Hard-drive Prices Have Gone Crazy

[KultiVator]

To give some sense of how much the Chia landscape has changed in just a few days...

This time last week, a 14TB drive full of plots was projected to have a probability of 'winning the Chia lottery' every few days (more than once per week).

As of right now, the same quantity of plots aren't expected to yield a 'win' until somewhere between 10 and 40 months!

The exponential growth of the network means most solo farmers are like cheetahs chasing a photon.

Still, it's given a lot of optimistic people an excuse to upgrade their home rigs before reality bites them ;-)

[DanceswithUnix]

It sounds like in a week you will be able to use a mining pool, which should get home mining a steady income.

Not that home miners will make any real dent on the industrial sized people, as per usual.

I had wondered if this would pay for me setting up a personal cloud in the garage, but the way things are looking I haven't bothered. Specially with the tax implications of going over £1000 in mining income.

[KultiVator]

It sounds like in a week you will be able to use a mining pool, which should get home mining a steady income.

Yeah - saw that the Chia protocol extensions to allow pooling are due next week. Problem for me is that this feels like a scummy use of hardware, power, etc and encourages more "me too" crypto-currencies to spring up + even more stupid shortages / price increases for PC components.

Might change my mind if my tiny set of test plots generate a win against all the odds ;-)

[DanceswithUnix]

Might change my mind if my tiny set of test plots generate a win against all the odds ;-)

lol, well there is that.

It is a rubbish use of hardware. I can see me putting something together from old spares in the garage where I have enough old enterprise 1TB drives to make a small ZFS array and see what sort of payout pooled mining gives. I don't see me buying storage for it though.

I don't see anything this does that eg Stellar does for a fraction of the hardware and energy cost other than stir up a pointless mining craze. Not that I have any moral objection to sucking money out of such a scheme, just have to make sure you aren't one of the people that lose in such a craze.

I did notice there are people selling plots at about $16 each. That's kind of amusing, though given the size I think it would be better to just calculate one locally.

[Saracen999]

Just like the knickers of the individuals in the phrase you can't use....?

Exactamently.

[Corky34]

Speaking of Chia, i found this der8auer video interesting. In that it's not so different than other mining efforts using computer hardware.

[Rob_B]

I watched a bit of that yesterday, seems like a great way to waste storage space

[KultiVator]

Speaking of Chia, i found this der8auer video interesting. In that it's not so different than other mining efforts using computer hardware.

That video reinforces my own reservations - having monitored parallel plotting I can vouch for Der Bauer's reservations about SSD thrashing and TBW (Total Bytes Written).

That said - his ThreadRipper rig is nice... but I'd still find better uses for that hardware ;-)

[DanceswithUnix]

That video reinforces my own reservations - having monitored parallel plotting I can vouch for Der Bauer's reservations about SSD thrashing and TBW (Total Bytes Written).

That said - his ThreadRipper rig is nice... but I'd still find better uses for that hardware ;-)

I haven't watched it, but I would have assumed if you can afford a Threadripper then you can afford enough ram that the whole thing is done in cache. He is using Linux, right?

Sometimes it is better to tile a problem into memory and knock out lots of them serially than do stuff in parallel.

[CAT-THE-FIFTH]

I haven't watched it, but I would have assumed if you can afford a Threadripper then you can afford enough ram that the whole thing is done in cache. He is using Linux, right?

Sometimes it is better to tile a problem into memory and knock out lots of them serially than do stuff in parallel.

You need around 260GB of SSD cache to do one plot,and its more efficient to plot simultaneously. So you would need something like 1TB of RAM.

That video reinforces my own reservations - having monitored parallel plotting I can vouch for Der Bauer's reservations about SSD thrashing and TBW (Total Bytes Written).

That said - his ThreadRipper rig is nice... but I'd still find better uses for that hardware ;-)

To wear out a 2TB Phison E16 NVME SSD(generally have around 3600TBW),you would need nearly 200TB of HDD storage capacity.

Many enterprise SATA,write intensive drives easily have over 8000TBW in 2TB sizes,and many well over 10TBW. So you would need 100s of TB of HDD storage to wear such SSDs out. The main limitation is going to be the HDD storage capacity you have,and the cost of 100s of TB of storage,which is easily much higher than a 2TB Enterprise SSD. Then the space requirements to have dozens of HDDs,especially if you run sort of redundant RAID.

