape. If they had thought forward as to how people would use the homes (an obvious observation to any user older than 1 year - they fill it to the brim and use all LI allowance) they would have realised that it was going to be a cluster****. As far as I know they did test the sim for lag, but only their objects without filling the houses to the brim like the users have. Very bad planning and QA. Then there is the whole physical impact on the sim where they could have at least made bushes and tree foliage to have no physics. Instead they leave them on so if multiple people hit bushes it impacts on the sim performance. A minor thing but it all adds up.

Even if the terribly basic culling method SL uses of distance to avatar is taken into account just some form of terraforming, object cliffs etc to hide around corners small groups of plots it would have helped. Instead Belli is design almost as a flat plane with house after house after house. Even in modern games they dont do such things due to lag concerns.

Whilst I agree meshes play the main role in the bulk of VRAM over textures, this doesn't absolve the Moles of their negligence in their designs. Go have a click on the trailers in Belli and you will see that the mesh complexity is insane compared to well made homes others sell with the download cost and land impact taking up near the same as large user made homes that are twice to three times the size. They have meshed screws, grooves, etc. that all could have been done with materials. They then upload 1024 textures for an object that does not need anything greater than a well made uv mapped 512 texture.

27 minutes ago, Coffee Pancake said:

Li isn't a measure of complexity, it's more accurately described as measure of scale. Almost anything, regardless of geometric complexity can be 1 Li if made from a single object and smaller than the size of a default prim.

It might not be a measure of complexity in real terms but it is a large indicator of a badly made mesh. If a trailer home 1/3 of the size and literally a rectangular one room building equals the LI of a large multi room house, then something is amiss.

  • Like 3
Link to comment
Share on other sites

1 hour ago, Sid Nagy said:

Result: all the efforts with nice jewelry, extra layers etc. are mostly for nothing

Not according to my experience. In my SL cloud people are definitely not at the point where they can't wear "nice jewelry". Of course this is subjective because nobody has the data, so I'm saying this based on what I see and on what feedback I get. (And yes I know that there are even people who rarely leave their sim because their computer is this weak that their viewer would just crash)

Edited by Noelle Delaunay
Link to comment
Share on other sites

12 hours ago, Drayke Newall said:

True, however that should have been taken into consideration in the initial designing of the buildings by the Moles and the landscape. If they had thought forward as to how people would use the homes (an obvious observation to any user older than 1 year - they fill it to the brim and use all LI allowance) they would have realised that it was going to be a cluster****. As far as I know they did test the sim for lag, but only their objects without filling the houses to the brim like the users have. Very bad planning and QA.

Maybe but why didn't anybody warn them?

 No wait! I think somebody did:

On 3/20/2019 at 7:00 PM, ChinRey said:

One test you should do, is fill the houses of an SSP sim with the kind of furniture people are likely to add and the garden with the kind of plants and other garden items they are likely to use. Then log on four or five mesh avatars on Powerbooks or similar and see what kind of performance you get.

https://community.secondlife.com/forums/topic/426577-ssp/page/31/?tab=comments#comment-1871988 (Typos corrected)

Edited by ChinRey
  • Like 2
  • Thanks 4
Link to comment
Share on other sites

Is there any sort of guidance published to guide users (like myself) who do not have a very good understanding of what we should buy?  Any way of endorsing particular retailers who have low script whatadoodles, & responsible with textures & polygons.  
I asked here in the forums basically this same question a few years back & no one responded.  Not even to say I was asking in the wrong place to direct me to help in a different place.  🤷‍♀️ 

Ive read that my mesh body is one of the ones that contribute to negative  performance because of the poly count.  Someone im’d me once at a club to bawl me out that the attached pet I had was killing her machine.  I didn’t understand why tho.  Was it scripts, textures?  Someone else advised me that my favorite wig shop was terrible with the textures & should be avoided.  I’ve been told the shops I visit for gardening suck the life out of a regions performance.

