Giant deathtrap: I want one! (also lions and tigers)

[Gumboots]

So I had another look at the mesh and rationalised a few things. It'll still look like the last pictures, but come in around 2,700 for verts+tris (was up over 3,200 in the last pix). Verts are about the same as the default PopTop Mikado, but it'll be smoother looking because it packs more tris into the same number of verts. So more tris to render, which is bad for performance, but packed into fewer separate meshes, which is good for performance. This is on highest LOD of course. Lowest LOD will be down around PopTop levels anyway.

Short version is I don't think it will be much rougher on the game engine than the most detailed of the default locos.

I have about half the UV mapping roughed out already, but there will be a bit of juggling here and there before it's finalised.

[Gumboots]

Got a bit more done. Only the chassis and some small thingummys left to do.

[Gumboots]

Had to modify the beast again. I made the mistake of doing some research, which always leads to trouble.

Turns out the shiny brass cover over the "Haycock firebox" is totally fake. It was installed back in the 1920's when the LMS Railway (who owned what was left of Lion at the time) decided to do a "restoration" as part of a celebration they were having. Part of that was the shiny brass look. They must have thought Lion needed some party clothes for the occasion.

The original firebox is still underneath there and is an iron round top, with a gruesome rivetty turret on top that does the job of a steam dome. The original chimney was also simpler, without the brass shiny stuff. This looks more freighty anyway. There are pictures of the original style here, and a good article about Lion's history (including the party clothes) here.

Apparently all the original bolts were made by walloping the end of an iron rod into a lump, and then filing the flats for the bolt head by hand. The result of this is that some of them required the use of three different spanners during the restoration work, depending on which faces of the bolt head you were trying to get the spanner on.

The other fun thing the article mentions is that you can chuck Lion into reverse while trundling along at full choof forwards. They say this is handy for when you need brakes, because Lion doesn't actually have any to speak of.

I hadn't thought of this before, but it does give an idea of how early loco drivers managed to survive the trip with little or no braking installed. They probably just threw her into reverse when they wanted to slow down. As long as it was done sensibly, it'd work just like you use engine braking in a car or on a bike.

Anyway, the original firebox and chimney have been modelled now. I've been doing some cunning trimming here and there, and the model now has 100 fewer verts than a default Mikado, while still looking just as good as it did before. Currently it has 1,157 verts and 1,440 tris, for a total count of 2,597.

[CeeBee]

Look ma..... no brakes. Thats classic.
Keep up the good work. I't looking better every day.

[Just Crazy Jim]

there's a bit about that no brakes on trains in Britain's Greatest Machines - S02E04: Trains - The Steam Pioneers (https://youtu.be/wOGYZC-IJPQ)

Life as a railway man before the Westinghouse Air Brake must have been very risky.

[Gumboots]

Hey I had an idea about the practical limits for verts and tris in RT3 loco modelling. I already knew that a total count up around 4,300 (double-headed H10 Mikado) was pushing things a bit too far. It can cause noticeable frame rate drop at times, even though it is still perfectly usable on my box with 250 of them running at once. So I'd come up with my theory that somewhere around 3,200 would be ok, being about the same as one and a half Mikados.

Thing is I hadn't thought of actually testing this before scratch building models to that limit. So this morning I think of that, and then start wondering if there is another default model that can be double-headed to give a good test of verts+tris around 3,200. It already exists. The 2-6-4T Suburban Tank that I made has 890 vets and 813 tris, for a total of 1,703, so the double-headed version obviously comes in at 3,406. If that runs ok on most people's boxes then 3,400 is a reasonable limit to use.

I have tested the 2-6-4T Double (obviously) and I've never noticed any problems. Nobody else has reported problems either. The map I was using to test both (single and double versions) was quite small, and was only running about 20 of each (total roster somewhere around 40). I haven't yet tried with doubles en masse, or when running the debugging tools to check frame rate, so that's something I should do this weekend.

Edit: Have just done this. Wall of text over here.

Short version for sensible people who just want to play with trains: it seems to be ok.

[Gumboots]

Have done some more cleaning up on Lion. I found some mistakes that needed fixing. For example, when I changed the chimney from the fancy one to the original style, I scaled the top for length but not for width. It was severely elliptical up there and I hadn't spotted it before.

