Why Minecraft Players Built a Real Life Supercomputer

if someone gave you a screenshot of a Minecraft world a seed and a set of coordinates and asked you to recreate that screenshot you could probably do it really quickly if someone asks you to do the same but without coordinates and without the world seed and the picture was a really low quality and foggy image from 14 years ago with a Herobrine poorly photoshopped into it how would you even begin the naive approach is pretty simple let's just check every single seed in existence problem there's 18 quintilian of them every Minecraft world is pretty much infinite and the universe is going to start dying out in around 800 trillion years so we either need to search 200,000 seeds per second every second until the universe starts dying or come up with a better plan the problem of finding a particular Minecraft seat from an image or video is not a new one as early as 2012 a Minecraft forums user named Melo KB theorized that the location of just 15 slime chunks are unique enough to only generate from one seed and whilst that is good you'd be hardpressed to find the exact coordinates of just one or two slime chunks in any screenshot or video let alone 15 in 2016 and 2017 a player on the ccraft server leder 64 discovered that the Heights and Order of the naturally generated obsidian pillars in the end were unique enough to only be generated by at most 4.3 billion different world seeds which could be checked much faster especially in combination with any other world generation information that was already understood like slime chunks by 2019 PewDiePie and a bunch of other YouTubers took to playing Minecraft again and to avoid spoilers or more likely because they had no idea how to they didn't share their world seed that wasn't going to stop a small group of players however who were dedicated to finding that seed by any means possible it was known as early S 2014 that the coordinates of a few structures like Villages and water temples should also be enough to find a seed but finding that in a playthrough where the coordinates menu is almost never pulled up was going to be challenging the individual responsible for this hunt was no one other than dream himself who did the equivalent of holding a flashlight whilst his dad Neil seed captain wutx and klock did all the actual hard work eventually they were able to find coordinates of structures using a technique discovered by coolman 24 he found that the rotation of naturally generated blocks is exactly the same in every Minecraft world and so if you found enough of them you could narrow down the coordinates of any place this combined with two ocean temples and a dungeon was enough to find PewDiePie seed but we don't have several hours of footage to go over we have a jpad that was uploaded to for Chan with like a hill six trees a chicken and all taken in a version of Minecraft before blocks could even be rotated clearly we're going to need a new plan something that can focus on very small details like the generation of trees or maybe terrain and a player named K Minster had an idea to do just that if you're interested in what K Minster did to find the Herobrine world there's an anonim video you can watch all about it but by this point you get how it's possible to reverse engineer an image to obtain the world seed using a bunch of cool tricks but what about engineering a seed that doesn't exist yet for example seeds with naturally spawning 12 I end portals or an end Island that generates without any endstone or a really really really really really really tall Cactus our story starts around January 2020 right at the start of another seed cracking project saly One spec PNG project if you're interested in that you should watch his video series about it but a couple of the people working on that project decided to create a Discord server dedicated to seat finding and they imaginatively called it Minecraft research the goal of the server was to try and find the Panorama image from the background of the main menus as well as a few other seed finding projects that were also discussed in s1's Discord server among these people was Captain wutx Captain wutx was some somewhat experienced in seed cracking projects and had been working with other famous crackers like Matthew Bolan and Balin on some personal projects way before sal1 began his search for pack.png since 2019 the two of them have worked together on and off to find a nine block tall or bigger Cactus I actually asked Matthew why he thought to do this and he insightfully explained that large Plant good a third notable member of the pet. PNG team was Cactus duper an individual who owns a machine learning and AI company and was donating her spare computing power to some tools used to research for pack.png half jokingly Captain wutx asks how much to rent for a day I'll probably need your services for something else at some point colum 3 I just need to Brute Force taller natural Cactus it's only a complex algorithm he didn't know it at the time but by introducing the sal1 Discord server especially a person literally named C stuper to the idea of tall catti he may have just set in motion the events that lead to Minecraft players creating a real life super computer the idea set in the back of everyone's Minds for a bit