Lag - prims, meshes or sculpts?

[Chin Rey]

Still perfectly reachable (and cachable last I tested) using HTTP, I'm afraid.

Assuming the point of https is to prevent copybotting, would it really have made any difference? The viewer needs the unecrypted data to be able to render an object at all and as long as the viewer code is unencrypted, there's always a way to hack it to gain direct access.

 

[Chalice Yao]

Assuming the point of https is to prevent copybotting, would it really have made any difference? The viewer needs the unecrypted data to be able to render an object at all and as long as the viewer code is unencrypted, there's always a way to hack it to gain direct access.

Oh, I don't care that it uses HTTP - it's a bit faster, no ISP will gain anything from it, there are no passwords or the like, and it's cachable - really, it's good for assets.

 

[Argent Stonecutter]

That's semantics. LL calls them "mesh assets".

Frankly, I don't care what they call them. They can call their home page a web asset, but it's not managed by the asset system.

 

[nebula]

So I have all kinds of computers to benchmark with as my company invests in new hardware frequently. And ones as old as 2014 that cost about $400 at purchase can still get 30 fps on (laptops)

Now what is interesting is the fps drop on certain machines trying to render mesh vs sculpts or prims based on what type of integrated graphics they have.

I have noticed that the apu chips have been way out permorming even some of the newer intel chip sets.

Very interesting. Also really important to keep in mind this is just for sl as they have very different performances for different games.

 

[Mechanical]

Then you'll need to carefully create an environment that does not vary from test to test, have rendered elements appear at the same times, and not down how they effect things. ARC is an unhelpful metric, and hidding how laggy an avatar really is.

 

[Argent Stonecutter]

I have noticed that the apu chips have been way out permorming even some of the newer intel chip sets.

Intel graphics performance is an oxymoron.

 

[Bartholomew Gallacher]

Intel graphics performance is an oxymoron.

For the moment - the more leaks we do get about the Intel GPUs to be released in 2020, the more those do look like decent and competitive pieces of hardware.

 

[Chin Rey]

ARC is an unhelpful metric, and hidding how laggy an avatar really is.

The display weight/ARC formula was developed by Nyx Linden and as far as I can see, he did a thorough job. It has a number of minor flaws of course and two medium sized ones, it doesn't actually count the triangles of a mesh - it just uses an estimate - and it shares the same LoD model weight factor as downlaod weight which overestimates the significance of lower LoD models and underestimates the higher ones.

The really serious flaws in the forumla are because it hasn't been updated to account for changes in SL. It completely ignores normal and specular maps, it does not take into account that texturing tend to be far more heavy than it used to be and of course, the really big one, due to the fitmesh LoD bugs, the calculation wil always be way off the moment there is any fitted mesh.

 

[Argent Stonecutter]

For the moment - the more leaks we do get about the Intel GPUs to be released in 2020, the more those do look like decent and competitive pieces of hardware.

Forgive me for being skeptical but I've heard that before.

 

[Bartholomew Gallacher]

I am very much convinced that Intel is dead serious on producing competitive GPUs, after they hired the former chief GPU architect from AMD in fall 2017. It's a market which gets more and more important in supercomputing and AI, but Intel has nothing there to deliver.

 

[Chin Rey]

I am very much convinced that Intel is dead serious on producing competitive GPUs, after they hired the former chief GPU architect from AMD in fall 2017.

They certainly are serious about it but it's not their first try so maybe we should wait and see if they can deliver this time.

 

[Da5id Weatherwax]

The display weight/ARC formula was developed by Nyx Linden and as far as I can see, he did a thorough job. It has a number of minor flaws of course and two medium sized ones, it doesn't actually count the triangles of a mesh - it just uses an estimate - and it shares the same LoD model weight factor as downlaod weight which overestimates the significance of lower LoD models and underestimates the higher ones.

The really serious flaws in the forumla are because it hasn't been updated to account for changes in SL. It completely ignores normal and specular maps, it does not take into account that texturing tend to be far more heavy than it used to be and of course, the really big one, due to the fitmesh LoD bugs, the calculation wil always be way off the moment there is any fitted mesh.

To be perfectly honest, when it comes to rendering the folks around me I have a very simple rule of thumb that seems to work much better than trying to find an arbitrary ARC cutoff.

If my framerate sucks, I flip into wireframe mode then jellydoll everyone that still looks perfectly solid....

 

[Clara D.]

I am very much convinced that Intel is dead serious on producing competitive GPUs, after they hired the former chief GPU architect from AMD in fall 2017. It's a market which gets more and more important in supercomputing and AI, but Intel has nothing there to deliver.

They're actively working with AMD to go up against nVidia.

Intel and AMD team up against Nvidia to produce a new laptop chip

The first partnership since the 1980s

www.theverge.com

 

[Bartholomew Gallacher]

Clara, this is only an intermission; there were always bits and pieces, where those two worked together and Intel licensed technology from AMD.

This doesn't change the fact, that Intel wants to become a rival for nVidia, and is planning big things for competition. And if they're finished and their own products are good enough, working together with AMD might change again.

 

[Clara D.]

Clara, this is only an intermission; there were always bits and pieces, where those two worked together and Intel licensed technology from AMD.

This doesn't change the fact, that Intel wants to become a rival for nVidia, and is planning big things for competition. And if they're finished and their own products are good enough, working together with AMD might change again.

Things are always changing in the tech world, several years ago there were rumors going around that Intel was looking to buy out nVidia to take on AMD/ATI

 

[Argent Stonecutter]

I am very much convinced that Intel is dead serious on producing competitive GPUs,

Speaking as someone who has been repeatedly screwed over by Intel making grand plans and abandoning them, back to the '80s, I am very much convinced that Intel has no bloody idea what it's dead serious about other than the x86 and it's even pulled back from that on occasion.

If AMD hadn't developed a 64 bit extension to the x86 and forced Intel to drop the Itanic as their "dead serious" plan for the future, we'd probably still be arguing about 32-bit vs 64-bit today.

 

[Clara D.]

We'd have nVidia CPU's if Intel hadn't basically paid them not to. Part of the legal settlement (2011) was for nVidia not to have x86.

 

[Chalice Yao]

The really serious flaws in the forumla are because it hasn't been updated to account for changes in SL. It completely ignores normal and specular maps, it does not take into account that texturing tend to be far more heavy than it used to be

They should have made the ARC increase from textures directly scale with the pixel count. Right now the formula is an absolute joke.

256 + (16 * ( ( height / 128) + (width / 128) ) )

So, a 16x16 texture adds 260 rendering cost.
A 1024x1024 adds...512.

It's as bad as the LOD mistake, if not worse given the average content. What they essentially should have done, is make it a simple formula for the dimension weight like (width * height) / 1024.0. That would result in a 1024x1024 being 1024.0 render weight, while a 16x16 is merely 0.25.
On top of that, yes, add a fixed amount per texture to that - to account for sheer amount of textures used. They already did that with the '256' per texture - just the added scaling calculation is worthless.

 

[nebula]

Forgive me for being skeptical but I've heard that before.

Yeah, I would be actually shocked.