A couple things: Q: Why didnt you go straight for launching rockets? / Why did launching a rocket take X hours? When I started the run, my immediate goal was to hit that 200 spm, i didn't want to build a base to launch the rocket just tear it down again to setup the megabase. If the goal was to just beat the game, this run definitely would have been a lot shorter. (maybe it should have been?) Q: Why didn't you put the sushi circuit on its own power grid? I'm stupid. (also oops) Q: What's with the french subtitles? I was automatically generating the english subtitles for the video in my editing software, and it lost it's damn mind and randomly translated the entire thing into french, which i thought was pretty funny, so I uploaded them, as is. Also, if you have constructive critisim about the video (or questions in general!), please let me know! i really do want to hear it
I would love a round 2 with these ideas. Plus having the production cells not in mass groups. The spreading out of iron furnaces was great for example. And not putting fluids in barrels, would be a massive headache saver
@@AustralianCapitalist yeah i only barreld the fluids because i knew it would be a PITA, and as for spreading out the production, yeah i thought about it during the run, but at that point i would have had to basically start over to rebuild it Definitly would have made the base a lot better oh also long inserters to reach more than one belt, why didn't i do that???
@@DocJade How? It lets you place blueprinted items just by hovering over the ghost, and only if you have the item in your inventory. It expedites a regular process slightly. Nothing more, nothing less.
Me foolishly opening this video: "There's no way he can possibly make something more horrifying than Dosh's enormous sushi machine." Me five minutes later watching you build a MODULAR GRID-BASED SUSHI BASE: "I am a fool."
honestly with a lot of calculation I feel like it could work the problem is crafting really slows down because items aren't found as easily solution is more items, but how much? percentages, we need to assign a certain percent of the belt to certain items, this meams to limit the amount of items to x*total belt count for example let's say we need 10 percent of the belts to be iron and we have 69000 belts, we should allow 6900 iron on the belts so every belt carries on average 1.5 iron at any given time despite having this solution, I have no idea how to implement it without having the human reset the item count for every single item every time a belt is placed
@@he3004 Couldn’t you use a circuit on a specific provider chest and only have bots place belts, and only have belts available via that chest? The downside is that I think you could end up over saturating the belts depending on how that circuit is set up (via the inserter- I’ve never tried to use a circuit to count items taken out of a chest by bots so idk if that’s possible)
This reminds me of how a cell makes proteins, just random diffusion of different proteins that happen to bind to ribosomes when they get lucky enough. I wonder if you optimized so ingredients are produced on average closer so their consumers if the base would end up being more efficient. I also think that maybe making the lanes between the cells 2-way and making the intersections truly 4 way maybe could have balanced out the distribution of items better.
Other people's brains: Logistics bot network for delivering dopamine and serotonin neurotransmitters with circuit-network combinators conditionally controlling their production and usage, combined with LTN train stations for cellular macronutrients. My brain: Sushi Megabase.
I would expect evenly distributing each recipe across the base would lead to better throughput. Each block of the same recipe is effectively competing with another, so spacing them out makes it less likely to starve each other and more likely for any resource to randomly run into a consumer.
One very easy trick to ensure your count combinators keep accurate counts. Set up their own little power supply - especially once you get solar and accumulators. They are very low draw so really dont need much to ensure they never suffer blackouts. After watching you and Dosh do this Im tempted to try this myself in a world, but I would want to use the [each] >= [item] trick and ignore the on belt ratios, jut add whatever is lowest on the belts until they are full. Combined with a system that remembers everything that was ever seen this should
I didnt do this for 2 reasons: 1: It forced me to upgrade power, I'm usually really bad about this. 2: Lazy. bonus: I was a tad worried that if I put it on a separate network and accidently hooked it back into the main one, i wouldn't notice and ruin the count without ever knowing Yeah, I really shoulda done that. oh well
As someone who is currently in their first playthrough, the fact that this build is still more efficient than mine makes me sick. Also the phrase "we need to expand" gives me "just add another lane to the highway bro" vibes.
Around 34:58 youre talking about the north-west bias, and after racking my brain ive come to the following conclusion; Since your blocks go clockwise, and each time it encounters an edge that doesnt split, it adds a little bias. Now, if it was a perfect square, that wouldnt matter, but due to the V-shape there are more edges at the bottom than the top, so a slight bias. Towards east. And just from how slight that should be, i fear the total playtime already...