So basically, my shopping choices are horrific.  And it’s not that I bought cheap things.  I shop/shopped at the trendy stores. My logic is that those bigger name- higher traffic places have a handle on being responsible with what they churn out.  Seems I am wrong.

It would be great (Mr. Lumberg) if someone could explain it like I’m 5 & go a step further & point me to shops that aren’t computer & region killers.  If there is a treasure trove already compiled I’d take a look.  I looked thru the wiki & some blogs in the past & only walked away feeling more confused.  Like, I want to drive an environmentally friendly vehicle.  I don’t & won’t understand how a catalytic converter works.  But I can appreciate why it’s important & (let’s pretend) that not all new cars have them.  If someone could explain which do, & if some are more efficient, I’ll be all over that in an effort to save the planet.  
 

I want to do my part to make SL run smoother.  For now it consists of my turning off the sky & water ( & sometimes land (at shopping events) with only friends rendered & no shadows, 32 draw distance, shopping in a full alpha etc.  
 

Link to comment
Share on other sites

14 hours ago, Coffee Pancake said:

Li isn't a measure of complexity, it's more accurately described as measure of scale. Almost anything, regardless of geometric complexity can be 1 Li if made from a single object and smaller than the size of a default prim.

Gazebo I made .. normal avatar scale it's 14 Li, if it wasn't for the complex physics (that could be omitted if this was just decorative desk trash) It would be 1 Li with a decent portion of the geometry sub pixel in size. 

...

It's trivial to fill a house with little things like this .. that the viewer will likely render even if you can't see them.

That's not quite true.

The download weight part of the LI equation, which is the one that is relevant here, is mainly based on the number of tris in each LoD model and the object's size. The idea is that the smaller the object is, the less likely it is that the higher LoD models are downloaded and rendered. This is a genuine saving both when it comes to bandwidth and render cost, or rather, it would have been if the LoD/LI system had been used the way it was intended.

Let me give an example. Say we have a decorative object inside a house. The size (for the object that is, not the house) is 0.5x0.5x0.5 m. Tri counts for the LoD models: 25,000 - 1,000 - 200 - 15.

If the viewer's LoD factor is set to 1:

  • High LoD (25,000 tris) only downloaded and rendered for view distances less than 1.45 m
  • Mid LoD (1,000 tris): 1.45-5.78 m
  • Low LoD (200 tris): 5.78-11.55 m
  • Lowest LoD (15 tris): >11.55 m

In other words, that crazy 25,000 tri high LoD model shouldn't matter because it's hardly ever actually downloaded and rendered. Anybody outside the house should only have to deal with the measly 15 tris of the lowest LoD model - even a Commodore Amiga can do that without raising a sweat.

Let's increase the LoD factor to 2:

  • High LoD (25,000 tris): <2.9 m
  • Mid LoD (1,000 tris): 2.9-11.55 m
  • Low LoD (200 tris): 11.55-23.1 m
  • Lowest LoD (15 tris): >23.1 m

This is not ideal but hardly a big issue for an object as small as this (although it can be significant for slightly bigger objects).

Crank the LoD factor up to the ridiculous 4:

  • High LoD (25,000 tris): <5.8 m
  • Mid LoD (1,000 tris): 5.8-23.1 m
  • Low LoD (200 tris): 23.1-46.2 m
  • Lowest LoD (15 tris): >46.2 m

This means your viewer may have to deal with that high poly model just because you happens to be in the same room or close to the wall in the next room. The 200 tris wasted on somebody who's passing by on the street outside may not seem like much but imagine there are hundreds, even more than a thousand such objects around - not at all unlikely in a densely populated suburb - and we have a significant and completely unnecessary reduction in performance.

That's the theory anyway. Unfortunately it seems very few content creators and users and no current Lindens or Moles understand how it is supposed to work. The result is a lot of poorly made content and also some rather dubious code in the software.

Edited by ChinRey
Typos
  • Like 1
Link to comment
Share on other sites

2 hours ago, Pixie Kobichenko said:

Is there any sort of guidance published to guide users (like myself) who do not have a very good understanding of what we should buy?

