Why trains? (UPS focus)

[blazespinnaker]

This is for UPS minded folks.

I just did a quick test of 10K blue belts fully loaded and the transport lines in time usage hovers around 0.06-0.09.

Entity update is at 0.45, but that could be more related to other aspects of the build (I load/ unload the belts, and that's a lot of ore.)

It seems to me that if you just did all belts directly from miner -> furnace -> assemblers -> labs, you could avoid a lot of unnecessary inserters.

[jodokus31]

I would propose to change the topic name to something with UPS. This one is really misleading IMO

[eradicator]

10K blue belts

Maybe you should test with more realistic numbers. Maybe 500k or something. That is if you're talking about replacing all trains in a megabase that pulls in resources from 5000+ tiles radius. Also fully compressed belts perform differently from belts with even just a few gaps.

But i think the conclusion will just be: Because people like trains and comparing them is difficult.

[blazespinnaker]

Why pull them into a central base? Why not just do science at satellite bases? Laying down a full factory actually isn't that hard in terms of robot usage.

Agreed, trains are fun. So is nuclear, radar, roboports, splittters, etc etc. Most of the fun things take up UPS.

I'm just interested in seeing how optimized I can make my base, UPS wise.

[eradicator]

Why pull them into a central base? Why not just do science at satellite bases? Laying down a full factory actually isn't that hard in terms of robot usage.

Agreed, trains are fun. So is nuclear, radar, roboports, splittters, etc etc. Most of the fun things take up UPS.

a) I never said anything about "central", but i guess usually because it's easier to design? b) Cos ore fields don't last forever? c) So... what?

I was just saying that "I tested 10k belts" is like saying "i tested 500 assembers". The number is just waaay too low, and you didn't even post a savegame or pictures of the test setup, nor hardware statistics nor anything. If you wanna discuss UPS you gotta do the numbers.

I'm just interested in seeing how optimized I can make my base, UPS wise.

Looking forward to your the results. What size were you planning for? Are you basing the design on any particular known megafactory? What kind of ore distribution settings do you use? Etcpp...

Do you even know about direct-insertion train bases?

[blazespinnaker]

10K belts is a pretty good number. On average (standard vanilla), I think patches are about 1K apart, and if they aren't, you can go find something else. You find a point equidistant between the patches, and you can load quite a lot with 10K belts. You can do interesting optimizations eg: prebuild items at patches ( steel, maybe engines).

I suppose, in theory, you could do direct insertion onto trains from furnaces and then into assembers (or any config) but I fail to see how that could possibly be UPS efficient. Belts are very JIT, cargo wagons, ugh.

As for base design, it's really evolving. I experiment with something new at every satellite base. In general, I think that's a non issue when it comes to UPS, and is more about microoptimization. Maybe bots are the way to go, I dunno.

Architecturally, I think the key question is how best to get resources between the patches. The reason I say that as it plays largely into the size and placement of bases.

There is a lot of focus on trains, and I get it, they are hella fun and facilitate large gigantic centralized bases, but I'm just curious from a UPS POV.

[eradicator]

10K belts is a pretty good number.

Hate to say this, but: Pic or it didn't happen. (More like savegame actually)

I've started a new experimental ore-belts-only base that is already over 10k belts, and i've only just started building blue science.

Edit:
Also the whole discussion is moot if you don't state the factory size you're talking about. 1,5,10,20 kSPM?

[Stevetrov]

It seems to me that if you just did all belts directly from miner -> furnace -> assemblers -> labs, you could avoid a lot of unnecessary inserters.

Your conclusion is correct, in fact using machine -> belt -> train -> belt -> machine is one of the worst design patterns for a megabase (upswise) but it does look cool!

You can also just use trains and di into the machine which in some situations is even better upswise.

[Fooluaintblack]

Blaze, if you want to dig into the rabbit hole take a look at Swolar's megabase. I don't know of anything that's more thoroughly optimised than this.

[Zavian]

I suspect (but haven't tested) that inserters grabbing items from belts is significantly more expensive in cpu cycles than picking up items from a chest/wagon/assembler. Hence a well designed train base is likely to be more cpu efficient than a belt base.

[Stevetrov]

I suspect (but haven't tested) that inserters grabbing items from belts is significantly more expensive in cpu cycles than picking up items from a chest/wagon/assembler. Hence a well designed train base is likely to be more cpu efficient than a belt base.

Inserters picking up of belts are worse (ups wise) than from chests. But train tracks take up space that means u can't fit as many beacons without using more inserters, so you end up with a trade off.

[blazespinnaker]

10K belts is a pretty good number.

