Hacker News new | past | comments | ask | show | jobs | submit login
Scaleway announces measures against abusive Chia plotting and farming (scaleway.com)
171 points by ephesee 4 months ago | hide | past | favorite | 180 comments

Somebody needs to invent a coin that works on proof-of-carbon-emission-reductions, or proof-of-global-poverty-reduction and do some real good for the world.

> proof-of-carbon-emission-reductions

That one is fun! First, I credibly claim to have plans to release ridiculous quantities of CO2, and then you negotiate with me to convince me not to. Then we can split the coins.

This is being done right now in the Amazon without blockchain. First you start logging a lot more trees. Then the world is outraged at how many trees you cut, and you begrundgingly go to the negotiation table to figure out the carbon subsidies you're going to get in exchange for not cutting those trees. But wouldn't you know, estimates of how many trees you're able to not cut per year are based on how many trees you just cut so you get way more than you would've in the first place.

Exchanging money for not doing things is fun.

Sounds like the classic https://en.wikipedia.org/wiki/Protection_racket on a global scale.

It’s like proof of stake - you have to prove you have a track record of producing the emissions.

Burn tires for a year then reap the reductions the next.

Sure - that’s pretty much how it works. If you own and can defend a jurisdiction where you can do that, other people will likely pay or trade things with you to stop.

A cheaper way to deal with you is to embargo/tariff your exports, until you come back to the negotiating table, hat in hand.

Not sure that always works.

Hang on, Fidel Castro is on the other line.

That was a life-or-death political issue for the communist party of Cuba (USSR subsidies also mitigated some of the damage the embargo caused). This is just a question of money - which is far more tractable.

Seems like there are some pretty large blocs who won’t embargo you these days.

That’s how the current carbon offset credit system works.

That’s actually pretty simple if you can live with centralization ... just certify CO2 offsets for example for each coin purchased. Then again ... what’s the point of putting that on a blockchain?

> Then again ... what’s the point of putting that on a blockchain?

Marketing. :p

> That’s actually pretty simple if you can live with centralization ... just certify CO2 offsets for example for each coin purchased.

We have this today, but "just certify" does a lot of work here. It's really hard to certify CO2 offsets in many cases (e.g., it's tough to verify that the trees someone plants weren't going to be planted anyway or that they will live to their expected potential and not get burned down prematurely).

Whenever you plant a tree, do the following.

1. Take a picture of the tree

2. Burn the tree down.

3. Create NFT of the picture of the tree burning down.

4. Use the NFT of the picture of the tree to certify the CO2 offset.

Inspiration from: https://www.bbc.com/news/technology-56335948

I don't understand, the offset of a burnt tree is 0. I get that it's a joke, I just don't understand what the joke is.

Lets change the order of things then.

1. Take a picture of the tree

2. Create NFT of the picture of the tree.

3. Use the NFT of the picture of the tree to certify the CO2 offset.

4. Burn down the tree.


Is it more obvious with this ordering? The NFT proves nothing. It doesn't matter when you burn (or otherwise damage) the tree. You have an NFT stating a carbon offset that is completely fake.

I'm just making fun of the idea that NFTs prove... anything useful... in these circumstances. Its the same thing as people who want to use blockchain to certify supply chains or whatever. The "Blockchain technology" doesn't do anything aside from provide marketing points.

> just certify CO2 offsets for example for each coin purchased

We have something like this in France. Energy providers (for example EDF, Electricité De France) must buy a certain number of CEE (Certificats d'Economie d'Energie, I think the concept is white certificates) every year, or face heavy fines (big enough that buying certificates is a big part of their activities).

As an example: let's say I want to change my window to have more isolation, which will reduce my energy consumption. This can produce a CEE, that I can "sell" to EDF, which will help financing the change.

I found a coin claiming to do this - tradeable carbon credits, where you could extinguish a coin to offset carbon. It seemed like it had no traction.

You could make carbon-neutral coin that automatically offsets its own emissions by buying and extinguishing carbon offsets for every transaction. Obviously this would be cost-prohibitive for Bitcoin, but seems practical for less energy-intensive cryptos. I’m surprised nobody has done this (at least, I didn’t find it by googling). I think you could make billions of dollars off this because of the current furor over environment impact of cryptos. And it would actually be good for the world.