So fixed that, and got rid of some stray verts that had somehow appeared, and a few other bits of crud. Have also changed the guard rails for the footplate. The ones that are on Lion now are more stuff from the fanciful "restoration" back in the 1920's. I've changed it to use rails the same as the Adler, which is an as-near-to-genuine-as-you-can-get replica of a Stephenson loco from the same period, and has a photo of the original to prove it.



So this style of rail is easier to model with simple mesh, and for that matter would be more practical for keeping the crew's feet on the footplate. On the basis that I know Stephenson used this style, I'm putting it on Lion too.

The old photo of the Adler shows timber lagging around the firebox as well as the boiler, but for some reason the replica doesn't have it on the firebox. My guess is that the timber is just too prone to damage from the extra heat there, while being fine on the boiler itself where it would be much cooler. So I'm going to go with my gut and skin it to have just the timber around the boiler, and the firebox in plain iron.

I'm thinking I'll also do it as painted timber around the boiler rather than the varnished timber that Lion acquired during the restoration. Anyone who has tried to maintain varnish out in the weather would not be inclined to use it on a freighter. Paint is much more sensible.

It's now good enough to call it finished as far as mesh goes. This means I just re-do most of the original UV mapping (which got borked by some of the changes) and sort out some LOD's (not hard to do) and Lion will be ready for skinning and testing. Hopefully I should have it sorted over the course of the next week.

[Gumboots]

I haven't forgotten about Lion just because of those cargo cars. He's still moving ahead. The revamped UV mapping is now 99% sorted. Well ok, maybe 95%, but I can knock off the other 5% easily tonight.

So there will still be skinning coming up next, followed by a usable locomotive in the not too distant future.

Edit: Knocked off the other 5% last night. Lion is now fully mapped and reading for skinning.

[Gumboots]

Thought I'd better do the LOD's before I started skinning. Took a look at it, and realised I'd left 8 completely redundant tris in the top LOD on the loco, and could simplify the tender's top LOD too and get rid of 8 verts and 8 tris there. It also gives a better transition down to the lowest LOD for the tender, which is a bonus. It will require a small adjustment to the tender UV mapping, but that's not a problem. Will sort it out tonight.

[Gumboots]

Have been thinking some more about LOD's. I did have the best model detail and draw distance settings worked out, but that was before I got the graphics upgrade courtesy of the Voodoo patch. So, I should check all of that again and see if those same settings are still the best.

Once I have the best graphics settings sorted, what I'll then do is load up a test version of the model with two LOD's, both of which are identical. This might sound stupid, but there is method in my madness.

I can whip up arbitrary LOD's based on looking at how the default models did them, and using a bit of semi-edumacated guesstimation. In fact I've already done that. The thing is that I now know that a lot of default modelling isn't actually that good. Some of it is pretty bad, with the LNER A1 4-6-2 being a great example. So my cunning plan is that if I load a model with two identical (most detailed) LOD's I can then check how it looks at the distance when it crosses over from the first to the second. This will tell me exactly how much mesh I can strip before it becomes noticeable.

I can then repeat that with the next level down, and all the way out to the last one. Once I've done a few locomotives like this I'll get a good feel for it, but it seems like a good idea to check the first few in search of the best result.

The good news is that it's looking like I'll be able to strip mesh faster than the default models. This is because the extra detail I'm going for (like more faces on boilers, so they aint chunky) is stuff that's only relevant at fairly close range. So although the top LOD will be heavier on the graphics card than the default models, the intermediate and lowest LOD's shouldn't be. Just as an example, comparing the Default H10 and the current Lion beta:

Locomotive Body .3dp only.
----------------------------------------------------------------------------------

Default H10 Mikado LOD's.
--------------------------------

0 - 947 verts, 727 tris.
1 - 573 verts, 500 tris.
2 - 341 verts, 289 tris.
3 - 181 verts, 126 tris.
4 - 52 verts, 32 tris.

Double H10 Mikado LOD's.
--------------------------------

0 - 1894 verts, 1454 tris.
1 - 1146 verts, 1000 tris.
2 - 682 verts, 578 tris.
3 - 362 verts, 252 tris.
4 - 104 verts, 64 tris.

Lion current beta (17/11 am).
----------------------------------

0 - 1031 verts, 1146 tris.
1 - 584 verts, 554 tris.
2 - 259 verts, 244 tris.
3 - 131 verts, 110 tris.
4 - 60 verts, 42 tris.

