Quality and Speed: How much waste does haste make?

Post all other topics which do not belong to any other category.
Post Reply
Copy link
coffee-factorio
Inserter
Inserter
Posts: 40
Joined: Thu Oct 17, 2024 10:56 pm
Contact:
Contact coffee-factorio

Re: Quality and Speed: How much waste does haste make?

Post by coffee-factorio »

Right, the comedy of the day.

I was talking about this with a coworker.
He is a lay carpenter; but a damn good one from what I can judge.
What he said was simply this "You're talking about machines that cost thousands, why are we having this meeting when I can put guys to work."

My response then, as now, is that I can put the right guys in the field faster and they'll to work faster. He quite liked it.

In our consideration we have looked at scenarios where legendary units are available.
The lifetime of a fair game of factorio we will have a situation where we have thousands of lay workers but none of a given quality. We have an egg.

So the next examples where I remove the legendary quality from the entire example.
I believe this will show Kovarex intentions for vertical scaling where realized and that it is more than a simple gambling mechanic.

It is appropriate because the drills I used on previous examples to hit a rate of 69-73 ips always will. That reward never changes. Resource drain does, it scales the time dimension a drill can capture coal. I've seen productivity used for this in Nullius (god bless those big drills) but resource drain is more nuanced because it doesn't change my rate of return.

So my input conditions of item/second are all valid, and my technique is to scale horizontally to capture output with the means available.

This shows how much of a speed mallus we have manipulating a/(1-r) by reducing quality. Manipulations in a aren't always so great. That's why a speed preference can cause hurt with plastic. In other models it's different but this will augment our ability to manipulate those.

Toying with a/(1-.5) shows that we get 2a. 1/(1-.6), will be (1.66)a. 6/6 vs 5/6, or an 18% decrease caused by a .10 shift. Quality module increases work by increasing a rate linearly in this manner. So it's something like a/(1-.1) v. a/(1-.2).

It's easy to see what happens if you use simple fractions to render that. a/(9/10) = a10/9. a/(8/10)= a10/8. Instead of a 1/9 increase we get a 2/8 one.

But I expect a punishing reduction across all builds and bigger than might first be suspected. The trip from a +585% building to a + 110% building is from 6.85 speed to 2.1 speed. So if speed is the only thing considered we expect a 1/3 drop in our builds (and might even forego recycling as a technique).

But in modeling we have at least the suggestion that our mallus will be more like going from a 10*(input)/8 to a 10*(input)/9 bonus. It will punish recycling harder than speed though
drop all output from that source.

So I start with a speed 3 build as my basis input. Speed 1 will increase quality by fractions of a percent here as well, but the jump is not considerable in the recyclers which are so
fast that for the same reason I use a normal assembler for my pipes I just horizontally scale
those.

There is another reason that a view of the test will show.

Image

Basically if I go from speed 3's to speed 3's I'm going scale this layout so much I won't be able to fit it in a screen shot.

And we see the drop was not a 1/4 performance cut but nearly a 1/10 performance cut relative to the legendary build. This on its own shows vertical scaling.

How does an optimized build hold up?

Image

Well relative to the one that I quoted above and showed earlier we went from 60 ips to 8. So the drop is consistently staggering, awful I am thinking is the word?

But relative speeding up input we're going faster: 3 and 8 are two different things. More practically in relation to waste we are directly making more parts in apples:apples comparison. As far as technical difficulty the reader will note the vertical scaling down didn't change my build size.

Blueprint is here in case I missed a module setting it up.



I do believe that is a cluster of five 4 tiles, so a speed based method benefits from exponential scaling by a factor of about six in terms of area used.

Well, blueprint in case I miss count.



It is larger on screen than in the picture.

I cannot simply rest on this. Just because I have a hammer not everything is a nail. To scale to 1 ips to 5 I need to scale my entire drill line when I have legendaries and that costs material. So there's two more test cases to consider at this scale to show how I might cut that cost down if it's possible at all.
Top
ichaleynbin
Inserter
Inserter
Posts: 25
Joined: Wed Feb 02, 2022 2:55 am
Contact:
Contact ichaleynbin

Re: Quality and Speed: How much waste does haste make?

Post by ichaleynbin »

coffee-factorio wrote: ↑Sat Dec 14, 2024 10:20 pm
I do believe that is a cluster of five 4 tiles, so a speed based method benefits from exponential scaling by a factor of about six in terms of area used.
Wild, that also represents investment cost to a degree. Startup time on a base even just 1/6th the size before compounding is pretty significant, but this is probably a plastic specific number? Qual modules having an explicit speed malus makes the first touch of speed beacon that much stronger, that's probably why cryos are taking such a massive bonus.
Top
coffee-factorio
Inserter
Inserter
Posts: 40
Joined: Thu Oct 17, 2024 10:56 pm
Contact:
Contact coffee-factorio

Re: Quality and Speed: How much waste does haste make?

Post by coffee-factorio »

Oh, we have only begun to explore this rabbit hole. There's a dragon down it.

As I did a maybe more effective build I developed a second one very quickly due to power constraints. The character of optimization changes as you play Factorio even in 1.x. A beacons square is great after you hit white science. But you basically relive the industrial revolution if you get some green modules and double your manufacturing capacity by effecting a brute force forge that's somewhere between 5/3 and 5 times the size. If you try that on a mega base you'll miss because UPS issues. Before that add green modules to everything and watch your pollution size not shrink while your iron production goes up an order of magnitude.

This is the issue ichelynban raised about iron. If you don't catch all of it, in the land of items:items this is called lacking ambition. And the iron ones I've posted earlier in the week where me on a work night so they do. You need a machine capable of capturing the input and that takes power. If you're in the late game power isn't an issue. If you're in the early game you get this as a hint. I got this hint by going full prod on plastic.

Image

Now the result is surprising. My old two building build is twice as fast as this one. And it uses the same drills; needs them in fact to start the quality chain. So it's the worst of all possible worlds, basically it's maker looked at it and already went "well I can do better" and then decided to cannabalize it for parts based off rate. And power was the hint- that beacon lane pushes power consumption to 17 MW a building or more.

Not figuratively cannibalizing either: As far as difficulty and scale this one is like touching grass in Nullius. Unlike Nullius where we have a packaging recipe (typically 5 items) on a 75 ips belt, you have to restack items onto 30 and 60 ips belts. Willy Wanka arms I think I'll call this. Yes. This was the result. It is blessed because it will make the build work period. You have to restack to hit i/o on the red belt.