This is one of the themes of Kim Stanley Robinson's "Ministry for the Future"


This defeats the purpose of disincentivizing adversaries who want to steal coins from other participants. The stake/spacetime/electricity isn't wasted; it is used to establish global consensus. You can't redirect those resources into "doing some real good in the world" the best you can do is to allocate resources directly to your good cause rather than trying to magically get the good done for free.

> The stake/spacetime/electricity isn't wasted; it is used to establish global consensus.

I still would like everyone to understand how proof-of-work really works, I don't think many people do.

The current Bitcoin hash rate means that all those miners are currently calculating on the order of hashes per second.

Of those, in Bitcoin only around 6 per hour have any actual effect. Six. Those are the ones that met the arbitrary target. The other 10^20 did not do any actual work, meaning that none of them did bring any miner, not even the same miner, the same chip, closer to the "answer". In that sense, they are wasted.

Instead, it is literally a lottery. The difficulty is adjusted such that if you try random numbers per second, 6 per hour will randomly hit the target. It's more "raffling" that "mining".

I’m not sure I see what you’re saying. The “arbitrary target” is so high because of all the mining. If there were less miners, the difficult would decrease, and then it would be easier for a malicious party to find one of those 6 lottery tickets.

I am simply explaining how proof-of-work works.

It is correct that the difficulty is adjusted based on the hash rate. If mining power was halved, it would become twice as easy to hit the target. If a new ASIC would make calculating hashes twice as efficient, the network would make hitting the target twice as difficult, effectively counteracting the efficiency. The goal is to always have 6 successful calculations per hour on average.

I really like the characterization as a lottery, it is just luck. Your odds of winning each “next” are weighted by your ops/sec, but nothing better.

> The stake/spacetime/electricity isn't wasted; it is used to establish global consensus.

In the case of PoS or PoW it is spent to establish global consensus.

It’s certainly waste if there is no other productive work is done by the network that wouldn’t have been done without it.

How about this: companies give out coins in exchange for useful labor contributions in the form of answering calls, stocking shelves, writing code...

That'll never work :/

Set up a coin such that every time it goes up, some amount of reserve is sold, and 100% of profits donated to carbon reduction projects. Give away free electric cars. Plant trees. Whatever.

Then hype up the coin. Make people bid it up endlessly just like they do other coins. Post lots of #moon #hodl crap on Twitter. Get the flock of sheep moving.

Sell hype, use the proceeds for CO2 reduction.

According to Bitcoiners Bitcoin is green, so you you could use the proceeds to buy Bitcoin.


I used to get paid to write stuff like this.

I mean, personally, I do own some cryptocurrency right now, if it moons enough in the near future I would be very likely to use the proceeds to switch to an electric vehicle.

If it moons even higher I could move to a more energy-efficient apartment with double-paned glass and all.

A coal company executive installing double pane windows doesn’t make coal green.

Cryptocurrencies create a fictional kingdom that is being defended by an army of miners. The laws of the kingdom only apply inside the kingdom.

Folding@home and SETI@home and The Great Internet Mersenne Prime Search already existed and could have had big nerds turn them into robot money somehow but naahhhhh. Of course not.


How do you cheaply and effectively verify that the real world work (sequestering carbon, fighting poverty) actually happened?

When the ultimate goal of zero emissions will be achieved, we will finally get a world where nobody is breathing :-D

A tree with a distributed ledger. The one who plants the most trees wins.

So, a Merkle tree.

Not sure what the "Rule 34" for crypto is, but:


A smart bike that prints coins based on the energy put into the bike.

i factory full of those bikes with motors to spin the wheels and gps spoofing

I’m sure all the crypto fans are lining up around the block to tell us that their pet technology doesn’t waste valuable resources, it just uses storage that would otherwise have gone unused, as they do with Bitcoin and electricity.

I think it's funny how the world as a whole just simply moves the problem (excess resource usage for something a lot of people have deemed 'excessive' like bitcoin mining) from one type of resource to another. Now we're going to see storage costs skyrocket, which will (hopefully not) affect cloud storage providers, datacenters, etc. etc. etc.