until the server gave people a channel dedicated for it to which the people there decided to Branch off entirely and make a new server called Minecraft research before we go into more detail about Minecraft research we need need to talk a bit about finding cacti while seed cracking is finding the world seed from something that we know already exists with a bunch of useful features that can help narrow down a search seed finding is trying to find some conditions that might generate a world with some desired features but with absolutely no clues at all essentially every single possible seed has to be checked which means it's going to take a computer a very long time it also just so happens that cacti generate in such a weird way that makes searching for them even harder normally every desert chunk attempts to spawn 10 cacti in 10 random positions between one and three block stall and a cactus is only generated if the randomly chosen block is a valid placement for a cactus rarely a second Cactus can generate on top of a previously generated Cactus creating an extra tall cactus in some rare cases this can happen to a third Cactus and a fourth Cactus and so on this is what allows very rare tall cacti to generate where you otherwise never see a cactus taller than three blocks the generation rules are simple but that doesn't make it an easy problem to solve we still have 18 quillion almost infinite sized worlds to search in just a 100 trillion very finite years what if instead of looking for tall catti across massive worlds we somehow look for all chunks capable of generating Tall Cat cacti and then try to find what world seek can generate those chunks it sounds counterintuitive how do you look for something that hasn't been generated yet but it's entirely possible and actually way more efficient because we don't have to generate an entire world just a 16x 16 chunk whilst the seed of a world determines its terrain biomes and rivers details like caves vegetation and structures are generated using a different seed which is unique to each chunk the seed for each chunk is calculated by essentially combining the coordinates of that chunk with the world seed and a little bit of math and Randomness that is crucially very predictable and then this seed is used to generate whatever features fit the biome of the chunk as it turns out chunk seeds are calculated using only 48 bits of the 64-bit world seed essentially meaning that there are only 300 trillion chunk seeds to consider much less than the 18 quadrillion worlds we would otherwise have to search so how do you search 300 trillion different chunk seeds it's not like you can just generate a world and start counting every unique chunk seed you come across the chunk seeds with the exact same seed and biome can still look totally different if they have different terrain instead we can make some modeling assumptions let's just assume every single chunk is a super flat desert even if that isn't true and then run a simulation of the Chun generation code for a desert on every chunk and this code can tell us exactly how tall every cactus in a flat desert chunk with that seed is going to be we can then filter these results to only take Cactus taller than nine blocks the previous record we call these filtered seeds candidate seeds as they are candidates to generate a tall Cactus but we don't know for sure if any world seed will actually generate the right terrain and biome to create the theoretical Cactus but this shouldn't be too hard to check because there aren't that many seeds to check through the good news is that simulating this Chun generation is actually doable with really simple math which means we can not only do it quickly but we can do it on a GPU gpus are not just for gaming but are also very important when performing repetitive operations on very large amounts of data the reason they good at this is simple whilst a CPU might have between 4 and 16 very powerful cores capable of doing many complex instructions modern gpus have thousands of weaker cores capable of doing very simple instructions with extra fast access to vrrm 2 which limits downtime between instructions in simple terms each tore can process one seed at a time so using a GPU means we can process thousands of seeds at the exact same time whilst 300 trillion sounds like a lot it turns out that a single RTX 2060 can process 26 million seats per second meaning that we can do this All In just 117 days and actually what if we had two RTX 2060s we could do it in 59 days with three 39 days and who do we know that has access to a load of GPU processing power cus stuper in the time between starting pack.png and founding Minecraft research a pandemic decided to rudely interupt 2020 which meant a bunch of really nerdy people suddenly had a lot of free time and a lot of these people were willing to help out the code for searching for cacti chunks had already been completed by Captain wutx balent and Matthew Bolan it was just a case of organizing everyone's effort and this required a high-tech solution a spreadsheet this spreadsheet was pretty simple seeds were grouped into 3,000 boxes of 100 bill seeds that anyone could claim with a Discord bot and then whoever claimed each box would have to run the simulation on all 100 billion of those seeds and then