Post by mastiphal87 in Eclipse Light Engine comments

Viewing post in Eclipse Light Engine comments

Question about optimization. I've tried using emissive particles, and it seems to be pretty inefficient, even in the example provided. If I have 7 le_light_fire at 1080p, my FPS dips below 51 fps. For context, I can run modern games (e.g., Baldur's Gate 3) on highest settings no problem on this same rig.

When I ran the debug it says 15% of the step was consumed by light_pre_composite, and it looks like surface reset target during light engine other 13 was consuming the highest (3%) of that. Is there a way to do particles/lighting that don't consume so much resources? Everything else runs perfectly.

Badwrong Games1 year ago

The particles rely on GM's built-in particle system, so it could be limited by how efficient they have made things internally. Unfortunately, GM still has a very old graphic API implementation, and yes, you will find performance for lesser things will cost more than newer games like you mentioned.

To troubleshoot, please answer:

What is the actual time in milliseconds that you can see?
Have you ran the same test compiled with YYC? (you'd need to output your own timing or FPS values)
What is your GPU utilization actually at when the FPS is that low? (Task Manager > Performance > GPU)

surface_reset_target() is a pipeline change, but saying that is the most costly thing is actually good, because it is not costly.

15% of the step is fine (but we really need to know milliseconds), and it is normal for rendering to cost the most performance. The question is where is the other 85% going? GM did recently change some things about particles, and there could be a lot of data going between CPU and GPU which would be slow if done poorly.

Badwrong Games1 year ago

For example, I added 20 of those particle fire lights and ran the profiler:

The important ones to look at here are:

Name	Time (Ms)
surface_reset_target	0.01
part_system_drawit	0.042

These are great metrics, and this is with VM of course. So, it would be faster with YYC.

My computer is not a good benchmark though, because I build a new one a few months ago with a 4090 RTX and other high-end parts. I have tested Eclipse on mobile, Surface Pro (2014), and my old Linux machine/server. Never seen a problem with particles, so I'm curious what it looks like for you?

mastiphal871 year ago

I have not tried YYC. Should I simply not use VM?

I did realize that windows has been using my integrated card (not an awful card though can run way better looking games at 60fps), instead of the better card that is used by Baldur's Gate 3. I added my high performance card to the "Game Maker" app, but when I checked in task manager, it still shows the integrated graphics as being used (my guess is I have to add the project app to high performance, but couldn't figure out how to do that).

I tried running a debug again, and this time it is actually much worse (and what I was getting in my actual project). Essentially, if I use the game maker particle system, I experience 0 performance issues. But if I use the eclipse particle system (e.g., le_effect with particle systems and emissives), performance dips very badly. It is nothing to do with my project however, since I recreated it in the example file you've uploaded. I Added 7-8 le_light_fire objects into the room, upscaled to 1080p, and then profiled with these results.

mastiphal871 year ago (3 edits)

I created an executable for the example project, and set the performance to discrete graphics card, and now I've got 60 fps at 1080p with all the particle effects. Do you think the high step % and ms I posted compared to yours is entirely due to the graphics card and the surface reset target function? I would figure out myself if I could figure out how to make windows run my debugging games from game maker using the better graphics card.

Badwrong Games1 year ago(+1)

Ah, that makes some sense then. Integrated graphics are going to be extremely inconsistent depending on what rendering is being done. It is weird that it all sits at surface reset, but how they handle VRAM when its integrated would affect that.

I use a lot of multiple render target (MRT) outputs in the shaders, and I bet integrated graphics do not handle that well at all. So, yes you'll want to figure out why it decides to use your integrated graphics at all. If you do not use it at all, then I would suggest disabling it in your BIOS settings entirely. Then GM will have to choice when selecting a device.

Aaron1 year ago (1 edit) (-1)

Hi, mastiph I am getting the performance issues using shadows, or emissive as well. Like if I inherit from le_game to use textures and create an entity that's a child of le_game, I get a serious performance drop. When I mouse over the blue marker where the drop happens (when I created two instances of such an object, see below for what it does, create a sufrace, and free a surface...no idea what is happening there, perhaps badwrong can shed more light.

I have no issues having hundreds of lights on screen at once however. Yes it taxes more, but not like how only two instances of the object which inherit from le_game do as per the screenshots below!

I am on a 5600x with no integrated graphics, only 3080 graphcis card.

Badwrong Games1 year ago

Your screenshot shows an FPS of 919 with an average of 1479.

Aaron1 year ago

Dropped to 1010 on creation of two instances in the last one. Which is why I was worried I might not be able to have more than 10. But no worries all is well! Its great!!

itch.io

Viewing post in Eclipse Light Engine comments