That would be a bummer, but if it doesn't exacerbate the energy/co2 problem I consider it a net benefit.

Manufacturing SSDs and platter drives is definitely not carbon neutral. Let's not even get into the problems with rare earth material extraction...

In order to be considered a net benefit, it only needs to emit less carbon than compute-based mining. It doesn't need to be carbon neutral (and indeed, I don't think anyone posited that it was).

That's a pretty low bar. To be a net benefit it should emit less carbon than traditional payment processing. If you set my house on fire you don't get to call saving my cat a "net benefit".

I don’t think you understand what “net benefit” means. It does not mean “the most optimal state”, it only means “better than the status quo”. In this case, the status quo for crypto is compute-based, so storage-based crypto is a net benefit if it beats compute-based crypto. Saying something is a net benefit doesn’t imply that it is optimal, so I’m specifically not suggesting that storage based crypto is optimal (indeed, I’m not even sure it is a net benefit; I only said “if it emits less carbon, then it very likely is a net benefit”).

How much carbon is emitted by traditional payment processing?

As an example: https://www.statista.com/statistics/881541/bitcoin-energy-co...

I'm sure modern, all digital, methods like Venmo or PayPal use even less.

Less than the carbon that would be emitted by doing everything with physical banknotes, coins, and precious metals?

I'm extremely disappointed Bram Cohen is behind this. I used to really admire him for his work on Bittorrent but this is going to be an indelible asterisk on his legacy as a coder.

So every conclusion to hard drive also applies to electricity? How do you establish the equivalency?

Computational resources too Actually pretty close to what some scifi writers envision(The Golden Oecumene by Wright comes to mind)

Perhaps the "paperclip maximizer" doomsday scenario was right in was right in its broad strokes but just wrong about the agents involved? They were worried about a super-competent AI that burns resources producing low-utility widgets. In fact we have created an system (a combination of societal incentive structures and technological tools) that burns resources producing low-utility pseudo-random numbers.

turns out people were the best "paperclip maximizers" all along

Hasn't the paperclip optimizer always been a euphemism for capitalism?

No, it's purely about AIs and the control problem.

The problem boils down to "How can I control superior AIs through mere power of inferior humans?".

The assumption is that the superhuman nature means there is no way disable a superhuman AI because deactivation will not fulfill its cost function and therefore must be avoided at all costs, the superhuman AI will always outsmart humans and prevent deactivation, the same way a skilled soldier will always beat an untrained civilian in combat.

We have solved it for humans by accepting that the victor gets to lead humanity. Can we accept AI rule? Probably not.

Even so bitcoin takes it to its logical extreme. You put valuable resources in one end and get money out the other side. In between, no useful work is done. It's the essence of capitalism without any of the beneficial side effects we've come to expect.

Gold mining is similar, although obviously not to the same extreme. Only 8% of the output is used industrially. https://www.statista.com/statistics/299609/gold-demand-by-in...

So gold mining is only 90% waste, whereas Bitcoin mining gets close to 100% waste.

Then again, perhaps printed dollars could perhaps also be considered highly wasteful?

No. The fable long predates the current anticapitalism trend.


Some people think Chia is so much better because of it uses SSD / Storage rather than GPU. I think this could be far worse.

Not everyone needs a GPGPU, but everyone need NAND and SSD. And it already causes price hike for SSD we will soon see something similar to chip shortage. The DRAM Fab which could be quickly retooled to NAND will be part of this equation as well.

This is not good.

> And it already causes price hike for SSD we will soon see something similar to chip shortage.

Nah. The expected ROI on a given TiB of storage is halved with each doubling of Netspace[0][1]. The current return is ~$1.50/TiB/Day. As the expected return declines, farming CHIA will become less attractive than other crypto 'investment' opportunities, and the whales will turn their attention elsewhere.

I'm guessing that netspace starts to level out after another doubling or two--assuming the exchange rate remains the same.

[0]https://www.chiaexplorer.com/charts/netspace [1]https://www.chiaexplorer.com/charts/xchTib

>The current return is ~$1.50/TiB/Day.

