A theory about frame loss

[avc15]

So we've been around and around on this. Something occurred to me recently. We've talked about all the systems that raise game load and tried lots of things. At some point, nothing works:

• Deleting debris
• Removing piping, gas vents, conveyors...
• Replacing empty space with tiles
• Radial pathing (exactly one path to any given location)
• Killing critters
• .... it goes on and on

One thing nobody has ever touched on in one of these discussions is complexity of power systems

See the thing that occurred to me is, we can observe how all of the bullet points above work, and we have some idea of their complexity. Some of those systems have been isolated to their own contexts (threads) and so they shouldn't bottleneck our game unless there's a synchronizing issue (ie if that context becomes much more work than all other contexts). So in our testing mindset we go delete all the things and see what happens to frames. No change.

But what about the power system? We can't really observe how it works, so there's very little speculation about how it's affecting performance.

So, a bit of a thought experiment. The power system could be set up as an isomorphic graph (see: graph theory) or it could be a cell-by-cell simulation that the game has to solve every.single.tick. If you aren't seeing the implications here, the latter would be far more memory-intensive than necessary. Maybe it's set up in some other way that I haven't thought of, but those are the two things I can most easily imagine.

There's one variable I never looked at among all my bases that have terrible FPS. I always built a huge looped heavi-watt backbone that crossed the entire asteroid. So this time I built small localized power grids with local fuel supplies & small local power grids. I also resisted the urge to industrialize, so my base is far less power hungry and much more focused on stone-ax labor than usual, this time.

I've discovered 80% of my planet.

This is an example of how my power grid looks. Small power plants that each service small areas.

With 80% of my planet discovered, and some pretty complex piping/vent systems, lots of automation, ranching about 80 critters, lots of debris on the ground, having dug up lots of areas, I'm getting 45 FPS looking at un-built areas of the map, ~37FPS looking at my main base, and about 33FPS staring right at my ranch. My last base was sitting on less than 20 FPS before I gave up on it.

Power systems might be the game's current bottleneck, but it's a hunch with anecdotal evidence at best. Please feel free to experiment and share your results.

 

[selereth]

What are your system specs? Most important is the CPU.

[avc15]

It is most important to see how different systems react to having the power system completely deleted from the map. My computer is pretty powerful but one base <20FPS with a huge powersystem vs this one with a small power system at ~35-45.

My CPU is irrelevant to this discussion except to say I have 6 physical cores. And then comparing one result to the other.

Would like to see other people's results. It's a viable test to just load up a complex base and delete all the wires in your base, then see if frames went up.

 

[flapee]

you seem to be onto something !!!

21 minutes ago, avc15 said:

delete all the wires

or better yet break up the one network into multiple separate ones; you still want to keep the power entities "ticking"

[nakomaru]

39 minutes ago, avc15 said:

The power system could be set up as an isomorphic graph (see: graph theory) or it could be a cell-by-cell simulation that the game has to solve every.single.tick.

The common understanding is that it is the former. That is, wire lengths do not matter. A single circuit is always computed as one unit, with a list batteries, generators, consumers, and maximum wattage. This is probably why order, positioning and individual wattage do not matter when determining overload damage.

[mathmanican]

19 minutes ago, nakomaru said:

This is probably why order, positioning and individual wattage do not matter when determining overload damage.

Positioning however does matter when you don't have enough power. Things further out don't get power while things closer to the transformer do. This means they do have some algorithm for determining "closeness" in the grid.

[nakomaru]

I didn't realize that. But generally you do see 5 transformers discharging in order which agrees with your statement. I thought this was tied to an arbitrary listing such as build order. I'll be sure to pay close attention now.

Still, the ordinality need only be determined once per circuit, and then only use that from then on, which I believe it does.

[Lifegrow]

I honestly don't think power grids are anything to do with FPS issues, or late game overlay/interact delay.

I have automation on every transformer I place, always the power is routed via a circuit battery - so the battery delivers the power rather than the transformer. As such, there's never more than a handful of transformers active at the same time - even in a huge base.

I'd love someone to prove me wrong though and fix my late-game slog