Jump to content

PE and full-region mesh builds.


Drongle McMahon
 Share

You are about to reply to a thread that has been inactive for 3791 days.

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

Recommended Posts

I would like to introduce an aspct of the effects of the high PE of mesh that I don't think we have addressed yet. That is the effect on full-sim builds.

This occurred to me durin a revisit to Maxwell Graf's Rustica, a masterpiece of sculpty building, both artistically and technically, not so many wasted triangles here, an object lesson in the benefits of instancing. On my seven year old computer with it's recently upgraded, but low-powered GT440 gpu, on it's 1.75 Mb/s connection, I can walk around here on medium settings with a perfectly acceptable 30 fps and tolerable waiting time for sculpties to rez. Looking at the scene statistics console, still on medium settings, the visibe triangle counts vary between 2,000,000 and 3,500,000 (eg 3500/12500 Ktri). That is up to fourteen times the target count that is being used to limit mesh.

Now it is possible that users with much less capable machines will find unacceptable frame rates when they visit this build. But what I am forced to ask is whether that should prevent Maxwell from building it, or whether it should deprive the rest of us the pleasure of viusiting it. The answer has to be no. This is among the best that the grid has to offer. To make it impossible would be unthikable.

But with mesh, that is exactly what the effect of the high PE costs is. It will be impossible to construct builds of this quality with mesh. In an attempt to protect the framerate of the lowest common denominator of client machines, an attempt that is bound to fail because it is restricted to mesh, everyone else will be prevented from making or enjoying mesh builds of this quality. This is regardless of the easier building, regardless of the wider pool of talent that could build them, regardless of the savings in efficiency that would be achieved.

I have tried to be balanced in this debate, but looking at Rustica and thinking that nobody will be allowed to attempt such a thing in mesh makes that a pretty hard to maintain. Surely if the poisoning of the potential market in static mesh is a shooting in the foot, then this is complete amputation of both legs.

Please, somebody, tell me why I am wrong about this.

PS. I guess the minimum L$150 per mesh upload might be even more effectively preventative, although it would reward the instancing.

rusticahanging.jpg

Link to comment
Share on other sites


Vivienne Schell wrote:

Drongle, better let the creator speak for himself. Using someone else´s work for your biased cause is fishy.

I am confused. Is there anything wrong with using examples from others to express ones ideas and thoughts ? And as there is only positive feedback towards the creator in Drongle's post, i can not see exactly what you are criticising exactly.

Can you explain in which way Drongle is biased ? Honestly i can not see it, maybe because i myself am also biased..., But maybe you can show where the argumentation is wrong ?

Link to comment
Share on other sites

I wouldnt open the can of worms Gaia, Vivienne has an unlimited supply of them.

 

Sim/Land owners having more options in regards to meshes would be a great solution, as mentioned in other threads, if a sim/land owner can designate thier plot to be mesh only it should allow more flexibility in the amount of mesh that can be used. Not only would this be good for the servers or said sims but it would also go a very long way to making mesh much more appealing and give owners more control over how much lag visitors can inflict (the mesh only option could also have a sub option of applying to avatars aswell)

It allows the community to determine what they want giving Linden complete freedom from the burden of imposing rules that could affect the entire grid.

Link to comment
Share on other sites

Wrong is that it´s Drongle who claims that the "high" PE and whatever will prevent Max from rebuilding this in mesh. If it were Max expressing his wish to rebuild this in mesh at all and complain I´d have no problem with it.

Isn´t the mesh beta open for a while now (Almost two years, closed beta included?) Didn´t you all have enough time to build and upload a full sim in mesh for *no* upload fees to give an example?.If you did not, please explain why not.

Link to comment
Share on other sites


Vivienne Schell wrote:

I don´t need to find a problem in the meshie forum, the problems jump on me for some reason. Maybe it´s due to the high PE of these darned problems.

Actually I think it's because you're a troll that believes mesh is the work of the devil. Quit holding back, it's pretty obvious from your posts that you will never accept mesh in any way shape or form. Just come out and say what you really want, which is for LL to drop mesh entirely.