And unlike in Nullius our worms, they do not eat the grass to support a beautiful food chain; and the grass cannot be composted for methane and more importantly the good brown landfill at scales that let me make hundreds of grass plants (edit: facilities).

It was too funny not to complete though; I mean the power use on this thing is off the chain if you imagine someone trying to implement it without legendary stuff. 17MW per building. It's worse than even naive speed and just because it's my work doesn't mean I won't happily show to all of you for edification before I find some sheds to finish eating it in peace.

This is where ichylenban is very right about productivity. (2)a/(1-r) gets you the same relative size of build in terms of building size. But not pushing r as high as it will go hits you with an exponential stick. OP's concern about seeing speed modules pop up continues to be borne out in practical examination; even when different routes to a line are attempted. Whatever else I can say about the speed strategy at least it scales at all.

Now but if you look at the model there's a way where we can optimize power and cut down complexity. And the two building build suggests it.

In that build the important difference is that plastic is benefited not through productivity modules but rather through quality modules. I mean if I change my module strategy to that on every building here, I get purple from that strategy by extension. The quality buildings all take a 5% penalty to speed only. But the modules themselves don't effect power so there's a much less steep power penalty. And it will go from ~7 ips to 12 ips when I Monte-Carlo it.

Here's the thing. It does it a 1 building larger footprint on cryloabs. For the nearly the same beacon strategy. Power is dropped from 17MW to 7. Beacons are moved off purple+blue benefaction and there's like one serving a line of 3 green benefaction cryolabs. But the most important cost cut is that since you make half as many bad things you need 4 recycler tiles instead of 8. The yellow on the orange chips always mean use with caution.

What is referred to internally in my tests as "2" and "8" depending on if has legendary is looking at all of this with a raised eyebrow, it got the recycling problem down to 2 recycler tiles of non-legendary stuff I believe. That's the 2 building build washing on legendary.

Discussing it adds a tool to our toolbox: speed bonuses through material selection. But speed beacons re-enter. You lose the bottom section of recyclers on the build because you don't double your production volume of bad stuff. Solving the dragon was still worthwhile because it allowed rapid development.



The quality one's blueprint is here. The whereabouts of the productivity one are unknown. I think I may have rotated a belt because I was moving on and decided to just grab it for posterity. (edit: it is fact semi-torn apart right at the infeed to normal benefaction)

Can we do better? The path to a brute force solution we have is going to green modules 2 and stretching the assembly line to 30-70 long like I where building early game furnaces. Worth mentioning? Yes. Worth testing. Look that coal patch can take 4 more ranks of drills. But then that needs a train line to what will be 120 to 280 cryolabs that I have to ship from Aquilo. It isn't a "that's the only way thing". It's a spectrum of not great tradeoffs from blue module use to green modules 2 with a building count attached; the build pictured is just when I tune everything to blue3 modules on that spectrum.

Judicious optimization means writing off the normal coal and benefiting only green+ quality. This coal patch was based off the size of a 4 mil' I found in the wild. It was 32x32. If I benefit 1/10th of that output fully with green cryolabs the at don't get what I want (halving the number of modules needed to expand miners horizontally) but I'll bet close enough to say that it's probably the most practical set of tradeoffs and go for further innovation elsewhere.

And if I expand horizontally with 4-5 more ranks of drills or whatever then that's 28-35 green processors. The blue purple orange step only grow when you hit input rates that are bound by 1/100 and worse decreases in volume. A circuit junky might be tempted to do something that will blow their foot off and do not do it.

That is wire a circuit to switch recipes. Every time you switch recipes the productivity track resets. If that changes I'll happily edit that statement. I tested that one in my actual playthrough though when I decided to FAFO. And boy I found out.

Anyways. The Judicious build we'll call 12.

Image

This one is power optimized too. It hits 11 ipm in initial investment off the 69 line. The linear solver probably went for blue+ cycling because over time that might be the best tradeoff it sees from the data it has; it might be interpreting uncommons as a trash stream and that gets into the murky area where you can take your fingers off. Just doing purple seemed like it provided the most benefit. I can say productivity won't do better than quality their because it's a straight shot up to legendary while productivity their is a straight shot into the recycler (mark of my ADHD I might never have made the build on this page if I took the time to think about why my third one worked).

ichylenban made a post rejecting the input size that was premature, but stacking does make a 69 input look like a 24 input and that bit me. If you stack to a full sixty and then try and horizontally scale you might need so many green buildings you end up with the power build in terms of scale, even doing quality. That's one short of the more greedy approach I discussed above but this isn't about optimal but rather practical; which is why I have a picture and a partially finished blueprint of those. If it scales proportionally when quality increases that means we have to go for something like 36 instead of 48 drills to scale the final product.

I can Monte Carlo faster than I can model with math so a real numbers person will just beat me on the prediction count and might have a thing or two to say to educate me on the logic count. I'm not the shiny kind. But this makes me smile.

Top
coffee-factorio
Inserter
Inserter
Posts: 40
Joined: Thu Oct 17, 2024 10:56 pm
Contact:
Contact coffee-factorio

Re: Quality and Speed: How much waste does haste make?

Post by coffee-factorio »

So you know what I was saying about how scaling can take your fingers off?

So I took this to legendary modules. Since legendary modules are twice as effective as the normal ones at making next tier quality; population of uncommons doubles or more. My reaction was to slot in legendary cryolabs. Figured out that those make so many legendary uncommon chips that you need to stack them.

Image

This all for the same input size. If I don't sound happy it's because one should feel very cautious when leave a world of technical simplicity and don't get double the output.

In between "2-8" and "13" the thing to consider is that "2-8" is 60 ipm and "13" is 80 ipm. 80:60 - 4:3. Speed is world "1". "1" is valid to catch a line of inputs because once a beacon hits a drill line then I fully expect 240+ ips. We can say trivially that by slowing down we do 4 times more; but my 4 times more is scaled to red stacked ones in 13 and I have to hit it more to get it to catch more trash. If you don't catch 240 IPS what you can say about material waste has to compared to seconds of your life you won't get back. The fact the number is 76:6(something) just makes this more true.

