Second Life Forums Archive - the Plane

Eep Quirk

Absolutely Relative

Join date: 15 Dec 2004

Posts: 1,211

04-07-2006 08:25

From: Argent Stonecutter

the number of polys in a cube isn't "pulled out of my ass".

Psst, yes they were--but I'm not referring to cube polys (duh). Go back and see the order in which you replied. You mentioned the cube polys AFTER you mentioned backing your comments up with numbers. Oops.

From: Argent Stonecutter

A lot of people do. It's a standard building tool these days to reduce prim counts, and there's people using hollowed sliced cubes to get 5 "plane" textures from a single prim.

I seriously doubt a lot of people do since a lot of people can't model worth shit in (or out of) SL. One look at all the idiotic tiny (tortured) high-prim jewelry (LIT, no less) makes this BLATANTLY obvious (and all the equally idiots who wear it). SL simply wasn't designed to BE very render-efficient initially (aside from LOD, which needs work), which is why it lags so much.

From: Argent Stonecutter

And if development time was free all possible approaches to reducing polygon counts would be equivalent. Since it isn't, it's vital to consider the costs and benefits of competing optimizations. So, there's three optimizations to be compared:

Plane object.
Flattened cubes.
Culling 100% alpha surfaces.

I didn't say "this optimization is implemented", I said "this optimization is possible"... and by making that optimization Linden Labs can get a MUCH bigger win than a "plane" object for a lot less cost... and you get the plane as well.

Plane object: reduce poly counts by a factor of 6 for signs, and by 3 for walls, but doubling the prim count for walls. Maximum potential savings, 10 polys per visible surface.
Flattened cubes: reduce poly counts by a factor of 2 for some signs and walls, and reducing the prim counts for some walls. Maximum potential savings, 4 polys per visible surface.
Culling 100% alpha surfaces: reduce poly count by a factor of 2-6 for any object where one or more surfaces is known to be completely obscured. Maximum potential savings, unbounded... may be hundreds of polys per surface for some objects.

Super, Argent. I don't care WHAT gets implemented, so long as SL DOES get a way to remove unnecessary polys. I'm still waiting for you to admit SL doesn't currently cull 100% alpha sides, incidentally...

Argent Stonecutter

Emergency Mustelid

Join date: 20 Sep 2005

Posts: 20,263

04-07-2006 11:23

From: Eep Quirk

I'm still waiting for you to admit SL doesn't currently cull 100% alpha sides, incidentally...

Huh?

When did I make the claim that SL culled 100% Alpha sides?

I said that SL provides a mechanism by which you can indicate that sides can be culled, and that I didn't know if they were currently performing that optimization, but that they did seem to be ignoring alpha sides in some circumstances.

I said that if they were, then a "plane" won't give you anything you don't already have.

I said that if they weren't, it would be easier for them to do that than to implement a new prim type, and it would be a MUCH bigger win.

But I don't see anywhere in the thread where I said they were culling them. To be precise, I wrote (please pay attention to the text in boldface, it's important):

From: Argent Stonecutter

They have it. It's called "setting the alpha value to zero".

I don't know if they are making good use of this information, but the client really doesn't seem to really know about the presence of 100% alpha objects that aren't owned by you until you turn on rendering for them with "show transparent", so they may be. In any case:

1. If they aren't then doing so would save orders of magnitude more polygons than a plane prim.

2. If they are, then there's no difference between a plane prim and a cube with 5 sides set to zero alpha. None.

You say they aren't making this optimization. I have no reason to contest that, so I have no problem with saying "to the best of my knowledge, SL doesn't currently cull 100% alpha sides". I honestly have no idea why you think it's important that I say that, though. Nor why you're arguing about it if you don't care what gets implemented. It's really confusing.

Kensuke Leviathan

Wandering fox

Join date: 11 Dec 2002

Posts: 127

04-07-2006 12:49

The LOD system isn't really that bad depending on the size of object and the distance, .2x.5x.2 spheres up close are 525 tris(roughly depending on camera postion) at 15 thats down to 24 and past that it's not even recieved as information on the client, heck the SL av gets a cut down to 1/7th of it's original tri count at 15m. Still though, without dymanic culling, this doesn't really have a drastic impact, your still viewing around 170K tris in a 128m area.