Unfortunately no. Well, there is one good rule at least: If a content creator tells you you have to increase yout RenderColumeLODFactor to see their work properly, you almost certainly want to stay away from them.

Unfortunately very little of LL's official documentation is helpful and much of it is downright wrong or misleading. All the info is out there, thanks to a number of content creators, TPV developers and other inquisitive users spending countless hours reverse engineering the whole thing. But it's all scattered about. Complining it all into a cohesive and user friendly manual would take month, maybe even a year or more of full time work. You can't expect anybody to do that unless they're getting paid for it and nobody's willing to cover the bill.

 

2 hours ago, Pixie Kobichenko said:

My logic is that those bigger name- higher traffic places have a handle on being responsible with what they churn out.  Seems I am wrong.

Oh yes, that is definitely wrong.

 

2 hours ago, Pixie Kobichenko said:

Ive read that my mesh body is one of the ones that contribute to negative  performance because of the poly count.  Someone im’d me once at a club to bawl me out that the attached pet I had was killing her machine.  I didn’t understand why tho.  Was it scripts, textures?

Probably not scripts but textures and poly count, yes.

There are two special problems with avatars and wearables. One is that there is no equivalent to land impact for them so no realistic limit to how much you can load down the scene with your avatar. The other is that the LoD system is effectibely broken for fitted mesh because of a set of three stupid bugs. This means that you viewer has to handle the full load of every single tri on every single item worn by every single avatar around you, even those who are so far away you can barely see them.

 

2 hours ago, Pixie Kobichenko said:

I’ve been told the shops I visit for gardening suck the life out of a regions performance.

Oh what the F... Here we go.

The only store that offers truly optimized mesh plants is OPQ Gardens and Landscaping. That's my store so yes, you can accuse me of blatant self promotion and the moderators can delete this post if they like. But it is still true. I don't really have that much to sell in SL anyway. I have lots of very low poly background/filler vegetation but I had barely started listing my more elaborate plants before I gave up on SL.

Others that make and sell very good mesh plants that although they may not be perfectly well optimized shouldn't cause any performance issues worth speaking of unless you have masses of them:

  • JubJub Forder (JubJubs Stuff)
  • Teresa Matfield (T-Spot Home & Garden)
  • Reid Parkin (Mesh Plants)
  • Alex Bader (Skye)
  • Eldowyn Inshan (United InshCon)

Well worth checking out:

  • Nadine Reverie (3D Trees) - high lag plants but so gorgeous they may be worth it. Don't be afraid to use a few of her trees for those special feauture spots but you probably shouldn't make a forest out of them.
  • Lilith Heart (Heart Trees Flowers Plants) - same as above. Lilith Heart is steadily improving so as a general rule, the more recent her builds are, the better technical quality they have.
  • FANATIK - very uneven quality but their best is really good
  • Mitsuko Kytori (Hayabusa) - generally good quality but the animation tricks she uses come at a huge performance cost. It's up to you to decide whether it's worth it.

I'm not sure if I should mention sculpt vegetation here. The sculpt system as some serious flaws that mean they can never perform as well as high quality mesh. (It's possibly not the LL developers' fault. The sculpt system was a rush job so they had to cut a lot fo corners to gt it out on time.) Well made sculpt plants can still outperform poorly made mesh ones though, so we shouldn't completely rule them out. Some good sculpt plant makers who still offer some of them for sale:

  • JubJub Forder (JubJubs Stuff)
  • Nadine Reverie (3D Trees)
  • Lilith Heart (Heart Trees Flowers Plants)
  • Cordts Baxton

Disclaimer: These are not complete list of course. There are several good plant makers I can't rememebr the name of at the moment and probably lots I've never even heard of.