But as brutally true as that is it's very realistic to say that if you don't manage proportionally small garbage stream it'll its the same. So there's going to be a gradient between "2-8" and "13" with "1" sprinkled in. "2-8" triples speed on 1 instead of multiplies by four - but you can put legendary beacons and modules on back end.

The thing about the drills and recyclers is that they want different when you think about the situation.

Not hitting the back end with world "2-8" at minimum costs you 2 legendary parts produced for the same input size of coal and time. This is why I fixed input sizes, at thousands of parts that's like half my smallest practical space frame. Give it a few seconds and you pay for one that prints space platform.

So if you really want to have a big population instead of rolling 1 on your processor back end you roll speed on your quality front end. So you have a back end that multiplies by 3-4 by playing in the "2-8" to "13" world but a front end production facility that's straight speed and green modules. In between you have lanes of processors that just start out 5 times as fast and reduce everything by 4 times as much. And they stack. They seem like they want to be legendary so they do 10 a second while you get as much quality as you can. If you see speed 1's, I mean the work on the back end was one thing. But for the same reason I criticized pipes at some point you'll see speed modules one hitting drills and recyclers. Drills because if you two big boys that can end up being another buffer line of material pretty quick. Recyclers because your waste stream went from "fun" to "fanfunningtastic" doing that.

Somebody mentioned something about not even mixing quality and productivity. In fact that's a great idea. Do that. Don't mix them: 8/10 * 1.25 is basically what you're looking at over time. The number doesn't get better with legendary.

Productivity brute force is a trap. Quality brute force is actually a trap that's just as bad. Your power footprint is going to grow so big that horizontal scaling is going to be rough. If the yellow lights mean caution, red ones mean stop and think. The more you play towards "13" the more you play towards 1.

Hypothetically if someone hands you an unlimited budget and the building at the end of the production chain: with best prod mods you double it in free plastic, then you go for that and just stack beacons on it till its a big happy square that beats anything because it is multiplying a giant speed bonus time a giant productivity bonus to produce results. And when I have 124 ips legendary coal that's what I'd use. But getting the loot requires I play around.

The only other option for plastic is a blursed line that does multiple free selection passes. We can't do that here edit (without pentapods). But we get the same effect in several ways on iron and because it isn't made with tentacles it must be pure.
Top
ichaleynbin
Inserter
Inserter
Posts: 25
Joined: Wed Feb 02, 2022 2:55 am
Contact:
Contact ichaleynbin

Re: Quality and Speed: How much waste does haste make?

Post by ichaleynbin »

coffee-factorio wrote: ↑Sun Dec 15, 2024 5:25 am So you know what I was saying about how scaling can take your fingers off?

...

Productivity brute force is a trap. Quality brute force is actually a trap that's just as bad. Your power footprint is going to grow so big that horizontal scaling is going to be rough. If the yellow lights mean caution, red ones mean stop and think. The more you play towards "13" the more you play towards 1.
Yeah, none of these relationships exist in isolation, is the thing. If you bundle in amortization cost of your investment on the "just build more machines" then I suspect it's closer to a week than a day to pay back the investment, and as you said, what you even optimize for changes over the game. If you look at any one conversion rate, yes, of course speed is less effective on a per item count, I knew I was signing up for waste when I said it was still good anyhow. If it's for a mall, and I get my factory built faster because the cost of the factory is less to produce more items/sec, then that snowballs pretty significantly. Especially in the mall stage, I think at that stage of the game if the idea is early game quality, I'm only measuring i/s/machine out. That FIRST legendary foundry is hard to come by but the faster you print them the faster your factory grows, and it's an optimization over multiple fronts important at that time to focus on i/s/machine.

You can print at least legendary copper, iron, steel, and plastic, without losses. 300% prod cycle on Blue Chips and LDS causes the quality chain to grow linearly, not exponentially. We can argue about how best to optimize those exponentials, because before the prod techs finish up they're exponential too, but they still win over other methods on several metrics long before maximum prod.

I think if you're trying to optimize i/s/machine the easy math to look at is to just maximize quality/second. 25% quality at 100/s is 25 quality/second. 10% quality at 1000/s is 100 quality/sec. Both of them are 10% to next, so 2.5 vs 10 jumping two tiers, etc. The closed form math is a bit messy, I might work on the closed form of the speed equation today, it should be a quadratic with a maximum, the hard part is normalizing the speed input variable.
Top
coffee-factorio
Inserter
Inserter
Posts: 40
Joined: Thu Oct 17, 2024 10:56 pm
Contact:
Contact coffee-factorio

Re: Quality and Speed: How much waste does haste make?

Post by coffee-factorio »

Yeah, but we're given a choice between at least two modules at the furthest endpoint. Research is a constructive example because you can put quality modules in it at the furthest endpoint; which is laboratories-rather than productivity. You can also use speed modules there.

Pipes is too because you can put quality modules in it but not productivity, as you pointed out. If we're going for max of best quality (making legendaries) and given countably infinite resources to scale horizontally and vertically with; from a shallow standpoint quality and productivity gives us c*a while speed gives as a. And then we can bicker about who has or is the bigger a as a technical discussion. Pipes set C to 1 with productivity but quality is an option.

As in research, and plastic, putting it on the furthest endpoint is likely not sound. Truthfully it's insane. But someone can see free stuff and compare it productivity in that light.

A guy like Godel will take my a, and then your a, and build a rectangle. Then he'll cut it in half to get a better factory by taking the longest line left. He also has the option of just using any 2 of our designs if his talent matches his ambitions and go for the full rectangle. He'll look at insane guy and quickly look elsewhere.

If I leave it at that. I invite misfortune on my house.

When I say benefaction, I mean like ore benefaction. I never heard the word until I researched the background for Seablock to see where these recipes where coming from. But by taking off the good chunks from the garbage and washing it we add a population so (your model for quality + recycling output)*(productivity).

And at the frost inducing rates of return * items per second we end up getting back double from the same input when practically tested (and if a mathematician can model that; that's what I need at this point to demonstrate the observation fully true. But Monte Carlo is a tablesaw for just showing what the rate is and I'm not building my space-pig fleet out while doing this).

A player with similar background will look at the gem sorting line and see directly what I mean. Because it's like switching back to electrolyzers for raw mineral product.

So basically, you're following along with the logic of my argument and methods taking your waste stream at the stage transition from (any_iron->pipes->recycle) iron instead of straight (iron->iron) recycling might do better. This "slows the process down" but any gold is a gold so why not check with Monte Carlo.