Nice to see my 8to8 balancer design made it's way out of the KoS discord :D that thing was a headache to shorten I'm half convinced I can go one more shorter still but it's been ages
12 (including belts before and after first balancers) is the limit according to my experiments with Factorio-SAT, which checks if there is a layout that fits a balancer network. So unless there's some breakthrough in that, no way to shorten it more.
I did get an imperfect 11 using an extra splitter iirc which is why I thought I could get it but I never used tools, just trial and error and intuition, good to know though, saves me stressing about it
I've been (thoroughly) planning a sushi base build since Dosh's video came out, so here's some of my thoughts about your plan with that context: You actually avoided one of the main mistakes with Dosh's sushi build - bursts. A city block design spreads out the timing distribution of items enough that machines have the ability to almost constantly produce stuff instead of being in the cycle of overproduction and zero production. Your city block design has a bit of an unbalanced distribution for items - using standard balancers between junctions is easier, but it fixes the path of the items into vertical-horizontal zig-zags, which I think is the reason why your items got biased into one side of the base. You could avoid this with some way funkier balancer designs. You've counted the total items in the whole factory - I think that it's better to count a "percentage of items on belts" or something to that effect. Basically, in what proportion the items are sitting on some portion of the belts. It both makes it easier to track for congestion - if the total percentages sum up to 100, the belts are full, and it decouples the base size from the equation - you want roughly the same proportion of iron with a small base as with a big one, so if you're tracking the proportion, no need to manually adjust the settings again and again. But that needs a fair bit of work with circuits to control everything properly. Belting fluids isn't a mistake - just imagine what pain would it be to make a grid of all the fluids, and how much lag would that cause! Anyways, if I'll get to finishing up my planning and going through my sushi base, I'll definitely record it to show how it's done!
It's clear that you've put a lot of thought into your sushi base build, especially with the goal of avoiding the burst production issue seen in Dosh's video. Your city block design does help in maintaining a smoother production flow, reducing the cycle of overproduction and zero production. Regarding item distribution, using non-standard balancer designs could be a creative way to prevent items from biasing into one side of the base, as opposed to the typical vertical-horizontal zig-zags. These unique balancers might help achieve a more even distribution. Counting a "percentage of items on belts" rather than the total items in the factory is a great approach, as it simplifies congestion tracking and decouples base size from the equation, which is a smart way to maintain consistency as your base grows. However, it does require careful circuitry to control everything effectively. Belting fluids instead of creating a grid of fluid containers is a practical choice to prevent potential lag issues. I look forward to seeing your progress on your sushi base and the recording you mentioned to showcase how it's done. Good luck with your project!
"You could avoid this with some way funkier balancer designs." It would just need to prioritize flipping items between the left 4 and right 4 belts. That would ensure that an item passing through a north-facing balancer that came from the west will likely continue east (and an item that came from the east will more likely continue to the west, and the same for all other balancer directions), which will spread items out further and faster. Also, this balancer (visible for example at 19:29) is NOT a throughput unlimited type. I wonder if that contributed to the top left being unable to get items to the bottom right fast enough. I don't really think so, but who knows... "you want roughly the same proportion of iron with a small base as with a big one," But you might want a larger share of iron on a less developed base than for a more developed base. The more developed base has more item types on the belts. On the other hand, running the less developed base with the final iron density will show if that will work. So yeah, I guess I agree.
@@Pystro So with balancers, I am actually looking at balancing withinjunctions - I've posted an early version on reddit, but the gist is that each "lane group" has half of the lanes going the opposite way - so on each junction each item has about equal chances to go in any of the 4 directions. The fact that the balancer isn't throughput-unlimited doesn't really matter unless some of the output lanes are blocked, which was not a problem here. As for iron proportion, what I meant is more like "how much iron is on 100 belts" rather than "how much iron is there out of 1000 items". Belt utilization/fullness should increase a bit as the base grows, but I think that timing between iron passing some random belt in the factory should be roughly constant.