Edited by ChinRey
  • Thanks 2
Link to comment
Share on other sites

 

The raw tris count isn't what brings the house down. Even with LoD factor 4, the tris count isn't really a huge factor and only costs a few  percentage points off the frame rate on recent hardware. High poly counts are bad, but not because of the impact of tris counts on FPS .. they are bad because they sit in VRAM  causing the texture space to become constrained and for thrashing to start .. or worse, you simply run out entirely (as happens constantly on the Linden viewer). There is no difference between an avatar scale gazebo and desk trash one from a logistical perspective and that's where the pain is.

This is in part probably why LoD factor 4 has been allowed to slide as an acceptable solution. It does drop performance, no question about that, but it's irrelevant next to the elephant in the room trying to manage all the data.

It's also important to note, it's not the textures. 1024's on everything is technically bad, but .. it's only noticed because mesh has gobbled up all the VRAM and textures are left fighting each other for the scraps.

@Sid NagyIt's also not the moles work and houses or trees. They might look like they are having a texture problem, but really .. that is a symptom of a bigger problem. If the viewer runs out of VRAM because it's full of mesh data, it just starts junking textures, sods law dictates it will pick the most obvious ones first. We've done testing on empty Beli regions and the mole builds are generally very lean. It gets rubbish after people move in .. and stays rubbish even if you derender all the mole content .

This is especially notable in the Linden Viewer which is capped at 512 MB of VRAM. There are Belli homes that are basically off limits to the Linden Viewer. 

z5IW1YT.png

 

  • Thanks 2
Link to comment
Share on other sites

2 hours ago, ChinRey said:

Maybe but why didn't anybody warn them?

 No wait! I think somebody did:

https://community.secondlife.com/forums/topic/426577-ssp/page/31/?tab=comments#comment-1871988 (Typos corrected)

The sad thing is that they shouldn't have to be warned or told what to do. That was what I was getting at. They are employed by LL to create content and that content should be the epitome of created content and an example of how to do things. They failed miserably and what is even sadder, they continue on their merry way, doing the same builds and sim planning on newer Belli sims as if the lag doesn't exist.

  • Like 1
Link to comment
Share on other sites

Just now, Drayke Newall said:

The sad thing is that they shouldn't have to be warned or told what to do. That was what I was getting at. They are employed by LL to create content and that content should be the epitome of created content and an example of how to do things. They failed miserably and what is even sadder, they continue on their merry way, doing the same builds and sim planning on newer Belli sims as if the lag doesn't exist.

Sprawling contagious spaces without instanced content is bad. 

SL does this on purpose. We expect this. We wouldn't want to live in Belli if it wasn't like this.

Belli isn't a technology show piece, it's tasty cakes and pumpkin spice for people who explicitly want tasty cakes and pumpkin spice. (which is why I own 2 Belli homes :D lalala !!)

  • Like 1
Link to comment
Share on other sites

1 minute ago, Coffee Pancake said:

Sprawling contagious spaces without instanced content is bad. 

SL does this on purpose. We expect this. We wouldn't want to live in Belli if it wasn't like this.

You wouldn't want to live in Belli if there were mountains with lookout spots, smaller niche areas, yet just as neighbourhood like. Cliffs with waterfalls and parks around for the trailer house's etc?

I dont want to live in Belli because it is so bland and flat... maybe I'm the only one that feels that way.

1 minute ago, Coffee Pancake said:

Belli isn't a technology show piece, it's tasty cakes and pumpkin spice for people who explicitly want tasty cakes and pumpkin spice. (which is why I own 2 Belli homes :D lalala !!)

I know it isn't a show piece however because LL 'made' it they should at least do it to a good standard and set an example. Then if people ask like above what is a good standard for mesh builds we could then point to those examples. It is the same with the free premium gifts. They are terribly made of which being something 'from' LL they should be well made.

  • Like 1
Link to comment
Share on other sites

24 minutes ago, ChinRey said:

Unfortunately no. Well, there is one good rule at least: If a content creator tells you you have to icnrease yout RenderColumeLODFactor to see their work properly, you almost certainly want to stay away from them.