Not presume. Check. None of these relations existed in isolation for the nuclear bomb. Monte Carlo was brought in specifically because far better mathematicians have had far more relevant issues to solve in a timely fashion.

And then you take the resources saved and invest in a much bigger factory. That is demonstratively practical because with technical testing on plastic; I took the option that intuition suggested in general that a linear solver suggested as best practice. In checking the tools value is borne out by adding the other streams in practice (worlds "12-13"). 4:3 output for 5x buildings and the modules invested in that.

Have a bit of patience and I'll do you some credit.

This is because the front end, based off the early designs of the week, is the fluid of a foundry (edit:for iron). But because it's inputs are so attractive, the high-water mark strawman with which we use to compare validity is going to be what we do off electric forges with "2-8" applied and I fully expect to show examples when haste is beneficial because of that.

Or OP's concerns will continue to be valid! From a research standpoint I can't lose.

It is also true that I have an expanding pouch of hammers and there are oh so many nails with which to test them on. Out of curiosity, can anyone see why going ( gears:iron) on a forge is a situation where haste is seductive when making legendary iron? And what the point of folly for that step might be with the modules available?
Top
ichaleynbin
Inserter
Inserter
Posts: 25
Joined: Wed Feb 02, 2022 2:55 am
Contact:
Contact ichaleynbin

Re: Quality and Speed: How much waste does haste make?

Post by ichaleynbin »

coffee-factorio wrote: ↑Sun Dec 15, 2024 5:23 pm From a research standpoint I can't lose.

It is also true that I have an expanding pouch of hammers and there are oh so many nails with which to test them on. Out of curiosity, can anyone see why going ( gears:iron) on a forge is a situation where haste is seductive when making legendary iron? And what the point of folly for that step might be with the modules available?
I've learned a lot too, almost sure I'm looking at the right stuff now.

The problem with those steps is how big the bottom end balloons. You can get your high end output, but it ends up being hundreds to thousands of machines. Haste drops machine count significantly, but then the i/s count starts to get legitimately problematic for any non trivial output. The question I'm stuck at right now is "Is it really blue chips for iron?"
Iron Chart.png
Iron Chart.png (724.02 KiB) Viewed 328 times
Linear beats exponential, I get that. Two dozen machines instead of two gross, and better inputs. I feel like I'm missing something still though, beyond "use 300% recipes for everything you can"
Top
coffee-factorio
Inserter
Inserter
Posts: 40
Joined: Thu Oct 17, 2024 10:56 pm
Contact:
Contact coffee-factorio

Re: Quality and Speed: How much waste does haste make?

Post by coffee-factorio »

Stack compression.
Top
coffee-factorio
Inserter
Inserter
Posts: 40
Joined: Thu Oct 17, 2024 10:56 pm
Contact:
Contact coffee-factorio

Re: Quality and Speed: How much waste does haste make?

Post by coffee-factorio »

The ultimate result of iron was to use exponential scaling to bring a normal quality plant to half output of an an unambitious legendary build while taking the number of forges from 34 in a contemporary example all the way down to 19. Other size characteristics inconclusive. However it is a bit long. The relevant rates are ~10ips legendary, ~5ips stack compression+selection, ~2ips speed only. ~1.7 ips selection+productivity without forges and stack compression.

Op your concern continues to be valid. As promised I delivered an example of how haste does not make waste. However my tale is a bit long so for the sake of brevity I report here.
Top
coffee-factorio
Inserter
Inserter
Posts: 40
Joined: Thu Oct 17, 2024 10:56 pm
Contact:
Contact coffee-factorio

Re: Quality and Speed: How much waste does haste make?

Post by coffee-factorio »

When scaled to equivalent rates it seems like we are world where linear scaling beats exponential.

Image

This is 1.6 items per minute. "2-8" plastic logic applied to iron.

Image

This is 2.1 items per minute. Haste in 2 ways: max productivity at point ore, which is not drills but fluid production. Just a plain faster machine. And at a long 34 furnace line to start with, I need it.

Speed 3 where speed used because I run it off the back of space platform. More quality engines double platform area for same speed. I can already print 3.6k space platform hour. Then add leaf to tree by adding second of engines. I already do fueling for one. Consumption doubles but I do not need cargo building , and area is about the same so I can go to 4x area.

Both monsters reside in the world of "12-13" in terms of discussion. Plans for space platform demonstrate exponential scaling even using uncommon quality engines and either of these can catch it at rates where we need to scale 30. Plastic did a lot better because we have more slots. Reality is more cruel still: that was on a planet where you get the research for legendary quality. So your rate of return is going to be like... maybe just go blue instead?

Gears optimizes by using a recipe that is both faster and more productive: 2 iron worth objects for 10 fluid instead of 20 in 1/4 time. We shorten the forge line but our poor recyclers have to handle... ah, stack compression. Each gear is 2 plates so it turns out to be the same. So it's a 200 ips line on a 240 belt. That due to exponential scaling doubles the volume of fluid we handle.

We can benefit both by adding a benefaction step. As stated iron->pipes->iron such a step. What you do by adding quality modules at that stage prior to feeding it into a recycling machine is apply another pass of quality and at a rate of 0.5 ips per pipe, fast enough in an assembler 3 for our applications. If we get a legendary pipe we are punished of course - the pipe is divided by four. But this is like, 2% quality instead of 10% and will ultimately be instructive in demonstrating how a minor difference results massive results. I ended up filtering on purple and the thing went marginally faster even though I was crushing legendary stuff.

And it was still because it was a productivity bonus, so I can basically write that off. Took it from 5 ipm to 7 ipm.

Not capturing gears as a chain lacked ambition. I will leave it as a technical excercise to determine whether or not going for the gears w/out speed 3 on a 34 line can boost this. It's worth it: I have to scale 2.1 ips by 30 and that just doesn't work.
I have a day job.

I didn't let up on speed 3 modules.

Recyclers where unevenly stacking. In Nullius I touched grass.
Basically what that means is I make 100+ ips out of positive feedback loop in a 24x24 area and scale horizontally, to get 3 boxes of landfill at a good rate. Each grass takes 2 ips, each box of landfill is 10, and can hit 75 instead of 60.

So when I saw the lines on 200 deadlocking I knew from experience that I needed a bigger leaf on my exponential scaling tree and replaced a lot of recyclers with stackers because they where being deadlocked by their throughput. This replicates the packing recipes and units that allow you to put 127 ips on like a red belt.