report any cat I found taller than nine blocks this would take anywhere between 1 to 3 hours depending on what graphics cards were being used and people were also encouraged to use cloud computing free trials to rent some very powerful Hardware to use for this too but this spreadsheet could really only solve the surface of the problem problem you see as it turns out our model for Desert chunks was horribly oversimplified we actually can't just model a super flat world we have to model every single super flat world at all possible Heights because each one of those will have a different Cactus generation so we don't just have to check 300 trillion seeds once we actually have to check all 300 trillion seeds for every y value we can at least narrow down the search to only looking at sensible ground levels but each level we consider is going to add another 300 trillion seeds to the search not only that but finding a promising seed means we further need to search for all possible world seeds that can generate that chunk and the coordinates that chunk would be at oh and turns out there's another complication we cannot just consider single chunks at all if we're truly trying to find the tallest cactus Minecraft chunks generate every cactus like this first they choose a random Block in the chunk and try to generate a cactus there if it fails to place a cactus there for example if there's an adjacent block that makes the placement invalid the chunk will instead choose a different Block in a 7x7 square around the original block there is no requirement that the chosen block is in the same chunk so a chunk can actually generate a cactus in an adjacent chunk this was discovered by Matthew Bolan who quickly wrote some code for for searching two chunks at a time and it wasn't going to be quick the naive approach to this would require all two to the power of 96 possible combinations of chunks being checked which is just stupid and not worth even trying but if you think about it the tallest cus is probably based on a cus that a single chunk generates itself to be really tall plus an adjacent chunk scat thigh that overflow across chunk borders so instead of check all pairs of Chunk seeds we can check all tall Cactus chunk seeds against all 300 trillion neighboring seeds and actually due to a quirk beyond the scope of this video there are only around 19 trillion different seeds that can be neighboring to Any Given chunk seed so to recap finding the tallest cactus involves searching 300 trillion seeds one time for every single y level to search and then comparing every single one of those seeds against 19 trillion and other seeds when the Minecraft research team finally made it through the y equal 64 seeds they ended up with 42,000 497 candidate seeds which would mean 87 quadrillion chunk paars that would have to be checked this would be of course stupid so the candidate seeds were trimmed down to include only cacti that generated on chunk borders as these are significantly more likely to be influenced by surrounding chunks after after filtering each y l search still left around 12 quadrillion pairs of Chunk seeds to check they were going to need some serious processing power even more than Cactus duper could ever provide introducing chip wolf a cloud engineer who joined the salc1 server to help with the p.png project and offered a large number of gpus to help with that however like many others he became enthralled with the idea of large C and decided to dedicate much of his time there instead chip recognized the incredible amount of Brute Force Computing that needed to be done to find both p.png and toti and inspired by a project called folding at home he knew exactly what to do folding at home is a project where anybody can download a client that allows their computer to perform scientific simulations used to research diseases like cancer and Alzheimer's disease the concept was simple a bunch of science nerds made a program to perform billions of simulations of the human body and then made this program publicly available to download for all platforms from PS3 to PCS when run this would connect all the users who downloaded it to a central server that would then send each user a simulation to do for as long as they ran the program especially when they left their PC idle at its peak the folding at home network was effectively the most powerful supercomputer in the world inspired by this the goal was to create a similar thing that instead of Performing life-saving research performed crucial research in the field of digital botany luckily a group of researchers at the University of California Berkeley foresaw this need 21 years ago when they created boink an open-source piece of software designed to allow anyone to set up their own distributed computing Network assuming they understood how to write good code to WR on gpus and could understand a code base probably older than most of the people working on the project with the help of Nile seed hyena day and a few others chip wolf got to work setting up a point Network for Minecraft research becoming a distributed computer research project meant it was time for a new name and to fit with the name of the other similar projects they decided on Minecraft at home a name that you may have heard of before and a reference to