Hate to say this, but: Pic or it didn't happen. (More like savegame actually)

I've started a new experimental ore-belts-only base that is already over 10k belts, and i've only just started building blue science.

Edit:
Also the whole discussion is moot if you don't state the factory size you're talking about. 1,5,10,20 kSPM?

Nah, I'm just in it for the math and perf analysis. We all play the game differently I guess

I have a pretty huge map with a lot of random and different experiments going on. Just a few 100 rockets launched tho. Don't really get to play as much as I'd like.

[blazespinnaker]

I suspect (but haven't tested) that inserters grabbing items from belts is significantly more expensive in cpu cycles than picking up items from a chest/wagon/assembler. Hence a well designed train base is likely to be more cpu efficient than a belt base.

Inserters picking up of belts are worse (ups wise) than from chests. But train tracks take up space that means u can't fit as many beacons without using more inserters, so you end up with a trade off.

That's an interesting statement. Mulark has a great site with experiments on stuff like this, needs updating for 1.0 though I guess

https://mulark.github.io/tests/test-000 ... 00011.html

I looked at swolars stuff. Very smart dude. Train/bot based tho.

We really need a belt that either directly feeds into assemblers like bots or a belt that you can do assembly on. I like belts

It'd be nice to see some optimizations for trains too. maybe mark v2 wagons that can pull directly from chests.

It'd be cool if wube created something so we can scale up our bases more without killing UPS too much. Something more meta. I like playing the game in map mode.

[Bilka]

I looked at swolars stuff. Very smart dude. Train/bot based tho.

What? The base linked by Fooluaintblack is belt/bot/train, not train/bot.

[jodokus31]

It so much depends on resource settings and reasonable building time (which is an important factor for me at least).

f.e: Rail worlds with spread out resources needs a lot of belts to bring the right amounts to a decentral "satellite base"
This costs a lot of building-time and more resources (blue belts) to build. Once it's done it might be better UPS-wise than trains. Important is, that the belts are fully compressed and have long stretches without splitters.

Otherwise, if you use very rich and frequent resource setting, belts are probably just better.

Next step, maybe more radical, why even use resource patches, if you could use infinity chest for ore supply from sandbox.

[jodokus31]

It'd be cool if wube created something so we can scale up our bases more without killing UPS too much. Something more meta. I like playing the game in map mode.

Exponential growth means double then amount of production in a certain timespan. If you would optimize the game to double performance, then you can grow a very short time range to reach that limit again.

[eradicator]

I suspect (but haven't tested) that inserters grabbing items from belts is significantly more expensive in cpu cycles than picking up items from a chest/wagon/assembler.

That also depends on the size of the inventory (-limit). Small inventories are more efficient. That's why certain megabase optimizations use wooden chests limited to a single slot.

We really need a belt that either directly feeds into assemblers like bots

Ok, i'm out. You're living in a fantasy world that i don't know. Bot -> Chest -> Inserter -> Assembler is the shortest you get, and that's actually two inventory interactions more than Belt -> Inserter -> Assembler.

[xeneonic]

Would it help if trains had 1/4th their current slots and increase stack sizes by x8 to put UPS more in favor for using trains? Or other alterations such as the bulk rail loader (1 entity instead of many inserters withdrawing?).

What are the optimizations one could do?

[Zavian]

I suspect (but haven't tested) that inserters grabbing items from belts is significantly more expensive in cpu cycles than picking up items from a chest/wagon/assembler. Hence a well designed train base is likely to be more cpu efficient than a belt base.

Inserters picking up of belts are worse (ups wise) than from chests. But train tracks take up space that means u can't fit as many beacons without using more inserters, so you end up with a trade off.

Most design decisions involve trade offs.
With a train setup you can get an 8 beacon arrangement with 2 less inserters than the equivalent belt based arrangement.
You can get rid of the long-handed inserter by using a car.
You can get 10 beacon arrangements with either an extra chest and stack inserter, or with a car. (No idea which is better. Using a car does limit your options to circuit control things. So another trade off).

train.beacons.setups.png (3.35 MiB) Viewed 6566 times

So yeah these are less beacons than the max 12 beacon arrangement that belts allow, and none of these designs facilitate chaining multiple steps using direct insertion, but all of these designs use less inserters than a train->chest->belt->assembler->belt->chest->train setup.

12 beacon setups are also possible, but some of the cargo wagons can't be used. (You can replace the unused wagons with locomotives).

[blazespinnaker]

Ok, i'm out. You're living in a fantasy world that i don't know. Bot -> Chest -> Inserter -> Assembler is the shortest you get, and that's actually two inventory interactions more than Belt -> Inserter -> Assembler.

No man, just google RL assembly lines. Inserters are the exception, not the norm