Ratios are probably a good way to approach it, but it led me to think of what I think is the true solution - Think of it in terms of throughput. A red belt carries 1800 items per minute. Lets say we're aiming for 200 science/minute like in the video - That means 200 red science has to be passing by the lab every minute. But that also means that 200 copper plates and 200 iron gears have to be passing by the red science assemblers every minute. Which is 400 iron plates. And 400 copper ore and Iron ore each past the furnaces. That's already 1800. A completely full red belt just to do red science. Now imagine all of the copper plates for the ~200ish LDS you need per minute for space science The other factor is that having an 8 wide sushi belt like the vid doesn't actually help if your assemblers are only being fed from one belt, because they'll only "see" that one belt of items. If your inserters are only picking up from one lane, then your throughput is capped at 2700 on blue belts no matter how many lanes you have. The extra lanes are just extra space to fill if you're not grabbing from them.
The first problem that I've seen with every sushi base ever is item count leakage. In THEORY you should be able to just count the items thrown onto the belts and taken off, but since it never works, I would not build a sushi base without adding logic that can update the count to what's actually on the belts. With this modular base, you could have just wired up all belts in front of (or behind) a balancer, and that would have counted every item every 6-ish seconds. If you then divide the count in your memory cell by 2 every 6 seconds, it will fluctuate between the actual amount and twice the amount. A very easy way to achieve that would be to throw a single fish onto the sushi belts and do the division every time that fish passes one of those counting stations. Putting it onto the sushi belts themselves and not a separate loop with the same cycle time will automatically account for when the belts get backed up. Since the count will trend towards an accurate number really quickly, you wouldn't even need to bother counting the items you place onto and remove from the belts. If you want a more accurate count, you can subtract 1/10th of what is currently on the memory cell any time the fish passes a counting station. Then the value in the memory cell will fluctuate between 9 and 10 times the actual count (and the count will reflect the running average of about the last 60 seconds). In that case you'll want to change the memory cell by 10 (or 9) times what the inserters grabbed from and placed onto the belt. A second problem is that for the item count targets on the belts, people always seem to just take the production rate (for a minute or whatever) and make that the count. Let's say you have iron smelting and brick smelting. You'll have more furnaces for iron than for bricks, but both types will want one of their "ores" to pass by their inserter every 1*0.625 seconds. (And since the assembler that makes rails wants a quite similar number of raw stone, it won't really change the number of stone needed.) In reality, you'll probably want to choose the target number so that it matches the highest consumption rate of *a single* machine. A reasonable 0th order approximation to that is that you assume that all machines consume at the same rate, in which case you'll want to allocate 1/Nth of the belt space to each item. (Where you actually have N-1 or N-2 items or so; you'll definitely want to reserve some belt space to be empty so that you can put items onto it, and you might want to account for the items that the previous machine has put onto the belt). A slightly better 1st order approximation is to take the production rate and divide it by the number of consuming machines. Also, to guarantee 100% uptime of your machines, you may need buffers. Assume you have 38 item types on your belt, and you set the target for all of them to 1/40th of the belt space. That means that ON AVERAGE, the next item that the assembler needs is 5 belt pieces away. But there is a 50% chance that it's not within those next 5 belts, and 12.5% that it isn't within the next 15 belts, and 6.25% that it isn't even within the next 20 belts, and 1.5%ish that it isn't within the next 30 belts. In order to guarantee high uptime, you'll probably want to pull from the belt into a chest, in order to increase the time that you can afford to wait for an item. For the final mistake, look at a tier 3 assembler for iron gear wheels. It needs an iron plate every 0.200 seconds. A blue belt can carry 45 items/s, but due to the second point we can again assume that only about 1/40th of the belt will be gear wheels. That means a single blue belt will carry about 1 gear wheel per second past the assembler. In order to carry 5 gear wheels per second PAST the assembler, you'll need 5 blue belts!!! And that won't mean that you'll get the 5 gears INTO the assembler. While any inserter is swinging, it has an opportunity to miss gear wheels on the belt. Realistically, you'll probably want the ability to grab from 6 or more belts. Without buffer chests you can grab from up to 16 belts, and with buffer chests it's unlimited (due to the ability to chain the chests).
Counting items off the belts is probably a bad idea at such a scale. Each belt connected with logic causes a transport lane break. At such a scale as the base in the video, this would likely double or worse the number of lane sections to update. On top of the increased complexity from having to sum all items on the belts, this would likely massively reduce performance.