Link to comment
Share on other sites


Vivienne Schell wrote:

Wrong is that it´s Drongle who claims that the "high" PE and whatever will prevent Max from rebuilding this in mesh. If it were Max expressing his wish to rebuild this in mesh at all and complain I´d have no problem with it.

But couldn't you (anybody) use Max's build for explaining what would happen to anybody who attempts to create such a full environment based on meshes? Using a practical example to explain a general problem seems legal to me.


Vivienne Schell wrote:

Isn´t the mesh beta open for a while now (Almost two years, closed beta included?) Didn´t you all have enough time to build and upload a full sim in mesh for *no* upload fees to give an example?.If you did not, please explain why not.

Because it is a Beta which gets scratched very often. Meshes have to be reuploaded all the time, Regions have to be rebuilt all the time, Most regions are sandboxes with return in 48 hours...

BTW have you ever seen the museum built by drongle ? That counts for me already as an environment (although not a full sim) . But i think that Drongle might be one of those who actually have practical experiences in working with whole mesh environments...

Link to comment
Share on other sites

I do not want LL to drop mesh entirely, i expect LL to embed collada imports and in-world mesh tools into the existing platform in a way which will not harm the existing platform. And yes, i´d prefer a focus on fixing of much more urgent trouble for any not urgently needed shiny.

Link to comment
Share on other sites


Vivienne Schell wrote:

I did not crticise Drongle for THIS (btw, did he ever upload his museum - or parts of it - on the mesh beta?).

I am talking about Mesh beta... So the museum is part of it. You can find a very short preview on the
(the last 20 seconds of the video shows a few imressions of it)

Link to comment
Share on other sites

Your results are in accord with what I see in SL and other graphics engines.  A reasonable graphics card can render a lot more than 250K triangles without slowing down. From my own tests, I extrapolate that an Nvidia 7600 class card (relatively slow these days), should be able to handle 800K triangles on the SL low graphics setting.

For comparison, the CryEngine 2 recommended limiting scenes to 3-4 million triangles, using a more complex shader with multiple texture maps/triangle (typically 3) and would perform well with a GTX 8800 class card.  That would translate to 6 million triangles in SL because we only use 1 texture map/triangle.

And whatever the parcel costs of using mesh objects, if avatars are not charged for heavy use they will load up on high detail meshes just like they do on prim and sculpt heavy items now.  Therefore people with weak graphics will still experience lag in busy areas.

The bottom line is the "cost" being calculated for mesh is so much higher than for prims and sculpts that people will not use them on parcels. A penalty of ~2x on prim cost would let the other advantages of mesh compete (custom UV maps, 8 textures vs 1 for sculpt), but with the current penalty of up to 10x they will simply not be used for many items.

Link to comment
Share on other sites

Vivienne. Sorry again for not making myself clear enough. I can't think of any reason why the creator of the region in question would want to rebuild it in mesh. As far as I am concerned, it is much better than satisfactory as it is. My only purpose was to give an example of the best kind of build that is possible with existing tools but that might be beyond the reach of mesh with bthe current accounting. I am certainly not good enough an artist to make anything like that myself, mesh or otherwise. As for uploading a whole region to Aditi, I think I must have uploaded my gallery about six times over, what with all the invalidating changes, strange disappearances etc. Most of the other original constructors of Mesh City gave up a long time ago. I guess I am just obstinate. Having to do a whole region over and over again would have stopped me too. I wish you would believe that nobody is trying to stop or invalidate the existing building methods, only to supplement them with another option.

  • Like 1
Link to comment
Share on other sites

Drongle, my apologises if you felt attacked. It was not meant to be a personal attack. I know your home parcel, read your posts and you certainly know what you are talking about. Unfortunately not everyone here on this forum wants to supplement.

Link to comment
Share on other sites

Noe, while I agree (to an extent) the comparisons you are using are a little off.

 

Comparing the CryEngine2 to the render pipeline in SL is a bit disengenious.  Scenes for the CryEngine (and most game engines for FPS games) have to be pre-compiled and optimized.  They don't change dynamically while playing (the terrain does not actually change, trees don't change position, etc....things can animate and such, but the definitions of the scene remain fixed) where the SL environment does.