I'm bearish on Chia, but I can see why it's attractive at this price point. I thought the profitability was much worse, but I did the calculations myself and it looks like around ~1.30/TiB/Day.

At the $20/TiB price point for HDDs that seems like you'd be in the money after ~16 days (excluding plotting costs and farming power usage). The big problem is the lottery, which could be smoothed out by pool farming.

Yes, the rewards are getting worse by the day, so it's hard to predict where things will be at in 16 days time, but again, pool farming should theoretically smooth that out a lot.

Yeah, the problem is actually worse. Bitcoin is energy intensive. Chia is capital intensive. Producing capital costs energy so it effectively boils down to proof of work (burning energy to produce HDDs and SSDs). However, scaling energy is very easy, scaling HDD production is very difficult.

The only benefit is that storage can be repurposed very easily, thus booms and busts may cause temporary oversupply of storage.

Chia isn't worth farming anymore.

I have a pile of old spare 512GB SSDs and it looks like I would able to make something like $2/day. Not even enough for a coffee.

Same with Ethereum. Tried to mine with my Titan V. 31 MH/s. Also not even enough for a coffee.

This is meaningless, and cryptocurrencies are effectively centralized around mining farms with ASICs.

I bought a 512GB SSD for around 60€ some time ago. Let's be generous and assume 40 days pay back. It could be 90 days and it still would be worth it. I don't know how fast SSD wear out when used for Chia though, it could be unprofitable for all I know.

I'm 90% sure as of this week you can get more money by selling the SSD than farming Chia on it before Chia wears it out.

I don't believe that an ASIC has been developed for Ethereum, but I'm not too familiar with cryptocurrency.

Long ago. See the hash rate chart and guess when that might have happened. Hint: long before they were offered for sale.

That's not how chia rewards you though. It's a lottery, and though chia may be around 1000 or so right now it could massively increase in the future. Current projections are something like 3-4 months to get a single reward. But if you could make potentially an extra 50K a year with some harddrives that's not a bad deal.

BitCoin mining is also a lottery. I could stumble on a good hash using my own CPU. But the odds are in favour of the farms with ASICs.

btc can only be mined on asics

That "3-4 months" is going up at an exponential rate, and it's so bad now that by the time 3 months comes by the expected time to reward is 6 months. Even if you get 1 reward, you won't get a second reward unless you continuously keep buying drives and racks for them. You can't really get continuous passive income on a fixed set of hardware.

50K/year is impossible at this point with a consumer rig.

> Current projections are something like 3-4 months to get a single reward

You cant project time to win without knowing how many plots a farmer has. You've gotta have 100+ TB plotted (prolly closer to 140actually) at this point to have a chance at winning every 3-4 months.

A single plot would put you closer to 7 years.

I currently have two plots farming and it looks like this:

  Farming status: Farming
  Total chia farmed: 0.0
  User transaction fees: 0.0
  Block rewards: 0.0
  Last height farmed: 0
  Plot count: 2
  Total size of plots: 202.674 GiB
  Estimated network space: 7357.090 PiB
  Expected time to win: 20 years and 1 month

That's just not accurate. See https://chiacalculator.com/ for more.

3-4 months is currently around 18 TB.

EDIT: 100 TB would get a coin every ~ 18 days on average.

> 3-4 months is currently around 18 TB.

That number does't account for how many more Chia farmers will exist in 1,2,3 months. It doesn't even account for linear extrapolation of total plots in the world, let alone exponential.

Not good indeed. Small hosting businesses and consumer storage will suffer, because big cloud companies usualy have HW production customized and contracted ahead.

And you didn't even touched on the e-waste issue.

Right. Your cloud rents will rise by few cents.

Seriously, anyone compared the capacity of Chia and beloved cloud providers?

"Chia plotting is extremely I/O intensive and destroys most SSDs in under a few weeks"


I’m really not one for conspiracies so I won’t go that far but If I was a drive manufacturer this would certainly be my favorite cryptocurrency.

Hetzner has forbidden Chia yesterday too because the disk are used a lot when chia farming.

Wonder if that’s why their backup storage box performance has been so bad for the last few weeks.

Likely. They specifically mentioned their storage boxes in the announcement; I suspect they had customers using them to store farmed blocks for Chia.

