How are custom DFFs/COLs/TXDs handled?

[DiSaMe]

A map in my server has many objects with custom models. But loading the whole map gives me some problems.

If I load and replace all textures and models when the resource starts client-side, SA loads very slow, so many objects (those of the original map) are invisible and the game freezes when I go to area with custom objects. If I only load and replace models, they're white, but everything else seems to work fine. I also tried to make a streamer which loads and replaces models and textures when the objects are near enough, but it's less stable - some objects lack textures, and it tends to crash client.

I have also looked at gta_sa.exe memory usage via task manager. When MTA loads custom DFFs, COLs and TXDs, memory usage increases. But when a few thousands are loaded, it drops to a very low level, around 15000 KB. So I have two questions: how are those files handled, are they loaded into game when engineLoad*** functions are used, or is the game just directed to use custom files instead originals when object with that ID is created? And in which way (streaming or loading when resource starts) would you recommend me to use custom models?

[Xierra]

Alright, this was my same purpose question as my stream fixes topic.

So if you load lots of custom models, can cause the GTA objects to f*ck up and make freezes. This is because the GTA's streaming limits, this also happens on GTA SA. So I think I also need to submit to add stream fixes, but in SA-MP, this doesn't happen.

[Deltanic]

Well, the streaming limit isn't the real big problem. MTA supports 250 objects streamed in per player, this includes custom placed models because they are still object elements.

Besides that: Watch your polycount. San Andreas can handle a lot of polys for vehicles, but that's really different for objects. Too many polys per object/collsision will crash the RenderWare engine of San Andreas.

Keep the polycount of your collisions below 6000 per object. San Andreas can handle 8000 maximum, but it's better to use less than that.

I forgot how many polys you could use in your DFF files, but this was MUCH higher than the max polycount for collisions.

[DiSaMe]

Well, the streaming limit isn't the real big problem. MTA supports 250 objects streamed in per player, this includes custom placed models because they are still object elements.

You misunderstood me. By saying "streamer" I didn't mean "MTA built-in streamer", I was talking about my own script which loads and replaces models when you are near object and unloads/restores models when you are far.

And to be more precise, I don't think there's something wrong with polycount. The map is Liberty City, converted from GTA3. I created programs to convert files myself , and they're not perfect. A few models crash the game, but when I use my custom model streamer, the game crashes more, so there's something unstable in MTA too, isn't there?

[50p]

Well, the streaming limit isn't the real big problem. MTA supports 250 objects streamed in per player, this includes custom placed models because they are still object elements.

Besides that: Watch your polycount. San Andreas can handle a lot of polys for vehicles, but that's really different for objects. Too many polys per object/collsision will crash the RenderWare engine of San Andreas.

Keep the polycount of your collisions below 6000 per object. San Andreas can handle 8000 maximum, but it's better to use less than that.

I forgot how many polys you could use in your DFF files, but this was MUCH higher than the max polycount for collisions.

There are no polygon limits that I know of. The only limit is player's PC.

@Doomed_Space_Marine

I've been experiencing the same problem myself recently even though my models are very low poly and texture resolutions are not greater than 256x256. I load 5 models and after loading 6th model, FPS drops down from 36 (default limit set by server) to 25-26 and only a few of these models are streamed in. So, it has something to do with MTA rather than GTA or your PC.

[Maccer.]

The high memory usage is a confirmed memory leak. The custom model system is very buggy and weird in general, and could require some rework. But... I don't really understand the question you're trying to ask.