And touch trees. Which use grass and tan-fill as a dependency - that you resolve by touching worms. Trees go faster than grass. So you learn the benefits of exponential scaling at those rates.

I had to recreate the effect of proper 4x4 boxes by having the Willi Wonka inserters repackage on the recyclers. That cut recyclers to 18 but you need stack inserters so a trip to Gleba is in your future. Regardless of how I dislike the place the sushi belts here are all derived from nutrition lines. So the world of naive speed and the huge recycle network may be in error. The 34 forges are real though.

This is all handled as 2 steps of pitiful selection. It works because exponential scaling:
10 iron is 1/2 a plate. 1 gear is 2 iron. So in the world of ips my rate of return at this stage is 1:1 because of productivity if I did my math right.

The we use stack compression.

This carefully hinted at in the tutorial for iron I believe. As I was working, I recalled (and the jokes about ADHD are not jokes, so this is nothing short of miracle) that I had
seen gears being fed into inserters before. We can do better by adopting a sledghammer:
body armor.

But why bother? Just smash the thing into a box? After all, not every nail needs big hammer and the armor recipe is slower by the numbers: a box 8 ips feed every 1/2 second is 16 ips. Armor is 40/3 = ~13.

I had to FAFO and I can say that it's because exponential scaling.
Going at the appropriate rate with a sledgehammer let me build a leaf 40 big to place on a 60 ips rate. A box is only 8 big. It gets a quality selection pass because we can load 40 plates into the same selection. Naive reasoning says this might be a bad idea, but when we win we win 40 at a time.

Go for broke then!

OP, your concern continues to be born correct. By going a bit slower I effected a rather large increase in my throughput ceiling. 40*60 = a line that can handle 2400 ips. And if you can hit higher than you stack that through exponential scaling, so that in the same footprint of 60 ips you get 4*40*60 = 9600 ips.

So basically, what I'm saying is. If I put full beacons on a legendary beacon squared productivity based system. If I swapped out the quality pass for a productivity bonus and then used DI machines on body armor for iron plate, ASAP. I could do this for a hot minute. And the speed scaling means that I just don't need to care about the gears thing.

But return to the idea of blursed plastic: every pass through a biochamber can be a selection or a productivity scaling. A linear solver given that idea can look at it and ask for selection at every level without realizing that it toss a boat at cthulu by using productivity scaling the point of like, making carbon. That isn't actually as scary as it sounds because you just shred every old thing in the same machine. It's why you go sushi belt on recyclers: you make 1/4 output and they try to stack so you can really just run this round in circles if you need to. I think though, since they are so fast you might be able to just overstack the line by a few machines and get a better result from an input priority feed line.

The situation in forges is like blursed plastic in a way which is actually blessed, I think. I will leave it as a puzzle for the technician's delight to unravel one of several other recipes that achieve stack compression. And if you solve it you get a nice bonus of something extremely special to boot.

Fulgora is your hint as to how deep this rabbit hole goes. The selection of material can be made by passing a particular item normal through a selection process at a rate of 1-1000+ ips, at which it is made.

Image



Ah I under reported. That is 6 ips, in a build that simply doesn't care about input on the back end, with a front end that is half the size. And we cut our drill line down because fluid is the ore, not the ore. But a productivity scaling multiplies that for free and ichylanban pointed that out, so I took on the more interesting problem as my incorrect benchmark used pure speed.
Top
ichaleynbin
Inserter
Inserter
Posts: 25
Joined: Wed Feb 02, 2022 2:55 am
Contact:
Contact ichaleynbin

Re: Quality and Speed: How much waste does haste make?

Post by ichaleynbin »

coffee-factorio wrote: ↑Mon Dec 16, 2024 4:11 amI have a day job.

I didn't let up on speed 3 modules.

Recyclers where unevenly stacking. In Nullius I touched grass.

Going at the appropriate rate with a sledgehammer let me build a leaf 40 big to place on a 60 ips rate.
I'mma be honest coffee, your posts are getting more and more deranged, and less understandable. Is this an AI sockpuppet of Romayne's, or just ESL? I don't understand half of this last post. It's all english words but they're not being combined quite right.

There are quality, productivity, efficiency, and speed modules. I'm not aware of a "benefaction" module and I saw no evidence at all that OP's concerns over the materials themselves are valid. You haven't even actually responded to any of my posts as far as I can tell? I don't think that I've ever recommended using a speed 3 module, I did say one way of calculating speed for a maximum, which was ignored it seems?

The purpose of speed on quality builds is specifically to increase overall items/second/machine, where appropriate.
Onebeac.png
Onebeac.png (1.82 MiB) Viewed 243 times
The presence of this beacon takes me from crafting speed 8, to crafting speed 20.5, with 22.3% quality instead of 24.8. 0.4 items per second total and 0.0992 quality items on the unbeaconed version, versus 1.02 per second total and 0.227 quality items per second, +129% better. It would require an additional legendary foundry and then some to produce as many quality bump furnaces, making a beacon with legendary speed 1 a significantly lower investment cost to maximize my output. Literally min-maxing here. It's ahead now, and will always stay ahead if you invest less up front, "no speed" can never amortize unless it provides a higher rate per unit time to make up for the increased investment. If not, you'll always have more actual legendaries in hand with speed, once speed is ahead it may continue to reinvest and snowball the lead even farther.

Now, depending on how you're measuring this, if you compare 1 speed1 to no speed, no speed never wins. If you match machine count, nospeed always loses. If you match investment cost, it always loses. If you match output, it always loses. If you overbuild and calculate amortization time, nospeed falls behind the speeded snowball in the big picture. But if you measure via another way (items/input items which I still contend is not itself the important thing to evaluate), speed never wins. Probably both those ways of measuring provide limited information at best; None of these things are isolated and I think attempting to study them in isolation may not be informing us of as much as we would hope.
Top
coffee-factorio
Inserter
Inserter
Posts: 40
Joined: Thu Oct 17, 2024 10:56 pm
Contact:
Contact coffee-factorio

Re: Quality and Speed: How much waste does haste make?

Post by coffee-factorio »

coffee-factorio wrote: ↑Mon Dec 16, 2024 4:11 amI have a day job.