Unfortunately very little of LL's official documentation is helpful and much of it is downright wrong or misleading. All the info is out there, thanks to a number of content creators, TPV developers and other inquisitive users spending countless hours reverse engineering the whole thing. But it's all scattered about. Complining it all into a cohesive and user friendly manual would take month, maybe even a year or more of full time work. You can't expect anybody to do that unless they're getting paid for it and nobody's willing to cover the bill.

 

Oh yes, that is definitely wrong.

 

Probably not scripts but textures and poly count, yes.

There are two special problems with avatars and wearables. One is that there is no equivalent to land impact for them so no realistic limit to how much you can load down the scene with your avatar. The other is that the LoD system is effectibely broken for fitted mesh because of a set of three stupid bugs. This means that you viewer has to handle the full load of every single tri on every single item worn by every single avatar around you, even those who are so far away you can barely see them.

 

Oh what the F... Here we go.

The only store that offers truly optimized mesh plants is OPQ Gardens and Landscaping. That's my store so yes, you can accuse me of blatant self promotion and the moderators can delete this post if they like. But it is still true. I don't really have that much to sell in SL anyway. I have lots of very low poly background/filler vegetation but I had barely started listing my more elaborate plants before I gave up on SL.

Others that make and sell very good mesh plants that although they may not be perfectly well optimized shouldn't cause any performance issues worth speaking of unless you have masses of them:

  • JubJub Forder (JubJubs Stuff)
  • Teresa Matfield (T-Spot Home & Garden)
  • Reid Parkin (Mesh Plants)
  • Alex Bader (Skye)
  • Eldowyn Inshan (United InshCon)

Well worth checking out:

  • Nadine Reverie (3D Trees) - high lag plants but so gorgeous they may be worth it. Don't be afraid to use a few of her trees for those special feauture spots but you probably shouldn't make a forest out of them.
  • Lilith Heart (Heart Trees Flowers Plants) - same as above. Lilith Heart is steadily improving so as a general rule, the more recent her builds are, the better technical quality they have.
  • FANATIK - very uneven quality but their best is really good
  • Mitsuko Kytori (Hayabusa) - generally good quality but the animation tricks she uses come at a huge performance cost. It's up to you to decide whether it's worth it.

I'm not sure if I should mention sculpt vegetation here. The sculpt system as some serious flaws that mean they can never perform as well as high quality mesh. (It's possibly not the LL developers' fault. The sculpt system was a rush job so they had to cut a lot fo corners to gt it out on time.) Well made sculpt plants can still outperform poorly made mesh ones though, so we shouldn't completely rule them out. Some good sculpt plant makers who still offer some of them for sale:

  • JubJub Forder (JubJubs Stuff)
  • Nadine Reverie (3D Trees)
  • Lilith Heart (Heart Trees Flowers Plants)
  • Cordts Baxton

Disclaimer: These are not complete list of course. There are several good plant makers I can't rememebr the name of at the moment and probably lots I've never even heard of.

Thank you for taking time to respond, give me answers I can understand & some names of shops that could be helpful (I screenshot your reply in case it is deleted).

I do know some of the shops you listed & have somethings from a few.  So that’s good to know 😃

Also, thank you for not talking down at me =).  ( I will check out your shop, too)

  • Like 1
Link to comment
Share on other sites

29 minutes ago, Coffee Pancake said:

Sprawling contagious spaces without instanced content is bad.

Yes but for all its flaws, instanced content is the one and only reason why Belliseria works at all. If the Moles hadn't saved bandwidth and viewer memory by reusing the same textures and mesh assets over and over again it would have been a disaster.

  • Like 2
Link to comment
Share on other sites

27 minutes ago, Drayke Newall said:

I know it isn't a show piece however because LL 'made' it they should at least do it to a good standard and set an example.

Yes and that's my biggest pet peeve when it comes to how LL handles lag/performance issues. "Do as I say, not as I do" never works well, especially not when you hardly say anything at all.

  • Like 1
Link to comment
Share on other sites

You are about to reply to a thread that has been inactive for 992 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...