[Genma Saotome]

darwins, on 01 February 2018 - 01:58 PM, said:

• How do we activate instancing?
  
• Does it work for trains as well as scenery objects?
  
• Presumably if so it would only work for identical shapes - so if we wanted to have a rake of similar carriages with different numbers - then those would presumably be better added as a freight anim. to a common shape...

1. It is an option you can set in one of the tabs on the OR menu. I don't recall now which tab so I suggest looking at experimental first.

2. No rolling stock, just static objects AFAIK.

3. Yes, the shapes have to be identical, including the textures. Instancing 50 identical trees on the same tile will be handled as-if they were just 1. There is an unknown maximum of placed, identical objects. I have a section along a river that includes rip rap along the roadbed. There are too many identical shapes there and the result is the boulders jump up and down in OR. I have to go and rename a bunch of them to get fewer of the same .s file name. A bother but in the end the rendering cost will still be vastly less that if all of those placed models were rendered one by one.

[Genma Saotome]

espee, on 02 February 2018 - 07:48 AM, said:

If you make a model for yourself, and never intend to release it to the public, don't run it in an activity, don't use it as an AI etc etc and can live with lower frame rates, then you don't need LOD's...

:rotfl:

[Genma Saotome]

I'd like to add that there is an understandable ratio between your screen resolution and what any given width is equal one pixel at set distances from the camera. IIRC at 1920x1080 (or 1200) one inch of model width is one pixel at 125m view distance, which means a 4inch width is one pixel at 500m, 8inch width is one pixel at 1,000m, 16 inches at 2000m and 80 inches is one pixel at 10,000m.

I tend to use a 100m view distance instead of 125, just to keep it simple.

Using that guideline (plus my use of 100m instead of 125) to set LOD's means a 33" freight car wheel is just 2 pixels at 3300m. Perhaps an LOD at that distance, perhaps an LOD at some distance less than that... the wheel is 4 pixels at 1650m, 8 pixels at 825. The "answer" is obviously a judgement call. My point is knowing this ratio lets you make well reasoned decisions consistently when it comes to setting up LOD's in your model.

[Guille592]

espee, on 02 February 2018 - 07:48 AM, said:

If you make a model for yourself, and never intend to release it to the public, don't run it in an activity, don't use it as an AI etc etc and can live with lower frame rates, then you don't need LOD's...

That's curious, cause I've distributed for some friends, we have done multiplayer and or I have activities with AI, and I really don't get much problems. :/

[darwins]

Reading all this, I can see not only a need for LoD, but also a need for lower poly trains for static stock and AI trains - especially for busy routes with full sidings.

Having said that I am thinking of making some "lower poly" models, which still have a more polys than some of the low poly MSTS stock of the past, and make better use of textures and materials.

[ErickC]

I think it's a good idea, but I'm biased because I had the same thought. I find that it's a good way to balance my workload, too. The hi-res models drive you crazy after a while, it's nice to work on some lower-detail locomotives because they can be built much more quickly.

I did have a thought not too long ago. The problem with distance-based LODs is that the camera might be zoomed quite far in but still be a fair distance away. I know MSFS uses a percentage-of-screen system for that very reason (that is, the LOD level changes with the amount of screen space that the object takes up). Might it be worth considering? It might be too complicated to implement.

[scottb613]

Hi Folks,

Hope everyone is well - been years - I know - Derrick pinged me with a 3DC question - so it started me poking around around in my old haunts once again...

Perforrmance - my favorite topic...

I’ve always been a poly minimalist - trying to do more through textures when I can... I’ve tested performance pretty extensively in various modeling projects - most of which have been documented here... LOD’s make a huge difference in performance - maybe not as much with the locomotive - as that’s where you spend 99% of your time - but when pulling 100 car trains - it’s essential...

When modeling - you need to keep three things in mind for good performance:

• LOD’s
  
• Draw Call’s
  
• Collapse the model heirarchy before export

I’ve always used 3DC and while I’ve heard contradicting opinions - I recall seeing substantial performance gains when I took the time to merge all my parts down to the bare minimum - before export... This testing was all done on the Feather River with 100 car trains... My final parts hierarchy I shoot for is bare minimum - just the bogies and wheels - with the entire car itself being a single part... This performed substantially better than when I had twenty or thirty parts making up the body of the freight car... Texture size seemed to have little or no affect...

Anyway - just my two cents...

Regards,
Scott

[ianmacmillan]

All my models have 4 LODs.

However I use Polymaster to replace high poly parts with low poly parts when they are hard to see.

My wheels take 136polys but they only appear for the first 300m. They are then replaced by 36 poly versions. At 300m the wheels are only a few pixels on the screen. The low poly wheels are not animated saving the processor some work.

http://www.atomic-album.com/showPic.php/80489/LowPolyWheelset.jpg

http://www.atomic-album.com/showPic.php/80489/PolymasterWheel.jpg

The low poly wheels have square faces made round with the alpha channel.
The rims are a 5 sided tube with the top ploy deleted while the axle is a 3 sided tube with the bottom ploy deleted.

I also replace detailed roofs with basic versions.

Saving a few polys on a simple wagon might not make much difference but they soon add up on a long train.

[ErickC]

You see a performance boost when you join parts because it reduces the number of drawcalls.

One thing that helps - that most people skip - is to ensure that as many UV coordinates as possible are welded. This reduces the number of texture vertices, which is also pretty important.

[Genma Saotome]

ErickC, on 19 February 2018 - 04:53 PM, said:

You see a performance boost when you join parts because it reduces the number of drawcalls.

Yup. Ideally there are no named parts other than what MSTS requires.

ErickC, on 19 February 2018 - 04:53 PM, said:

One thing that helps - that most people skip - is to ensure that as many UV coordinates as possible are welded. This reduces the number of texture vertices, which is also pretty important.

I've long been curious about the use of the verb weld as it is used in these forums. It's not a concept I recognize when working in Sketchup... it is tool specific? A product of how one does the work? Or is it something that is inherent to all such modeling?