 

That said, I still agree with you.  To wit:

 

The nVidia GeForce MX 440  card from back in the old AGP8x days (a decade ago, and way below the 'minimum' requirements listed for SL), could T&L 34 MTri/s.  So assuming very low graphics settings, such a card (with a 2.5MTri/s per region limit, which works out to an average of 167 tri/prim) would still deliver around 14 fps.  (yes, this is simplified, as there are a lot of other factors....setup time before the T&L pipeline as well as cache changes would tend to push this down quite a bit.)

 

Even half that (7 fps) is playable.....not very aestetically pleasing, but doable.  And this is based on a card that is over a decade old.

 

Now, the minimum suggested graphics is the same class as the nVidia 6600 (mid 2004 date, about 7 years old tech) which clocks in at around 375 MTri/s......That's 10 TIMES the performance.  This means the MINIMUM suggested card, on low graphics settings in SL, should be getting 70+ FPS.  (In fact, with the improvements in the setup pipeline and memory for cache, in general the performance will be much closer to the theoretical top......in this case, 150 fps.)

 

Put simply, if you are running the minimum suggested GPU or better, you should be able to (with the graphics settings ALL THE WAY DOWN, except for draw-distance) get 60+ fps easily.

 

They could triple the given 'max' triangle count in view and STILL not hurt frame rate enough to be an issue.

 

(note:  due to skinning, pixel-shaders, and more, these calculations are strictly theoretical maximums.  But, as I said, this is the minimum suggested card on SL's own System Requirements page......you would not expect to get 70 fps on Ultra with a card that old and outdated.)

 

Now, let's compare to a couple ol years later.....the nVidia GeForce 7600GT GO (yes, that's right, an integrated mobile GPU.)  It's rated for peak performance of 550 MTri/s.  Almost 16 TIMES the performance of the old MX440, and almost 1.5 times the performance of the 6600 series.  And it's a MOBILE GPU.

 

The current 'triangle budgets' are woefully outdated.

 

I would consider a 15 fps render rate to be the baseline.  For the 6600 series, that would translate (at peak) to 25 MTris in view.  Assume it for the whole region.  That's still 10 TIMES the current budget.

 

Either there are render cost we aren't being told about (which could reduce it considerably), or the server costs for physics are completely overblowing the rendering speed.  If that's the case, they need to invest in some nvidia physX cards for the servers and take some of that physics load OFF the CPUs.  Yeah, it'd be a major re-write (unless the utilized Havok version supports it directly) but the benefits to the visual quality budget would be enormous.

 

If we look at a midrange card (GeForce GT450, multiple kinds available at newegg for $100-$150), then the vertex rate (they stopped rating tri/s a long time ago) is about 37600 MVtx/s (which would work out to around 12800 MTri/s)......assuming a peak rate, at 60 fps, that gives us 213 MTri per region.....about 75 TIMES what the budget is.

 

Note again, I haven't touched gamer/enthusiast cards......this has all been mid range and below.  And, I've used peak ratings for the most part.  Even given all that....with the minimum suggested system requirements GPU, the 'triangle budget' they are basing these calcluations off of is woefully outdated.

 

If they are basing it off of physics, they need to realize that most physics shapes will be optimized and less complex, so the budget relationship should be a LOT higher than it is currently.  Penalize heavy physics cost (since that is server side), but don't limit static, simple, optimized meshes the same heavy-handed way.

 

 

 

Link to comment
Share on other sites

70fps on an nvidia 6600? woah. my nvidia 5400 only gets 40fps on failed login or draw distance of zero where there is absolutely no geometry to render. I get 15 fps on my fairly simple home region, and most places get less than that

Link to comment
Share on other sites

Nvidia GeForce FX5600 had a triangle rate of 100 MTri/s.  That's about about 1/4the the rate of the GeForce FX6600.  I can't even find info for a FX5400.  The FX5200 had a rate of 81 MTri/s.  So, if the 6600 could get a max of 70fps, I'd expect the 5600 to get about 20-25 fps maximum.  With the memory limit (FX5600 had a max of 256MB VRAM), most SL scenes would be texture bound, more than triangle bound.