_____________________

_________________

":> wark wark"

Argent Stonecutter

Emergency Mustelid

Join date: 20 Sep 2005

Posts: 20,263

04-07-2006 16:01

From: Kensuke Leviathan

The LOD system isn't really that bad depending on the size of object and the distance, .2x.5x.2 spheres up close are 525 tris(roughly depending on camera postion) at 15 thats down to 24 and past that it's not even recieved as information on the client, heck the SL av gets a cut down to 1/7th of it's original tri count at 15m. Still though, without dymanic culling, this doesn't really have a drastic impact, your still viewing around 170K tris in a 128m area.

One problem I have recently noticed with the LOD system is that it's based on distance from the avatar, not distance from the camera. If you sit-teleport whiel focussed on the teleporter your camera doesn't move... but the teleporter is culled.

Eep Quirk

Absolutely Relative

Join date: 15 Dec 2004

Posts: 1,211

04-07-2006 19:20

From: Kensuke Leviathan

The LOD system isn't really that bad depending on the size of object and the distance, .2x.5x.2 spheres up close are 525 tris(roughly depending on camera postion) at 15 thats down to 24 and past that it's not even recieved as information on the client, heck the SL av gets a cut down to 1/7th of it's original tri count at 15m. Still though, without dymanic culling, this doesn't really have a drastic impact, your still viewing around 170K tris in a 128m area.

97817 is what's wrong with SL's LOD.

Eep Quirk

Absolutely Relative

Join date: 15 Dec 2004

Posts: 1,211

04-07-2006 19:22

From: Argent Stonecutter

Huh?

When did I make the claim that SL culled 100% Alpha sides?

I said that SL provides a mechanism by which you can indicate that sides can be culled, and that I didn't know if they were currently performing that optimization, but that they did seem to be ignoring alpha sides in some circumstances.

I said that if they were, then a "plane" won't give you anything you don't already have.

I said that if they weren't, it would be easier for them to do that than to implement a new prim type, and it would be a MUCH bigger win.

But I don't see anywhere in the thread where I said they were culling them. To be precise, I wrote (please pay attention to the text in boldface, it's important):
You say they aren't making this optimization. I have no reason to contest that, so I have no problem with saying "to the best of my knowledge, SL doesn't currently cull 100% alpha sides". I honestly have no idea why you think it's important that I say that, though. Nor why you're arguing about it if you don't care what gets implemented. It's really confusing.

Sorry; I didn't read what you wrote carefully enough. You worded it weird, though. Anyway, no, SL still renders fully transparent sides/prims.