The purpose of speed on quality builds is specifically to increase overall items/second/machine, where appropriate.
Onebeac.png
The presence of this beacon takes me from crafting speed 8, to crafting speed 20.5, with 22.3% quality instead of 24.8. 0.4 items per second total and 0.0992 quality items on the unbeaconed version, versus 1.02 per second total and 0.227 quality items per second, +129% better. It would require an additional legendary foundry and then some to produce as many quality bump furnaces, making a beacon with legendary speed 1 a significantly lower investment cost to maximize my output. Literally min-maxing here. It's ahead now, and will always stay ahead if you invest less up front, "no speed" can never amortize unless it provides a higher rate per unit time to make up for the increased investment. If not, you'll always have more actual legendaries in hand with speed, once speed is ahead it may continue to reinvest and snowball the lead even farther.

Now, depending on how you're measuring this, if you compare 1 speed1 to no speed, no speed never wins. If you match machine count, nospeed always loses. If you match investment cost, it always loses. If you match output, it always loses. If you overbuild and calculate amortization time, nospeed falls behind the speeded snowball in the big picture. But if you measure via another way (items/input items which I still contend is not itself the important thing to evaluate), speed never wins. Probably both those ways of measuring provide limited information at best; None of these things are isolated and I think attempting to study them in isolation may not be informing us of as much as we would hope.
I've come down sick :\

And I'm in a bit of a spot. Because the first thing I'll say is: with the haste added by the stack compression step I noted, I can basically fill a mine with fully sped up drills so I basically have a better optimization at that point. Space platforms are where your overall "big numbers argument is given justice" but that's because of stack compression and the fact that you just lose the rocks if you don't use 'em. So I promised the man justice for his ideas and I think whatever else I'm going to say next; someone can look at that just read it as a fair shake.

I've been too busy building to much bother with the negative things you've said. Calling me crazy is one thing because I know I am. I've been playing 800 hours on a mod where a linear solver like Foreman will flat out lie to you and get you to piss away 4 gigawatts power on 5 steel. As far as my English, that isn't an issue on a forum with people from all over. So I thought I might critique your build with my normal flair.

My spot. Is that I noticed what you're saying about percents and then looked at what that does with simple back of napkin math in a moment. You made me just mad enough to do what any sane person should do before they try and build something - but this is a game right?

So the quality table states that the probability of getting a legendary in the best case is 0.1%. But that's applied down the entire line. So a legendary is 1/1000.
But I can scale those numbers to be 10/10_000. You multiply 10/10 right?

When you apply a speed module you go to 0.05. With normal quality that goes down to 5/10_000.

And if you divide those numbers 5/10_000 / 10/10_000. That's 1/2 as a simplified fraction. That's what you're getting.

At legendary it's better. Because it's ~20/10_000 instead of 25/10_000. So that's 4/5. edit: which means I need to change a number.

But ah, see the thing is. He has 1/2 the green quality ones too. So those contribute half as much. And they're ten times better than the normal ones even though there's ten times as few. On purple where 2-8 saw the most benefit. That's where this is real clear. Half as many will jump because half as many exist in the first place.

I've just been having too much fun to realize that you're asking people to lose over half to 1/4 their volume of legendary's.

And someone might think I'm being a bit unfair in just saying this until they realize what happens when someone hits volume on the electric drills pictured. Then you have to buy a second machine or move it onto the beacon the big drill is on . And you have to see a real profit. The beacon itself is four modules tier 1 in cost. A second building like an assembler 3 is more. But loading a beacon is like putting down a second and doubles its value.

Except you have more trash you need to pick up right? So your recycler and benefit scheme feels this as .95 items vs .90. So you have to buy more machines there.

At that point in terms of volume you get 1/2 as many drills. On top of 1/2 as many big drills. 1/4 in the worst case. Or 16/25 in the best using simple ratios. And to put no fine a point on "here's how bad that gets" you can put quality modules in biolabs and down that entire assembly line and feel exponential decay all the way down.

Romayne. I think I need to do one more example because you used forges. And I used forges. In terms of machine energy compared to the world of 2-8 that's 34 forges worth vs 18 forges optimized to: 2 furnaces. That are so underutilized they don't need to be scaled to handle a legendary quantity of components because the rate is just that low. That exponential disadvantage is a big hint. Gears in the tutorial might be another. Speed by volume and speed by speed might have both just lost. But I was shredding 40 items per selection so a big leaf might not be the way to go for this after all. I won't know till I mess around.
Top
coffee-factorio
Inserter
Inserter
Posts: 40
Joined: Thu Oct 17, 2024 10:56 pm
Contact:
Contact coffee-factorio

Re: Quality and Speed: How much waste does haste make?

Post by coffee-factorio »

So I got a new model based on applying the back end scaling 1 at a time
to 2-8 iron til it required more than a few furnaces using pipes.

Got it running at 2.9 ips vs the original 1.6 ips.

Image

There where a few things going on that I was curious about.

2-8 fails to look good because it's catching less iron.
And in terms of ??? forges I'm thinking the mistake is trying to
scale your production to 2.5 input expecting to get honest numbers.

Volume scaling alone turns out to be a form of speed scaling. If I increase volume of candidates by 1.2x the number of legendaries drops so fast that you can catch it the production graph at 64x speed.

Difference was 2.3 to 2.7 ipm. I guess that's like rolling a 20 then a 10 being better than rolling another ten every soften, you just roll the 20 always even your field of numbers is smaller.

The I went fast for free example that lets me hit insane units per second is rolling 1d10 per 40 items instead of 1d10 per item. So I started with pipes as a control then went to armor on purple of them to see what effect that would have. I mean if the dice doesn't have a memory then who cares? But in practice I'm playing less games for bigger hands.

That took it down from 2.7 ips to 2.2 ips. Still better than going from a handful of furnaces to 34 forges o_o. But in terms of cuts 2.2/2.7 means I better be in a place where I can write off about a fifth of your operation to packaging costs (what this seriously is).

If you go the other way with sticks it went up to 2.9 ips. Basically even though I'm getting back half as many plates I'm rolling twice as many times. So that's an example of pulling volume out thin air basically. So in terms of micro-optimization, that might be one because my production graph was jumping around a lot. Which puts me at... not 18 furnaces for half there.

We're in a different field now. A furnace makes more volume does so much more that even doing something that is going to handicap two of it's selection passes lets it produce twice as much for a given output. Speed 3's on the furnaces then body armor.