all of the other at home projects currently using boink this was going to be the Minecraft super computer after a couple weeks of work on June 15th 2020 the project was publicly launched in beta this was after all 300 trillion single chunk seeds had already been searched at yal 64 the tallest of these created a 13 tall Cactus but the double chunk search Project had their s set on the theoretically possible 25 or 26 tall C C which could be generated by a 12 and a 13 tall caus stack combining before the network was finished and whilst testing the program a 19 tall cus had already been found and so the search for the 20 tall cus had began the beta launch was successful and the project fully went live on June 21st 2020 and was an incredible success with the help of a geosquare video and announcement from saly 1 the project kind of went viral the server gained a ton of members and the caus finding Network kept growing as new users would sign up to use it within 6 days a 21 block cus was found and then 4 days later a 22 block cus was found and then a month went by without any new discoveries this was as the project was nearing the completion of its y equal 63 seeds and had already searched every y equal 64 seed next up the y equal 62 seeds were to be searched which first required checking the 300 trillion individual chunk seeds too but still nothing the next Hope was the y equal 65 66 and 67 seeds and by this point the project had actually reached its peak over 2,000 graphics cards from 1,000 users were being utilized at peak times giving the Minecraft at home network a computing power of approximately 8 petaflops a fancy way of saying eight quadrillion math equations per second if the network were all brought into the same room and physically connected it would have been the world's 33rd most powerful supercomputer at the time almost matching the power of the strongest us government-owned supercomputer and all this power was being used to find cacti in Minecraft and this power was much needed with over 120,000 hours of computation time needed per layer searched there was no shortage of work for the network to give out and finally 6 month after the previous discovery on 10th of February 2021 the first 23 tall CTI was found and this is the tallest cactus that has ever been discovered whilst the cactus and Cactus project continued its search fruitlessly a few other projects were being run on Minecraft's super computer one of these was the p.png project whose Steed was successfully found in September 2020 with the help of the Computing Network another of these projects was finding the title screen background panoramas for every Minecraft version before 1.13 which eventually proved successful too alongside this project which only used users gpus an extra project to utilize CPU power was underway to find seeds that were beatable with just a single Minecraft chunk and another project that was super success successful and created some very interesting seeds as of March 2024 the Minecraft at home Point network is no longer running as there are simply no tasks left to compute since the release of 1.17 and 1.18 Minecraft's World Generation has completely changed requiring time for people to relearn it all and discover optimizations for any seed tracking projects time that many people don't have as the pandemic has ended and most of the original Founders have graduated gotten jobs and have totally moved on from the project without the donated computing power of thousands of strangers and the optimizations of the code written by the researchers we'd probably never know the seats for all of these worlds since late 2021 most projects have been optimized and are small enough in scale that the power of a supercomputer hasn't been necessary some projects use a custommade network known as micr but it's no were near as large in scale as the network used to be there are still smaller projects being worked on in the Minecraft at home Discord server and you can join it and take a look for yourself but for now the Minecraft super computer lies dormant waiting for another world changing pandemic or perhaps someone crazy enough to invent three or four Chan cus finding to awaken it again the original researchers have asked me to consider donating some processing power to other B projects that perform actual scientific research like climate studies cancer and Alzheimer's research or even astronomy your Dona computing power will help make the world a better place you can find a link to a list of all of these in the description of this video and they genuinely do make our world a better place special thanks to everyone who helped me research and made this video especially hyena day Cactus duper Matthew Bolan Captain wutx cortex voice Sanic dutchen 18 Matt 2869 Nile seed Tom Lao voco everybody who coordinated this project everyone who donated their computer to these projects and of course you for watching this video see you and have a nice [Music] day

Channel: HellCastle & Tylerrrr

Views: 839,474

Keywords: hypixel, skyblock, hypixel skyblock, hellcastlebtw, furryeboy, hellcastle, Tylerrrr, tylerwith4rs, supercomputer, BOINC, cactus, tallest cactus, minecraft@home, minecraft @ home, minecraft at home

Id: dOqUIvce22c

Length: 23min 23sec (1403 seconds)

Published: Tue Apr 02 2024