the Plane
Eep Quirk Absolutely Relative Join date: 15 Dec 2004 Posts: 1,211	04-07-2006 08:25 From: Argent Stonecutter the number of polys in a cube isn't "pulled out of my ass". Psst, yes they were--but I'm not referring to cube polys (duh). Go back and see the order in which you replied. You mentioned the cube polys AFTER you mentioned backing your comments up with numbers. Oops. From: Argent Stonecutter A lot of people do. It's a standard building tool these days to reduce prim counts, and there's people using hollowed sliced cubes to get 5 "plane" textures from a single prim. I seriously doubt a lot of people do since a lot of people can't model worth shit in (or out of) SL. One look at all the idiotic tiny (tortured) high-prim jewelry (LIT, no less) makes this BLATANTLY obvious (and all the equally idiots who wear it). SL simply wasn't designed to BE very render-efficient initially (aside from LOD, which needs work), which is why it lags so much. From: Argent Stonecutter And if development time was free all possible approaches to reducing polygon counts would be equivalent. Since it isn't, it's vital to consider the costs and benefits of competing optimizations. So, there's three optimizations to be compared: Plane object. Flattened cubes. Culling 100% alpha surfaces. I didn't say "this optimization is implemented", I said "this optimization is possible"... and by making that optimization Linden Labs can get a MUCH bigger win than a "plane" object for a lot less cost... and you get the plane as well. Plane object: reduce poly counts by a factor of 6 for signs, and by 3 for walls, but doubling the prim count for walls. Maximum potential savings, 10 polys per visible surface. Flattened cubes: reduce poly counts by a factor of 2 for some signs and walls, and reducing the prim counts for some walls. Maximum potential savings, 4 polys per visible surface. Culling 100% alpha surfaces: reduce poly count by a factor of 2-6 for any object where one or more surfaces is known to be completely obscured. Maximum potential savings, unbounded... may be hundreds of polys per surface for some objects. Super, Argent. I don't care WHAT gets implemented, so long as SL DOES get a way to remove unnecessary polys. I'm still waiting for you to admit SL doesn't currently cull 100% alpha sides, incidentally...
Argent Stonecutter Emergency Mustelid Join date: 20 Sep 2005 Posts: 20,263	04-07-2006 11:23 From: Eep Quirk I'm still waiting for you to admit SL doesn't currently cull 100% alpha sides, incidentally... Huh? When did I make the claim that SL culled 100% Alpha sides? I said that SL provides a mechanism by which you can indicate that sides can be culled, and that I didn't know if they were currently performing that optimization, but that they did seem to be ignoring alpha sides in some circumstances. I said that if they were, then a "plane" won't give you anything you don't already have. I said that if they weren't, it would be easier for them to do that than to implement a new prim type, and it would be a MUCH bigger win. But I don't see anywhere in the thread where I said they were culling them. To be precise, I wrote (please pay attention to the text in boldface, it's important): From: Argent Stonecutter They have it. It's called "setting the alpha value to zero". I don't know if they are making good use of this information, but the client really doesn't seem to really know about the presence of 100% alpha objects that aren't owned by you until you turn on rendering for them with "show transparent", so they may be. In any case: 1. If they aren't then doing so would save orders of magnitude more polygons than a plane prim. 2. If they are, then there's no difference between a plane prim and a cube with 5 sides set to zero alpha. None. You say they aren't making this optimization. I have no reason to contest that, so I have no problem with saying "to the best of my knowledge, SL doesn't currently cull 100% alpha sides". I honestly have no idea why you think it's important that I say that, though. Nor why you're arguing about it if you don't care what gets implemented. It's really confusing.
Kensuke Leviathan Wandering fox Join date: 11 Dec 2002 Posts: 127	04-07-2006 12:49 The LOD system isn't really that bad depending on the size of object and the distance, .2x.5x.2 spheres up close are 525 tris(roughly depending on camera postion) at 15 thats down to 24 and past that it's not even recieved as information on the client, heck the SL av gets a cut down to 1/7th of it's original tri count at 15m. Still though, without dymanic culling, this doesn't really have a drastic impact, your still viewing around 170K tris in a 128m area. _____________________ _________________ ":> wark wark"
Argent Stonecutter Emergency Mustelid Join date: 20 Sep 2005 Posts: 20,263	04-07-2006 16:01 From: Kensuke Leviathan The LOD system isn't really that bad depending on the size of object and the distance, .2x.5x.2 spheres up close are 525 tris(roughly depending on camera postion) at 15 thats down to 24 and past that it's not even recieved as information on the client, heck the SL av gets a cut down to 1/7th of it's original tri count at 15m. Still though, without dymanic culling, this doesn't really have a drastic impact, your still viewing around 170K tris in a 128m area. One problem I have recently noticed with the LOD system is that it's based on distance from the avatar, not distance from the camera. If you sit-teleport whiel focussed on the teleporter your camera doesn't move... but the teleporter is culled.
Eep Quirk Absolutely Relative Join date: 15 Dec 2004 Posts: 1,211	04-07-2006 19:20 From: Kensuke Leviathan The LOD system isn't really that bad depending on the size of object and the distance, .2x.5x.2 spheres up close are 525 tris(roughly depending on camera postion) at 15 thats down to 24 and past that it's not even recieved as information on the client, heck the SL av gets a cut down to 1/7th of it's original tri count at 15m. Still though, without dymanic culling, this doesn't really have a drastic impact, your still viewing around 170K tris in a 128m area. 97817 is what's wrong with SL's LOD.
Eep Quirk Absolutely Relative Join date: 15 Dec 2004 Posts: 1,211	04-07-2006 19:22 From: Argent Stonecutter Huh? When did I make the claim that SL culled 100% Alpha sides? I said that SL provides a mechanism by which you can indicate that sides can be culled, and that I didn't know if they were currently performing that optimization, but that they did seem to be ignoring alpha sides in some circumstances. I said that if they were, then a "plane" won't give you anything you don't already have. I said that if they weren't, it would be easier for them to do that than to implement a new prim type, and it would be a MUCH bigger win. But I don't see anywhere in the thread where I said they were culling them. To be precise, I wrote (please pay attention to the text in boldface, it's important): You say they aren't making this optimization. I have no reason to contest that, so I have no problem with saying "to the best of my knowledge, SL doesn't currently cull 100% alpha sides". I honestly have no idea why you think it's important that I say that, though. Nor why you're arguing about it if you don't care what gets implemented. It's really confusing. Sorry; I didn't read what you wrote carefully enough. You worded it weird, though. Anyway, no, SL still renders fully transparent sides/prims.

Welcome to the Second Life Forums Archive

the Plane