How much leway does streaming give SL in the GFX department

I have heard streaming content as the reason SL is rather behind the curve in the graphics department. I do know that nothing else steams content like SL does but how much leway does that give them. I bring this up because I was bored and did a little test to see how much slower SL was compared to modren games so I won't an unsupported .

For the test I used HL2: the lost coast with everything on its highest setings except HDR, AA (both of which don't work for my computer on that game) and only trilineral filtering (incase I needed some wiggle room in the fps). In SL I had everything to its highest setting except for avatar vert ix (doesn't work) bumpmaping (also doesn't work) and the draw distance was only 128. In SL the test was in a little suburb and I waited for all the textures to be loaded. I worked to find at what resolution would both have the same FPS. At 1920 x 1200 I got 15.66 FPS in HL2 the lost coast using its stress test. To match that frame rate I had to put Sl at 1020 x 768 with the average FPS at 17. Now that sounds bad but it gets worse when you do the math. SL is rendering 783360 pixels while HL2: TLC is rendering 2304000 pixels. That is around 3 times the number of pixels in a sim empty of people and having reasonably spaced out and rather small houses.

I repeated the test in a mall to see the how another normal place in SL would compare. I did not change the resolution of either game and only changed one setting on HL2: TLC. To get HL2 to match SLs 10 fps in the mall I cranked the filtering up to x16 and got it within 1.5 fps of SL. Please not that sl does not have as many grafical tricks running as HL does which makes this test even more in favor of HL. I had thought about doing a test on equal seting to see the difference, but because I am not sure of the exact seting that HL would have to be turned to to be equal to sl I decided not to.

With both games having the date on disk I can not see why there should be such a huge difference in the to. How can how the information is put onto the hard drive have such a huge difference if FPS when the gfx card is access the data the same way in both? It would seem to me that the problem is not far more then that the information must be streamed in. If someone with more knowledge wants to point to what was wrong with my testing methods or reasoning feel free to.

Professionally made games have their content (textures, amount of polygons used etc) tailored very precisely to ensure best final performance when things get combined.

SL is basically combination of items multiple creators did separately, focusing on the appearance of item in question and little else. Often single wig of prim hair or piece of jewellry will contain more polygons than half of whole level in modern FPS game. A single SL house can have more (and larger) textures than whole zone of WoW or similar game.

SL puts a much higher stress on your computer simply because the content is nowhere near being as optimized as regular games.

Just about everything in SL can be changed on the fly, this means that little or no pre-processing can be done. In HL, the maps are highly optimized (you won't find a single poly that doesn't need to be there), and data structures can be built offline to speed up the rendering. I think HL uses BSP (Binary Space Partitioning) trees to divide up the world. HL can also use things such as pre-baked lighting (radiosity) and normal mapping, to make you think it's rendering a lot more polys that it actually is.

From what I know, this technology is the basis for the next-gen SL renderer as well. There were screenshots posted a long time ago, but the renderer has not yet materialized.

So what you are saying is that it has nothing to do with streaming the content and everything to do with disparent amateurs making the content. I expected that to be part of the problem but I didn't think it would be such a huge problem as it appears to be. I guess our only hope is that IBM's 300 ghz chip actualy works and 1.2 petabyte hard drive comes out at the same time.

Not all of us amateurs make laggy stuff


Personally, I always try to work low prim, simple script, and as small a texture size as I can get away with whenever I'm creating stuff, but not everyone shares that desire.

Many people in SL think only of themselves and don't care how they can affect others - usually this manifests itself in building problems, but more often than not it's the 500 twisted torii prim hair and bling that are the real problems, not the regular builder.

Lewis

I disagree. I think there is a large group of people that would change things if they were simply educated more. I played for a longggggggg time before I saw an explanation about how texture sizes affect load times and such. Previously I was under the impression that no matter what size image I uploaded that SL would optimize it for the game. While SL does convert it to set sizes and such, the load time difference between a 1024x1024 image and 256x256 image was more than 4x if I recall correctly. Once it was explained and I saw the numbers, I spent time retexturing my store and alot of the products.

Perhaps there's an educational opportunity for someone? I'll have to look into seeing how to be a paid instructor, be nice to do something else to benefit the community generally.

Lewis

Being paid mostly benefits you, not the community.

I think lot of it comes from tools not providing enough control, and not enough feedback. When you make things in SL, you're never given information how many polygons your finished thing can make, nor how much room the textures you used take together. You have no ability to enforce some specific quality levels... so often people won't even realize how large their items can be, and even if they realize it, there isn't really much they can do about it.

Finally, there's avatars... these are normally a really large performance killer in any game, and it's typically dealt with by utilizing identical models across the board, sharing the same textures just with different tint, etc. In SL you have everyone trying to look different, and each av means at least 3 custom textures + whatever their hair / shoes / jewellry uses... so this is something where SL is given to lag more than more 'regular' games ^^;

I'd have to go with the "optimizing". Excessively large textures & prim counts, etc.

When building maps/levels for various games, they stress many ways to optimize your levels.

Part of the problem is, SL content is not made by many people with a background in game design.

It's not necessarily that amateurs are making the content: opportunity for mesh optimization is very slim. You may end up using several full prims to approximate a curved surface that could be done in just a few polys, if polys could be manipulated individually. Kick into wireframe mode and take a look at how many polys are used for the complex curved surface of your back. It's beatifully shaped, and beautifully efficient. Now compare that to anything made of prims. This is one of the reasons why it doesn't run as fast as Half Life.

No, actually, that isn't what they are saying. Amatures is part of the problem, but all of the rest of it - BSPs, optimized levels, some of the very fancing lighting techniques... that stuff is hurt by the streaming nature of SL.

Linden Lab wants to support "older" generation graphics hardware, so users without the latest graphics cards can still play. This holds SL back from using the newest graphics features.

Linden Lab makes the newer features optional, so users with less powerful graphics hardware can just disable the features. I suspect, however, that Linden Lab doesn't want too wide of a divergence between users with powerful hardware and users with older hardware.

Linden Lab seems to want everyone to see the same "world," as much as possible.

The newer graphics features are also held back, according to Linden Lab, by buggy and incomplete graphics driver support. SL uses OpenGL as its API, not Direct3D. Before I get flamed by OpenGL fans (hey, I'm one of em) let me explain that the situation is not OpenGL's fault. OpenGL 2.0 has many advanced graphics features. The problem is that the graphics card designers don't fully and consistently support OpenGL 2.0 features in their drivers.

Nvidia and ATI seem to fully support OpenGL 2.0 in their latest cards and drivers. The problem is with older cards and off-brand (non Nvidia or ATI) cards and drivers. Linden Lab can't force graphics card designers and vendors to better support OpenGL 2.0 on older or off-brand cards.

I understand that Linden Lab works with Nvidia and ATI on the issue. I may be mistaken or overestimating how much Nvidia and ATI care about the issue. I think as SL's userbase grows, Nvidia and ATI may take more of an interest.

Some more good news is that the administrators of OpenGL, the ARB, (Architecture Review Board) are handing over administration of OpenGL to a faster-moving group. The ARB have historically been very slow to update OpenGL. I hope to see more advancements and more "push" to use OpenGL from the new management.