So Lion is starting out around 50% heavier on tris than the default (single) H10, and around 20% less than the double H10's, but the second LOD is only 10% heavier than the single H10, and the third and fourth LOD's Lion are about 10% lighter on tris than the single H10. The extra 10 tris on the lowest level won't make much difference.

[Gumboots]

By the way, this one hasn't been forgotten. Although I will admit to getting sidetracked.

What happened is that after doing Moguls for Christmas I got into the Pennsy H3, as a natural progression from Mogul to Connie to something else, so we had a decent US/World freight hauler for the end of the 19th century when the Connie gets outclassed. So that's sorted now.

I did take another look at Lion last week, but having learned more in the meantime I realised I could do a better job of it. Short version is I can cut rendering loads by around 15% without making it look any worse, and in fact making it look more accurate to the original form, but this will require rejigging a few things.

Rejigging a few things will happen once Samson is finished. Samson will be very useful for the 1830's, and is the direct ancestor of Lion, so I thought I might as well start at the start and go from there. Lion is next on the list.

[bombardiere]

Anyway, the original firebox and chimney have been modelled now. I've been doing some cunning trimming here and there, and the model now has 100 fewer verts than a default Mikado, while still looking just as good as it did before. Currently it has 1,157 verts and 1,440 tris, for a total count of 2,597.

LMR_57_Lion.blend.jpg

I don't understand how you do that with only 2597 total count. I really need to learn that skill.

Lion is one of engines I wanted to do for RRT3. :) So thanks for this project. :)

I do remember this engine always as "Titfield Thunderbolt." That was a great railway movie. :)

[Gumboots]

Well the short version is you do it by saving tris wherever you can. Which I suppose is obvious, and not much help.

I've done some more testing since then. It appears that for the sorts of models we are interested in, running on the RT3 game engine, the basic tris count (or poly count if you prefer to call it that) is a near enough proxy for rendering load and frame rate. So I don't worry about the verts any more, and just keep an eye on total tris.

I used the Pennsy H3 as a test case for several things. It comes in at 1,500 tris, and just to see what skin quality was possible I used the one image for both locomotive and tender. I figured it was good practice, since later and larger locomotives will have to be done at similar resolution even if you have the luxury of using separate loco and tender images.

I've found that anything up to around 1,500 tris is not a problem. Once it starts getting up around 2,000 it will cause a noticeable hit to frame rate sometimes. These counts are totals for locomotive and tender combined, including all components.

But obviously less is always going to be better for performance. Using alpha instead of mesh is a good idea where it is possible. For example, the Pennsy H3 has a boiler made from an 18-sided cylinder. That's enough sides to look smooth with the livery it carries. For something like an LNER A1, where the livery has much more distinct lining on the boiler, I'd use 24 sides to stop the lining looking blocky. I decide on this by checking the model out from a range of angles and distances and seeing what makes sense.

So if the front of the smokebox was done with 18 sides it would obviously rack up another 16 tris* (if flat) or 18 tris (if peaked in the middle). But if the front of the smokebox is done as a basic square with alpha it will only be two tris, and as long as it's done to a large enough scale it will look fine in the game.

You can sometimes use a single tri instead of a square too. For the Pennsy H3 I used single tris for all wheel layers. Each locomotive wheel has four layers. Since the tender wheels are not quite as noticeable in-game, I only used three layers for those. Adding them all up, the locomotive has 10 wheels at four tris each and the tender has 8 wheels at 3 tris each, so the total is 10x4 + 8x3 + 64 tris. If I used a square instead of a triangle this would double the number of tris, so compared to that I saved 64 tris just in the wheels. I thought this was worth experimenting with because once you start getting into Garratts and Mallets the numbers of tris in the wheels starts rapidly adding up.

The only catch here is that using a single large triangle will use up more space on the texture for a given resolution. A square is more compact for a given amount of pixellation/blockiness/edge jaggy/whatever you want to call it. The in-game geometry isn't a problem, because an equilateral triangle will be perfect if you set the radius (ie: centre to verts) to be exactly double the wheel diameter.

Oh and by the way, drivetrains are a totally known quantity now. Setting up pistons, conrods and coupling bars is simple and reliable. It's even possible to stagger the two sides of the drivetrain by 180 degrees and have them work perfectly, but you can't quarter them because a 90 degree stagger will break piston animation.

*Minimum number of tris for any flat surface is N-2, where N is the number of sides around the perimeter.