Technically plotting is the really bad part, farming mostly just occupies storage and uses a bit of CPU. Not that I personally am a fan of either, but plotting is the destructive part and then you 'farm' the completed plot forever and that can be done on inexpensive spinning rust drives by a cheap PC.

Can you "plot" at the chip factory (producing some sort of pre-plotted ROM) or does it have to be updated in some way?

A plot is basically a 100GB file (or file) that is associated with your public key. You can move it around how you want. Many people are creating them on machines with nvme disk and then they are copied over a network storage somewhere else.

The plots are each unique, and are too large (hundreds of GB) to represent as a ROM.

Sounds like it's only a matter of time. We didn't have an incentive to create chips like the before, but just like with the specialised mining ASICs, we may see one soon.

if you can build an SSD that's hundreds of GB you can certainly build a ROM...

Flash memory is uniform. Every flash memory die is (hopefully) identical to the next one. Once you know how to produce a flash memory die of a given size, you can produce a million of them, and that's worth a million times more.

Chia plots are unique. Producing two copies of one isn't just pointless, but undesirable; it means that someone else might get a copy and reap the rewards instead of the person who plotted it.

The idea is you plot into a one time programmable thing either at the factory or on first use (so single die, multiple values).

That said it needs to be cheaper than HD, as the SSD is apparently only used to build the plot, so maybe impractical indeed.

Curious that the plotting hasn't already moved to DRAM, apparently 256GB is needed to build a plot which is not too wild on server-class computers.

> The idea is you plot into a one time programmable thing

... like flash memory?

Flash storage has been heavily optimized. You're not going to do any better than that with custom silicon.

> Curious that the plotting hasn't already moved to DRAM, apparently 256GB is needed to build a plot which is not too wild on server-class computers.

My understanding is that people have done the math, and using an SSD ends up being cheaper, even considering the fact that it'll wear out prematurely. (And, in the case of hosting services like Scaleway, the miner isn't concerned -- their hosting provider is the one burdened by this usage.)

> Chia plotting is extremely I/O intensive and destroys most SSDs in under a few weeks

Woah, really?

From what I understand, newer consumer level SSDs are designed to have a lower TBW (TeraBytes Written) - mostly because any reasonable consumer doesn't write terabytes of data to their SSDs in any decent length of time. Enterprise/Higher end "GaMeR" SSDs have higher TBW ratings - I believe 1TB FireCuda Model NVMe SSDs have something insane like a 1800 TBW rating when I was looking a couple weeks back.

These above mentioned consumer drives, when written to in massive amounts (I believe even the lower sized Chia plots thrash the SSD to the tune of at least a couple terabytes of write usage) cause it to degrade at a significantly higher rate.

Long story short, consumer SSDs aren't designed for super high write tasks, while enterprise and higher-end ones are.

> I believe 1TB FireCuda Model NVMe SSDs have something insane like a 1800 TBW rating when I was looking a couple weeks back.

This is still only a couple hundred 100GB plots (the smallest possible size).

"Chia SSDs" being made are something like 12000TBW.

What is the expansion of the acronym “GaMeR”? I’ve not seen it before and I want to be sure I understand.

They're just spongebob texting "gamer".

I think he's making fun of the over the top gamer branding some technology products have.

It's not an acronym, it's the word "gamer" written in something akin to leetspeak [1].

[1]: https://en.wikipedia.org/wiki/Leet

Believe it's use of Mocking Spongebob style https://knowyourmeme.com/memes/mocking-spongebob


its "gamer" like "video game player". not an acronym.

That's incomplete, it turns out; apparently it's a weirdly capitalized form of "gamer" with sarcastic/disrespectful intent towards the referenced group.

Firecudas are terrible for Chia. TBW doesn't matter as much as people think. Some of the lower TBW SSDs will outlast the higher TBW ones (again, Firecuda is an example of the latter).

The standard plot size uses 1.2TB of writes.

You won't wear out your SSD in a few weeks with Chia...you just won't. If you use a 120GB one, maybe...but no-one is doing this.