@@drsupergood8978 Valid point. I don't know much about Factorio performance and especially little about how belts are optimized. The performance impact with this suggestion that I noticed is that hundreds of items would be counted on the belts every tick. But what's the point in making something fast if you can't make it work? It will just grind to a halt earlier. It might instead make more sense to come up with a method to debug/pinpoint _where_ leakage occurs. That way you could fix all issues and then remove the extra circuitry to get maximum performance. (In addition to only applying that modification to a small part of the base at once.) I don't really know how I would do that, though. You'd need to count everything that enters and leaves a certain section; and I'm not just talking about on the belts, but also via the inserters. And getting that last part to work without disrupting the total counts in the base would be the tricky bit.
Buffers simply cannot increase uptime. No single machine can get a better throughput of inputs than some portion of a belt, determined by the proportion of the item on the belt (unless you do some stuff to route as many different belts to a single machine. They might reduce the fluctuation, but inserters will insert inputs for *at least* 2 crafts in advance - that's a buffer. Buffer chests can only reduce the fluctuations in the uptime, but they cannot reduce it.
@@ignaloidas That's only true in some circumstances. You can indeed not increase uptime beyond what the average flow of items past/into the machine allows. And in many sushi bases, this is THE limiting factor. Simply because the number of machines built and the SPM goal is usually out of scale with what the belts can provide. But the conclusion changes if you have some types of assemblers that are very busy and you sized their number to the desired output (and done those calculations correctly), which I would call requirements for calling a sushi base "working correctly". In that case, any fluctuation would decrease uptime (because fluctuations of uptime above 100% aren't possible).
““This entire thing is the quote, not just the part in quote marks.” [Quote marks, brackets, and editor’s note are all in the original. -ED.]” - Randall Munroe
Just want to say that when I saw this video I thought "man... I don't want to watch another factorio video" but then I remembered your name and got very excited to watch another of *your* factorio videos.
You're a madlad. I love it. Plain and simple. I'll probably never try one of these after watching both you and Dosh do it, I just don't think I have the patience or sanity for it. But I love both of your runs
You could add a constant combinator to the cell's blueprint that adds 1 of some signal (e.g. red) to the network, this keeps tally of the number of cells you have built. Then you can add constant combinators that set the amount of items there should be on the belts per cell (i.e. available to each cell at any time, on average) and multiply by number of cells to get the number of items total to put on the sushi belt. Each red belt cell belt loop holds approximately 1400 loose items, so you could use that to gauge available physical space on the belt when adjusting the number of items per cell.
Wow this was a great video! The sushi belt run is something I've always thought about doing but am just too scared/value my sanity too highly. Nice run.
Sometimes, genius is mistaken for madness. Those few enlightened souls who stumble upon a higher plane of understanding, who see and perceive the world in a way we can barely comprehend, are the people who are so genius that madness is simply a consequence of our flawed perception of them. This case is not one of them, this guy is clearly insane.
You can theoretically sort of get around the item count inaccuracy issues by abstracting away the item counts into an "items per minute" rolling average and using that to manage the belt contents. This has the benefit of scaling inputs automatically with consumption so you don't really have to fiddle with constants, and since it runs on averages instead of hard counts, it'll eventually just heal itself after a brownout, though it might take a while.
Now this is a true work of art. Somehow this is more beautiful and amazing than train supplied mega bases. I would love to see you make another base but this time with practical sushi.
yo! I love it. my first base was a sushi base... kinda. was 3 huge blue loops of 4 lines carrying everything (welll... to me... then... they were huge... these days I've got bigger nuclear power modules. ) my last playthrough ended in a no-belt world where everything was moved by bots or inserters and storage. warehouse mod plus long reach fast stack inserters can move 300k items 22 tiles per tick per line.
If you use burner inserters specifically for fueling the boilers you wont have the issue where the power runs low and they struggle or stop refueling the boilers. If your power runs out completely, or you run out of coal, burners will automatically restart once coal is restored.
It's beautiful. Some interesting thoughts that come to mind is the items are analogous to molecules and the belts are analogous to capillaries. Perhaps you could take inspiration from life since it already fixed this problem for you! I remember reading a reddit post about the blood bus. Basically hooking up all belts and reading contents on hold signal which would accurately count all items except maybe ones in undergrounds and splitters. Another way is to have item sensors like the speakers, but you would want a specific items/min goal and when that's not fulfilled, the sensor sounds an alarm analogous to hormones. To get massively greater SPM it's helpful to increase the chances that an item will be what you need. This is done by preventing most of the items from reaching the sushi. Just have raw resources going directly to end goal items like science. Malls can just output in chests instead of on the sushi.