Ah, I should have put away the sledge hammer after all ichy. Still it let me throw four more hammers at the same nail to see which one works best, and if you can handle the volume that appears to be generating more stuff.

Furnaces remain a technicians delight. Because of that volume increase, I might just drop a selection pass if I can see how to add it elsewhere ;)

I ah, need to eat that third one for parts though too now that I did the math because I'm using Speed 3's lol.

But the justice of that logic is kind of what got us here, if I don't grab the loot I don't get it, and that's why you want something small for space platform even if isn't theoratically productive: I mean your pool of candidates is infinite * width. \0/ who cares!

I gotta come up with something better. It might not run off a space platform but Vulcanus has room for something like that. But like, I'm probably going to tune that in private then put it on pipes so I'm not just posting to many pictures. Whatever you can say about speed you're seeing constant factors lost in an environement where that'll compound and eat the line - any kind of speed beacon normal quality unfortunately is the example of that because you're literally scaling by -7.5/10_000 in the best case (just 1) multiplied by every quality benefit it hits.

To be very clear, if that's a row of furnaces that's fine. But if you have a furnace then an recycler say you're starting to experience exponential decay.
Top
ichaleynbin
Inserter
Inserter
Posts: 25
Joined: Wed Feb 02, 2022 2:55 am
Contact:
Contact ichaleynbin

Re: Quality and Speed: How much waste does haste make?

Post by ichaleynbin »

coffee-factorio wrote: ↑Mon Dec 16, 2024 11:18 pm
Calling me crazy is one thing because I know I am. I've been playing 800 hours on a mod where a linear solver like Foreman will flat out lie to you and get you to piss away 4 gigawatts power on 5 steel. As far as my English, that isn't an issue on a forum with people from all over. So I thought I might critique your build with my normal flair.

My spot. Is that I noticed what you're saying about percents and then looked at what that does with simple back of napkin math in a moment. You made me just mad enough to do what any sane person should do before they try and build something - but this is a game right?

So the quality table states that the probability of getting a legendary in the best case is 0.1%. But that's applied down the entire line. So a legendary is 1/1000.
But I can scale those numbers to be 10/10_000. You multiply 10/10 right?

When you apply a speed module you go to 0.05. With normal quality that goes down to 5/10_000.

And if you divide those numbers 5/10_000 / 10/10_000. That's 1/2 as a simplified fraction. That's what you're getting.

At legendary it's better. Because it's ~20/10_000 instead of 25/10_000. So that's 4/5. edit: which means I need to change a number.

But ah, see the thing is. He has 1/2 the green quality ones too. So those contribute half as much. And they're ten times better than the normal ones even though there's ten times as few. On purple where 2-8 saw the most benefit. That's where this is real clear. Half as many will jump because half as many exist in the first place.
Foreman doesn't lie to you; It exactly answers your question, and you are responsible for asking the right questions. Maybe 5GW of power is the easy thing to come by, with fusion it might be for some people. But yes, I mentioned that an increase to second order chances can be seen as a positive as well as a negative a few days ago.... When I said that haste gave as many second and third order legendary plastic per second as no-haste, among others.

If there's a clear tradeoff I can see, I feel like I'm asking the right question, it's Factorio and Wube are VERY good at puzzles. Solving for optimal points is fun, I do it all the time, but to imagine that we could figure out the entirety of the quality meta in the first two months is very funny. There's at least one or two undiscovered nuances and I refuse to pretend like I've got a global optima located yet. Maybe it is, but even saying local optima makes me iffy.

Haste is incredibly worthwhile on some steps, as part of a balanced breakfast, depending on how you're evaluating. The biggest thing it saves you is investment cost into machines, better i/s/machine, which allows a budding factory to snowball more quickly, but any situation where time is involved (literally all of them?) it's worth a thought. You can't fairly evaluate these things based on isolated builds, you're not producing legendary iron for the iron itself, you're producing it for something else.

The plastic question is interesting because at least two solutions are being ignored and probably more, and isolation is the reason. If you ask "how best to get legendary plastic?" you can easily get different answers than "how best to get legendary red chips?" Again, a 300% production cycle wins and if you're willing to yeet the copper and steel, or able to consume it, LDS is excellent for upcycling specifically plastic. If you're looking for red chips, and have a way to consume the excess copper and iron, maybe it's blue chips. Whatever it is you're building, it's almost assuredly not just plastic.

I still don't see a reason to evaluate over input item count, which as far as I understand it is OP's concern, and why I'm saying OP's concern isn't valid. When I do try to optimize over item input count, the little trick I can do with amortization (taking into account the item cost of the additional legendary machines), haste still wins for legendaries-in-hand, and per time. No-speed is behind on count until the initial investment cost of the additional machines has been recouped.
Top
ichaleynbin
Inserter
Inserter
Posts: 25
Joined: Wed Feb 02, 2022 2:55 am
Contact:
Contact ichaleynbin

Re: Quality and Speed: How much waste does haste make?

Post by ichaleynbin »

If you care about i/i, use only 300% cycles and go lossless. I checked this approach for legendary science to try to decrease i/s input rates, and didn't like it very much, blue chips doesn't generate enough iron, or alternatively LDS generates too much copper, I ended up with tons of excesses. For i/i, nothing can compare with a lossless cycle, but personally I'd rather not constrain myself to 5 recipes.
Top
coffee-factorio
Inserter
Inserter
Posts: 40
Joined: Thu Oct 17, 2024 10:56 pm
Contact:
Contact coffee-factorio

Re: Quality and Speed: How much waste does haste make?

Post by coffee-factorio »

I'm an FO's son. FO means forward observer. So at face value what I should do is just say "I got your number" and see if there's a friend on the radio.

But this needs to be kept around at least before the rounds hit. I kinda thought I heard the guns going off when you made that ESL comment. Because a fraction of your big drills produced is always going to be a fraction of anyone else who could afford a second building could do, and it shows exponential decay the more you try to optimize. And that's going to hurt someone on the basis of time spent playing with a smaller assembly line. I got your number.

But I tried to optimize my build. I know that the red on the modules means stop. Because I got your number. But I wondered what the white is supposed to mean.

This is a walking corpse and a person can't get a number on it. Dracula's loaded so if you don't play with the dragon someone else will take your ticket and you'll still lose. My next example will show why.
Top
coffee-factorio
Inserter
Inserter
Posts: 40
Joined: Thu Oct 17, 2024 10:56 pm
Contact:
Contact coffee-factorio

Re: Quality and Speed: How much waste does haste make?