Yes and no. As I understand it, there are two parts: you create plots which take up a certain amount of disk space, and then farm those plots indefinitely, and only the first plot creation part is extremely I/O intensive. So what a lot of people seem to be doing is creating the plots on SSDs and then moving them to cheaper spinning rust hard drives once created, and it's this that can apparently destroy consumer SSDs relatively rapidly (so long as you have much more storage used for Chia than the capacity of the SSD). I don't think just creating a plot or two in the spare space of your consumer SSD and leaving it there farming away would be a huge problem.

Yes, presumably as a measure to make it harder to scale the plotting process (to make it more "fair"), it apparently does a ton of reads and writes across the whole region being plotted, so it will quickly exhaust the total number of write cycles available on the drive. A cheaper drive will be effectively destroyed (high quality drives can survive more write cycles, so they're comparatively okay.) I would have assumed it just linearly computed the whole plot to be farmed, but presumably that would make it vulnerable to some sort of attack where you create a "winning" plot on demand.

You could theoretically plot on a RAMdisk to avoid this problem, but the necessary amount of RAM would be incredibly expensive, so it's unlikely that many people will do it.

See https://www.youtube.com/watch?v=twwyBdsRYL4

Here's the official Chia answer to this: https://twitter.com/bramcohen/status/1393988621921701889

> For some odd reason the 'Chia burns out hard drives!' is getting repeated as the fashionable fud.

No one is claiming that Chia is burning hard drives. He is intentionally being misleading. Chia absolutely destroys consumer SSDs.

> Chia absolutely destroys consumer SSDs.

If they are used to produce many orders of magnitude more plots than could fit on them. His point is you don't need SSDs to plot, and can reasonably make them directly on the spinning HDD on which you'll store them. I know, because I'm doing this. I make one plot in about 10 hours on the 7200rpm commodity drive in my home media server. It may take me longer to fill up my space than it would with an SSD, but compared to the amount of time in the future I'll be able to farm the plots it doesn't really matter. Cranking out plots as fast as possible probably only helped for the first week or two.

People are burning out SSDs since they're buying into the arms race and trying to get/stay ahead of the netspace growth, hoping for some "easy" XCH, but that's a choice they're making, using poor (or more likely no) ROI calculations and outdated information.

He’s a sellout who premined a massive amount, of course he’s going to lie through his teeth

Yes. The creation of the plot will eat through SSD TBW.

Proof of Space is just Proof of Work by another means. Honestly, I'd rather just have (electric) PoW -- at least the energy source could be green, and any green energy capacity built out for PoW can be repurposed.

I'm sure eventually we'll see articles about the amount of garbage produced by Chia. I'm imagining pallets of dead SSDs.

Extreme cutting-the-last-tree-on-Rapa-Nui vibes

> Chia plotting is extremely I/O intensive and destroys most SSDs in under a few weeks

I need to look into this more, but I think this is an implementation issue. It should buffer in RAM to require only one write pass.

Edit: I was wrong. They set a minimum k that wouldn’t easily fit in ram.

You need a relatively large amount of storage space for plotting, more than most RAM disks. It has proven to be more cost effective to consume SSD's than to use slower HDD's or a RAM disk.

Any idea why the minimum plot size needs to be so damn huge? Really killed the attractiveness of the whole scheme for me

If the plot were smaller there's the risk that someone could calculate plots "on the fly" within the 30-second claim window, thus reducing Chia to proof-of-work (or even breaking it entirely? the "green" paper [1] is a bit too dense for me). It's expected and planned that k will have to increase to 33 in a few years to stay ahead.

[1]: https://www.chia.net/assets/ChiaGreenPaper.pdf

Now you just need terabytes of RAM.

it is possible, the problem is just that you need around 240GiB of temporary memory

I read the headline as China rather than Chia and was wondering what China was plotting to farm now. Frankly it was much more interesting than Chia bad, no Chia for you.

I initially thought it was some agricultural issue similar to palm oil farming but for chia seeds.

> Chia plotting is extremely I/O intensive and destroys most SSDs in under a few weeks

Ok Chia is stupid but this means that you are not pricing your SSD I/O properly.

To me the only explanation of this fad (or scam) is pure greed. That information smog producing period of time will be recalled in history books like the coal race with smog cities of early XX century.