Your snap-to-grid skills abhor me. You had at least one blueprint that followed the grid, all you needed to do was copy the snap size and global offset to make more that snapped on the same grid! All you needed to do was fiddle with the local offset, but looking at your snapped blueprint, you don't even know how to do that! Edit: I still subscribed
Oh my gosh I just got obsessed a day or two ago about this other youtubers sushi belt base video and today - you release this! I am Sooooo excited, thank you!!!!
you can mark walls with deconstruction plan, so there will be gap between them, and then just ctrl+z that, its a bit faster then just mine and place walls by hand
In hindsight, you could limit how much the items run off to dead ends in the corners by simply.... not placing them. Make the only possible output the ones that feed back into the sushi
what you couldve done was desing a vacuum cell you can activate with a combinator or something that will slowly remove all items from the belts then when you give the signal they will release the items while counting them to get a accurate count once more
Technically Sushi belts are better for compact factories cuz it doesn't matter where you place buildings, the required items will come across at some point
One option for accurate counts would be to have a dedicated power supply for the counting combinator. Also, i can totally see this kind of setup with filtered stack inserters, for making modules. The imbalance is pretty much due to diffusion and localized production/consumption. Higher item count would be needed to counteract, or more spread out production of various things. Basically not having basic production on one side, and advanced production on the other.
I think when building a sushi base the ratios should not be those for x spm, because your assemblers will work on a probabilistic basis and crafting time matters less. Slower crafting time simply means that the inserters have more time to actually collect the resources, meaning the assembler will run more constantly. A Also even though it is sushi the placement of assemblers matter greatly. Green circuits should still be in front of blue for maximum efficiency. Imagine a 1 belt sushi base where the assemblers are placed green red blue: It could spit out a blue on the first lap. Compare to blue red green: Many greens first have to make a lap to get into blue, then when blue is full it can start going into red, then the red has to do a full lap to get into the blue.
I imagine one solution to the localized item shortages would be to spread the production of each item all over the base in an equally sushi manner rather than have it "organized" like with the oil all in the middle and iron/copper around the outside.
main problem is probably that you concentrated all of your production into "blocks". the stuff your crafters were looking for would take up the bits and bobs and then it's a 50/50 whether it'd even go deeper into the grouping. Evenly distributing everything everywhere would probably have done some good. Funny idea: the human body is like a sushi belt, the heart in the center, the lungs and stomach taking things in and the rest making stuff and sending it to be gotten rid of. Could do something like that; instead of the funny city blocks, gigantic branching 64+ belt wide things entering a behemoth of a "mixer" with the resulting gigantic belts fanning out and potentially splitting from blue belts into yellows around the production loops, only to re-combine and go back to the "heart". keeping track of approximately how much is on the belt would then be pretty easy; everything has to pass through the heart sooner or later, so just measure how often you see something at the heart over time, multiply it by something fancy like the total number of belts in the system then take an average over time.
If you're looking for ridiculous challenge runs, I'd love to see one where you have a central warehouse district and a rule that every resource must be taken from its place of manufacture to a warehouse, and then distributed from there...
In a large enough base, the simplified sushi circuit design of "red wire +1, green wire -1" becomes problematic - a lot of things could happen that mess up the count. An alternate idea I just thought of: detect items _on the belt_ with pulses on the red wire and use the green wire for your "computer" to tell the circuit what it does or doesn't want by simple presence. The computer can just sample however many items pass in however long it takes for your sushi to complete a full loop. This also makes it easier for you to adjust the desired counts and you don't have any problems if your system fails from biter attack or power stall, AND it doesn't count items dead in the corners.
good lord at least dosh's base was made in such a way that it ran stably with a *relatively* even item distribution and items couldn't skip entire sections of the base by pure bad luck
Holy shit bro! I definitely understand why this one took so long. Glad you got it done though. Definitely worth the wait. Quite the monstrosity you created lol.
The absolute hilarity of you accidentally launching a steel bar into space the first time, and then missing the launch of the second rocket with the satellite was too much for me Absolute peak