Post by coffee-factorio »

Power was mentioned once or twice here.

Fluids weren't. Liquid iron as a metric is 83K.
Power is 170MW. Or is important too: 4.3K. Theres variations between builds but 170MW is what you're looking at.

When we remove fluids from the equation we're doing furnaces.
Power is 8MW. And in the world of "I don't care I can copy and paste",
for a given power grid if I'm subscriber to ichaleynbin's church (it's a good church, and all have their flaws) then I'm free to copy and paste up to 160 MW; just by not optimizing prematurely.

So if I apply that by forges here's what we get:

Image

67MW. 2.3K ore. 47K fluid. 2.7ips output. 12 forges but I'm making half as much, even though I cut raw speed by more than half. Half material usage, third of the power and compared to using faster machines I'm doing 2.7ips and not 2.1.

Eyes of the vampire my friend. If you think for a second you are in a better place this will tear out your throat. It already is, if I just buy a second 2-8 I match it using ichaleynbins scaling argument augmented by using a slower production method.

Here's what I mean by that. If you say 67:170 ~ 60:180 you can copy paste.
And I'll still be sitting there going "One basis of output mines more compact, just don't overdo it and it'll always be cheaper".

But you'll be able to say "It's eating more ore. 2k per second is less than 4k. This is a micro-optimization from the point of view of volume".

Well what happens when you double it is you immediately use more than an optimized speed example. Which means you can have an argument with someone on that and they'll come up with a material handling optimization that lets them do more for slightly more power.

Even though body armor wasn't the most efficient solution from an ips point of view, it was good enough to hit six. If I double 2.7 I hit 5.4 ips and I'm using more material. From a practical perspective just buying 2-8's yields a more powerful ips count overall and 24 is just not 120MW.

If you look at any one metric you're going to lose. I mean if you take the most extreme case and say don't do it all, I'm sorry space platform engines just don't work that way.

Da bunner was humorous. But it is also instructive. At a fair rate it's 25k/10 = 2.5k iron recipes. It'll show up in a design but you'll need it more if you sacrifice processing speed. It also alludes to something not meantioned: calcite has a machine cost. This can probably be neglected except that if you replace a calcite machine with a simple iron layout you get 60 iron, and simple asteroid recipes roll a new rock every 1:5 of the time which might be quality.

A forge can't benefit from that. It outputs a fluid no matter what quality your input is. So if you can pull off a scaling argument to match input by using simpler steps - that's great. But you can do stuff where you remove constant factors of legendaries at a time.

OP in technical literature a dragon is a hard problem without an easy solution. Probably the worst one I can think of is hailstones numbers. Packing a box with items with different values is another one.

You're right to be worried OP. I looked up what Dracula is supposed to mean, I guess it's connection to the word dragon is from latin. Dracu - Draco. The ending la is something that happens in Slavic languages where they say "son of" - Son of the Devil. Close enough. Da bunner is Bunicula.

The technicians delight is still in forges. They do more interesting things than just use fluids. That's why iron sticks from forges was promising.

Man, I thought I had a table at the Monte Carlo casino but I guess I'm gambling at Castlevania. At least I didn't put any of this up in my game.
Last edited by coffee-factorio on Tue Dec 17, 2024 5:37 pm, edited 1 time in total.
Top
coffee-factorio
Inserter
Inserter
Posts: 40
Joined: Thu Oct 17, 2024 10:56 pm
Contact:
Contact coffee-factorio

Re: Quality and Speed: How much waste does haste make?

Post by coffee-factorio »

Not my best work, I'm still sick :\
Top
Koub
Global Moderator
Global Moderator
Posts: 7902
Joined: Fri May 30, 2014 8:54 am
Contact:
Contact Koub

Re: Quality and Speed: How much waste does haste make?

Post by Koub »

coffee-factorio wrote: ↑Tue Dec 17, 2024 3:51 pm I'm an FO's son. FO means forward observer. So at face value what I should do is just say "I got your number" and see if there's a friend on the radio.

But this needs to be kept around at least before the rounds hit. I kinda thought I heard the guns going off when you made that ESL comment. Because a fraction of your big drills produced is always going to be a fraction of anyone else who could afford a second building could do, and it shows exponential decay the more you try to optimize. And that's going to hurt someone on the basis of time spent playing with a smaller assembly line. I got your number.

But I tried to optimize my build. I know that the red on the modules means stop. Because I got your number. But I wondered what the white is supposed to mean.

This is a walking corpse and a person can't get a number on it. Dracula's loaded so if you don't play with the dragon someone else will take your ticket and you'll still lose. My next example will show why.
I'll be honest, I didn't understand a single word. I mean I DID understand the individual words, but for the life of me, I can't assemble them into anything meaningful. I'm not saying they have no meaning, but I couldn't find one. And if I can't, I'm pretty sure many if not most readers here, who, like me, aren't native English speakers will have trouble too.
Please be kind enough to write for the community to understand if you're to write anything.
Koub - Please consider English is not my native language.
Top
ichaleynbin
Inserter
Inserter
Posts: 25
Joined: Wed Feb 02, 2022 2:55 am
Contact:
Contact ichaleynbin

Re: Quality and Speed: How much waste does haste make?

Post by ichaleynbin »

coffee-factorio wrote: ↑Tue Dec 17, 2024 3:51 pm But I tried to optimize my build. I know that the red on the modules means stop. Because I got your number. But I wondered what the white is supposed to mean.

This is a walking corpse and a person can't get a number on it. Dracula's loaded so if you don't play with the dragon someone else will take your ticket and you'll still lose. My next example will show why.
Hey, you should probably check out and go touch some grass. I'm not even reading your posts anymore. This isn't about right or wrong or facts at all for you anymore, this is a personal vendetta now.

I'm done here, if you think getting the last word wins an argument, go ahead, the last word is yours, but you're still wrong.

Koub, I almost reported that message as a physical threat, I'm not sure how to parse it either. This thread's going super off the rails and coffee's just been spamming numbers in hopes nobody is willing to put in the effort to check their numbers. I know I'm not willing and I seem to be the most willing to engage with them.... I can't say they've actually responded to anything I've said yet.
Top
Post Reply

Return to “General discussion”