Useful Proof of Storage cryptos like Siacoin or Filecoin should be more incentivised. They store meaningful data.

This reminds me of how Steve Jobs would talk about the need for liberal arts influence in technology. I feel like so much of this blockchain and cryptocurrency stuff is made purely looking at the tech, with either no thought to the actual incentives involved, or a total misunderstanding of them.

There’s no misunderstanding, people are naturally greedy and will try to get away with as much as possible

> This reminds me of how Steve Jobs would talk about the need for liberal arts influence in technology.

I mean, in this case, you really just need engineering (in the traditional sense) influence. Engineering as a discipline covers considering this sort of thing.

Can anyone explain a simple explanation of how chia works? I tried reading the White Paper ("green paper"), but its a little dense and math-y.

What is plotting? How does farming a plot work?

Edit: Why do people plot with SSDs? Is there a speed element to it?

Farming: you hold massive (101GB) bingo cards of mathematical proofs on your hard drive. When you receive a challenge, you first do a pre-filter on your bingo cards to see which of them could contain a proof. Within that filtered list, you find the closest proofs. The peer with the closest proof available wins the challenge and [currently] gets 2XCH.

Plotting: you fill your free space with bingo cards. You need 300-400GB of free temporary space and TBs of writes in that space. This uses a lot of compute and eats SSDs for breakfast.

Once you have filled your space (assuming you don't go out and buy more disks), you can stop plotting. This is where the claimed "greenness" of Chia comes into play: you eventually stop using compute resources. Of course, there are whales who are adding drives to their servers daily.

Can you explain the "bingo card" analogy in a bit more depth?

is this like pre-computing the values to a hash function against different inputs? Where is the write thrashing coming in? If you only need X hundred gigs, why do you write terabytes? Because you can't store intermediary values in RAM (since ram is expensive and small in comparison)?

I get how bitcoin style block-chains use hashing to secure the transaction chain, and the POW factors into hashing... how do responding to the challenges factor into a currency?

> Where is the write thrashing coming in?

This happens during plotting (not farming). The plots are sorted, which is where the write thrashing happens. Once you've completely plotted a farm, it is never written to.

Everything else you said is basically spot-on.

I have 512GB Ram, cant it be used for plotting without ssd?

You could. But someone else could add 8 SSD's for the same money and plot multiple times as fast as you, until the SSDs break.

And what happens if the SSDs break? You lose your plot? Has to be right - since it's proof of space? But what then is the use of burning through SSDs so fast?

the SSDs are used just to create the plots, once the plots are created they can be moved into HDDs where they get farmed (checked to see if they can solve blocks for the blockchain)

It can be, just set up a RAM disk and use that as your temporary drive. It's exactly what I'm doing to fill my tiny 12TB Raspberry Pi farm. Your plotting rate won't be competitive (plotting in parallel on NVMEs turns out to be much faster).

Yeah, I plot an a DL580 with 512GB. Other than not having to worry about write durability there's not much gained though. GB per GB my ram drive is faster, but i can plot in parallel on my NVME drive because it's considerably larger.

You could do 1 plot at a time.

Is there an expected limit of storage where your average drive would die of old age before it yields enough income?

No. Once you are farming, the drive is exclusively read. Of course there is bitrot/cosmic rays flipping bits, effectively rendering some portions of the plot corrupt.

In short:

Plotting is an compute- and IO-intensive process which eventually generates a ~100 GB block of data after a large number of read/write operations to that data. This is the part of the process that involves (and destroys) SSDs.

Farming is a slower process which is performed on those blocks of data, which involves occasional read operations on small sections of that data. The data blocks are typically transferred to slower bulk storage (like hard disks) for this.

If you'd like it in video format, [this one][1] explains it in layman terms quite well and covers possible ways to mitigate the ssd endurance problem, as well as expected profit rates with the right "commodity" hardwares.

[1]: https://www.youtube.com/watch?v=twwyBdsRYL4

>Chia plotting is extremely I/O intensive and destroys most SSDs in under a few weeks

Is this true? Back in 2015, techreport tested a bunch of SSDs to failure and it took 18 months to kill them all. This was a while ago, but I'd been lead to believe write endurance had only improved since then.