[badass]

I've probably contributed to the storage problem unintentionally. Heard about Chia and decided I'm not getting stuck on not enough storage for ages because of crypto (I'm itching to change my gfx card but refuse to do so at current prices)
The problem is that I went to order a pair of WD D10 8TB disks as they are DC drives inside and would make a decent capacity mirrored pair. During the ordering process the stock disappeared and it looked like 5TB D10's were available so I just selected those. Turns out they are P10's and those are 2.5" drives with the USB interface soldiered directly to the drive. I've now ordered the D10 8TB for £40 more than they were on Amazon so I can just shuck and plug into my server. The problem is that I now have the P10's and seeing how much income could theoretically be made through Chia I might not return the drives :-S

Maybe CHia income can pay for an overpriced Graphics card so I'm sort of back at square one!

[badass]

I've just chucked a spreadsheet together. It looks like Chia storage growth is increasing 10 fold appx every 24 days or 10% per day.
Should the price remain stable and the growth continue, after 1 month, daily profit is 5% of what it is today. I believe that's less profitable than buying a scalped graphics card and mining ether.
Another way of looking at is is take todays daily profit and multiply by 10. Expect no more profit from that storage. It shrinks too quickly.
I think I'll send the 5TB HDD's back!

[CAT-THE-FIFTH]

I've probably contributed to the storage problem unintentionally. Heard about Chia and decided I'm not getting stuck on not enough storage for ages because of crypto (I'm itching to change my gfx card but refuse to do so at current prices)
The problem is that I went to order a pair of WD D10 8TB disks as they are DC drives inside and would make a decent capacity mirrored pair. During the ordering process the stock disappeared and it looked like 5TB D10's were available so I just selected those. Turns out they are P10's and those are 2.5" drives with the USB interface soldiered directly to the drive. I've now ordered the D10 8TB for £40 more than they were on Amazon so I can just shuck and plug into my server. The problem is that I now have the P10's and seeing how much income could theoretically be made through Chia I might not return the drives :-S

Maybe CHia income can pay for an overpriced Graphics card so I'm sort of back at square one!

I got one of the D10 8TB drives last week and they WD HC320 8TB enterprise drives.

[KultiVator]

To wear out a 2TB Phison E16 NVME SSD(generally have around 3600TBW),you would need nearly 200TB of HDD storage capacity.

One subtle detail I’m not sure everyone has picked up on... the amount of data written per plot is greater than the size of the temporary files created on your SSD. There appears to be a recursive aspect to the plot-making process concealed by the headline plot file sizes being quoted. From memory, I believe Der Bauer measured the real figure as being 1.4TB per plot! That really puts a finite lifetime on even a 3600 TBW SSD. Then there’s the high power draw throughout the plotting process and the 24/7 ‘farming’ which you have to keep up to remain synced with the blockchain and an active player ‘in the Chia lottery’.

I don’t fancy the odds with Chia... but might buy shares in Seagate and the power grid instead :-D

[spacein_vader]

I don’t fancy the odds with Chia... but might buy shares in Seagate and the power grid instead :-D

Not a bad idea. The best way to make money in a gold rush is selling picks and shovels. Next best would be buying shares in those that do I guess.

[DanceswithUnix]

I've just chucked a spreadsheet together. It looks like Chia storage growth is increasing 10 fold appx every 24 days or 10% per day.
Should the price remain stable and the growth continue, after 1 month, daily profit is 5% of what it is today. I believe that's less profitable than buying a scalped graphics card and mining ether.
Another way of looking at is is take todays daily profit and multiply by 10. Expect no more profit from that storage. It shrinks too quickly.
I think I'll send the 5TB HDD's back!

The need to compute plots should make it more linear than that, but it is still an arms race for who can grow their number of plots the fastest. In my book that instantly kills any green credentials it might try and claim. If the network right now is fine, why should there be any incentive to massively expand it with mining capacity?

If you have a good internet connection, then you could use those hard drives to farm Storj. That is basically being paid to store people's backups, and AFAIK can be done with a USB drive and a Raspberry Pi.

From memory, I believe Der Bauer measured the real figure as being 1.4TB per plot!

I got to watch his video, the cloud compute thing was a bit iffy. He was locally using Windows, and from what Son of a Tech was saying on his channel if you switch to Linux you get a 15 to 20% speed boost (which makes sense given the better block caching in Linux). So Der Bauer switched to a ramdisk which being on the cloud forced him to use Linux and at that point he was going 10% faster than on his local machine. So that's actually slower. I assume Linux gets the performance from optimising away some of the ssd writes, but even if 20% of them are optimised out that's still over 1TB per plot.

For the farming stage, it sounds like people are able to do that on the cpu power of a consumer NAS box. You might want to add Qnap or similar shares to your mix