 

Realize, as you go further and further back, the memory limits (due to textures) will overrun the render speed.  If a card only has 256MB of RAM, and the scene you are rendering uses more than that in textures, it's having to reload textures from main memory every frame.  Can make a big hit.

 

There are other considerations.....render size (which is affected by pixel-fill rate) means more pixels for a given triangle in a frame.  It also means more texels.  Older cards don't scale up as well.

 

There are other limits to the reported render speed in the 'statistics bar' that aren't representative of the actual rendering speed.  If the viewer is busy calculating animations, switching texture coords, changing other things, and handling a lot of other stuff, even with multi-threaded rendering turned on you can get delays that have nothing to do with the full-on rendering speed.  (I did mention this in my notes....though perhaps not explicitly enough!)

 

For most purposes, I base realistic performance at about HALF of theoretical.  So, I'd expect an nVidia GeForce FX5600 to get around 10-12 fps typically in SL.  The FX6600 (the suggested minimum supported) would get around 35 fps.  These are assuming you have a few additional things turned on, and running at a decent resolution.

 

(I should note here, my older machine is running a GeForce FX7950 GTKO, and with custom High graphics settings in SL at 1680x1050 resolution, I'm still getting 30-40 fps.  (as a note, the 7900 was rated at 1400 MTri/s, almost 4 times more powerful than the 6600....)

 

 

Link to comment
Share on other sites

Firstly, let me say thank you for the compliments of my work, Drongle. I certainly have absolutely no issue with you using it as an example in making what is, I think, a very valid point. While I dont want to recreate the sim using mesh, you do point out what a lot of us are concerned about; That is, if I wanted to redo it with mesh, it would end up being considerably limited. The $L and prim counts would be prohibitive.

I, for one, am looking forward to the challenges and opportunities that mesh presents, although Im already going to need therapy from trying to figure out some of those challenges. I do feel that, as the dust settles, it will become simply another tool in the toolbox for creators to choose from rather than a replacement for anything that currently exists. I cant wait to see what people do with it, but I dont see a future where SL is a mesh world, with older content passing out of use. I am also saddened by the growing limitations of mesh, which seem to be making it less viable as a build alternative as we get nearer to release.

I'd like to add my personal desire to see some form of parametric deformer for worn mesh items, as this would serve the greater purpose of making it much more usable, much more sellable, adopted by a greater portion of existing designers and much more (pardon the pun) flexible. It would also negate the need for rigging, which seems to be one of the more difficult aspects of making something to wear for many people. Ideally, worn mesh items should be developed with adoptability and usability in mind, considering the existing market for worn items. This is, of course, only my opinion, but I voice it after having experience the exact same issue with worn items in bluemars. The parametric deformer they created in response to the problem was a very viable solution, one which made clothing and worn item design much more accessable and marketable to a wider segment of the population there. It would have the same effect, solve the same problems, here.

Thanks again for using Rustica as an example here. Its nice to be mentioned among such esteemed company. :)

 

Link to comment
Share on other sites

The only proper way to test render speed in SL is to actually use slower graphics cards in suitably filled regions.  Unfortunately I gave away my 9 year old PC with the Geforce 3 that I first started in SL with.  So I don't have a slow card to use any more.  The best I could do was fill up a Mesh Sandbox with high triangle objects and see what happens with my current card (Nvidia 260), and then estimate how that scales to other cards.  I'd be happy to set up a bunch of objects as a test area for other people to visit and see what they get.

Link to comment
Share on other sites

Kinda makes you scratch your head. LL's main business is selling land, but the best looking, and most efficient ways to make that land look good is being put out of our reach. Hey, but the 1 thing that can affect a sim more than anything, the avatar, is given even more power to lag every1.

Is the avatar the way to success? I think not. IMHO, more avatar features or models are not going to change anything in SL, at least as far as bringing more people in. What will bring more people are better full sim builds, more interactive environments, more fun stuff.  If you ask me, all the limits put on mesh is completely counter productive.

Link to comment
Share on other sites

You are about to reply to a thread that has been inactive for 3791 days.

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

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
 Share

×
×
  • Create New...