If you were an alien species worried about the rise of humans, you couldn't invent a better technology than crypto currency to cause us to kill ourselves.

Not sure how this is a problem since I assumed most cloud providers severely control and throttle IO.

> As you can see, (the scale is in Petabytes), the space used is growing uncontrollably, and has already reached 7 Exabytes.

Judging by the graph in the article, it was 1 exabyte in late April, under a month ago. Wow.

Who buys up all the farmed coins on these new cryptocurrencies.

It is really just speculation fever?

I came across a subreddit, I forget its name though, thats sole purpose was pumping up new cryptocurrencies, hoping it would catch the attention of casual crypto investors. They'd pick a day to sell off whatever cryptocurrency they had pumped up, then repeat it was some other no-name cryptocurrency.

Yes. And miners tend to hold their coins so the supply remains moderately low. A bit like the OPEC.

There's been an absolute metric crap-ton of interest generated since the media has started picking it up and taking about it 2-3 weeks ago. It seems like it's one of the first crypto offerings that relies not on constant CPU/GPU usage, but just proof of "storage" from what I understand.

Very interesting.

I wonder if this is why they are also phasing out their C1 bare metal instances - other than being old I bet it would be easy to burn through those disks for Chia coin.

Is it possible to optimize plotting to use I/O less and RAM more? Or is it an inherent boundary of the protocol like energy use is in PoW?

The reference implementation is Python, so there's lots of room to optimize. I wonder if you can eventually build an ASIC that replaces the disk based lookup with in memory calculations of the same data.

Can you send a link? I'd like to help.

Most of the work happens in C++[1]

[1] https://github.com/Chia-Network/chiapos

If you have ~300 GB of RAM you should be able to plot directly in RAM.

Imagine a cloud provider has 300GB RAM. What if they plotted and let other people later download the plot?

I think bandwidth charges kill that idea.

On AWS, yes.

On Scaleway and Hetzner, there is unmetered 100Mbps bandwidth.

On Linode, Vultr, and DigitalOcean, there is a 1-month 1TB limit.

Oh no! Regular people are using products for something legal that's not approved by "Hacker" "News"? Petabytes of storage capacity for decentralized storage network? That's obscene. Clearly that storage capacity is meant only to expand exabyte storage data centers of big tech. Only socially responsible way to store data is to rent capacity on AWS.

TIL people are competing for who's got the biggest hard drive...

Oh you have no idea how far it goes https://www.reddit.com/r/DataHoarder/

"it's a digital disease"

the time to farm chia was three months ago

You couldn't farm it three months ago. Mainnet opened in mid-March.

is it not feasible to plot on AWS? would be interesting if they are also forced to ban this.

Did some, transfer costs beetween instances that serves as plot generators are f** expensive. You need TB of data transfers, for example 15TB to transfer on the internet is somewhere around 1500$.

AWS/Azure/GCP are significantly more expensive compare to these providers.

This is mostly a problem for low-cost hosting that had some implied assumptions about how heavily used various components would be relative to the rest.

From what I understand, the network/algorithm is designed to prevent this from happening - either due to massively huge egress costs (I think you need to 'upload' the whole plot to the network after you have a verified proof), or for sheer data storage cases (some farmers/miners have tens of petabytes of plots already - i wonder how much that would cost to store monthly?)

You don't have to upload the plot after you have a verified proof. You do have to pay to store it, though, and that can be pretty expensive on AWS. If you just plot on AWS and store the plots elsewhere to save storage costs, you have to pay outbound charges on the plot files.

Seems fair.

Chia plotting is forbidden on all SSD and NVMe powered instances

Translation: Scaleway didn't take SSD wear into account when designing their cloud.

Chia farming is allowed on Scaleway Object Storage providing prior request has been both made and authorized by our sales team

Translation: They can't expand their object storage fast enough to keep up with demand.

> Translation: Scaleway didn't take SSD wear into account when designing their cloud.

I don't think many cloud providers expect all their disks to fail every 1-3 months, which is what happens when you mine Chia on a SSD/NVMe.

Yup, all cloud (well, all rental) business relies on rented resources to not be 100% utilised. Oversubscription is how they can offer prices cheaper than buying the resource yourself.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact