• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

This is a reason NOT to mine Chia...

Chia was initially marketed as the crypto where you can use your spare hdd space to make some money but the reality it turns out people just emptying shelves of bestbuy HDD and clearing all newegg/amazon stocks of them. It also become an industrial size operation now. The Chia network space current at 10.44EIB have already surpassed backblaze servers in total storage use. It storing nothing but just "lottery ticket" numbers. It sadden me to see hdd being filled with literally meaningless coupons. This was promoted as a "green crypto" but looking at the hdd consumption and the burning up of ssds is anything but green. Add to it to plot faster aka mine faster you need a cpu with alot of threads and a high number of ram space to do it quickly.

People used ssd to mine Chia so they are able to mine faster at the expense of ssd life, it better to do it on old rust spinners despite that slower mining speed at least you don't burn up precious ssd writes. Anyway the pandora box for Chia is open now, the HDD hoard and SSD hoard is well underway now. Currently most Chia miners are buying are nvme ssd for their speed but I am afraid its a matter of time before sata based ssd would be affected too.


Yeah, I always thought HDDs would be better for this but aparently access times effect your earnings too.

Just seems messy and nonproductive all around.

Yeah I agree but many people used as SSD in hope of faster profit to catch up with the growing network. It take about 1Tib to write one plot of Chia the size of 101Gib. Judging by the Chia network size now of 10.44Eib, that means at the worse case 104Eib of SSD writes are used. Damn that is tons of ssd burned up.

Edit : You can plot 4 to 5 times more with SSD with a HDD.
 
Greed distorting a utopian vision is all too familiar in crypto land, I'm afraid.
 
I think of it otherwise... ie what kind of trash we are fed to actually... QLC+, smaller tech nodes etc etc... it just milking and making more profit not making better products, the stupid burst speed that does nothing in real life... like netburst style for more MHz, popcorn useless ones.

NAND ain't that durable... it's been proven during time, especially in mobile phone tech... for example anyone using torrent on their phones is a plain stupid idiot, you don't even need Chia to kill a NAND, well it takes usually 6 months to exhaust overprovisioning area and overprovisioning areas ain't that big there. I see dead memory IC's each day... with all data permanently lost.

Basically the TWB numbers are a thing you should be aware off and they tend to drop. I look at them really skeptically for newer IC's. Manufacturers clearly say, more bits in the stack, the less endurance it will have. They are kinda racing towards the wrong end instead of reliability they go for profit and appeal.

I imagine flooding the gray market with those dead drives with overwritten SMART to fool the client. Manufacture date would be fresh...
 
Last edited:
I think of it otherwise... ie what kind of trash we are fed to actually... QLC+, smaller tech nodes etc etc... it just milking and making more profit not making better products, the stupid burst speed that does nothing in real life... like netburst style for more MHz, popcorn useless ones.

NAND ain't that durable... it's been proven during time, especially in mobile phone tech... for example anyone using torrent on their phones is a plain stupid idiot, you don't even need Chia to kill a NAND, well it takes usually 6 months to exhaust overprovisioning area and overprovisioning areas ain't that big there. I see dead memory IC's each day... with all data permanently lost.

Basically the TWB numbers are a thing you should be aware off and they tend to drop. I look at them really skeptically for newer IC's. Manufacturers clearly say, more bits in the stack, the less endurance it will have. They are kinda racing towards the wrong end instead of reliability they go for profit and appeal.

I imagine flooding the gray market with those dead drivers with overwritten SMART to fool the client. Manufacture date would be fresh...
The only real longevity advance that was made recently was 3d stacking. Beyond that, it's been a downhill roll, yeah...

Used to be I'd only buy mlc drives. Been forced to tlc now unfortunately...
 
Any pcie4 ones? That's what's getting me.

You need that pcie4 ? I would take older 970Pro not the renamed evo drive that calls himself 980Pro. MZ-V7P1T0BW (1,024 GB) 5 Years or 1,200 TBW vs MZ-V8P1T0BW (1TB) 5 Years or 600 TBW...

As for my main drive I am happy that I didn't cheap out on my Optane drive, especially looking now at GPU prices, it looks peanuts. My mere 5.11 PB write limit... and it is plain stupid fast no matter how full.

Shame the tech ain't developing much further. I am considering even picking up Intel Optane 905P 380G SSD M.2 NVME 22110, while it is still obtainable.
 
You need that pcie4 ?
I don't NEED any of it but if I was to upgrade, it'd be nice, yeah. It'd also use less lanes than my current RAID card for same performance in burst speed, so...

I mean I don't even know why I'm asking really. I have to have OPAL encryption for my document handling at work, so that makes my options extremely limited. /sighs.
 
I don't understand the coin but I have read they are meant to be releasing out very high TBW but I still don't really get it, the point of it at all.... I'm not sure why you'd want to run it... But its very frustrating considering I was toying with a server drive upgrade and then this happened lol

Ah well :D
 
You need to use server class which can survive stuff like 2-3 full HDD rewrites per day. And those are a bit expensive. Also good deal more reliable and sorta faster than regular ones.

Chia is interesting concept, but my HDDs are just too precious to me.
 
You need to use server class which can survive stuff like 2-3 full HDD rewrites per day. And those are a bit expensive. Also good deal more reliable and sorta faster than regular ones.

Chia is interesting concept, but my HDDs are just too precious to me.
You can also go with used enterprise SSDs, in particular Intel DC S3700 is a great choice. An array of that costing maybe $400 US will support a farm as large as 2 PB.

If these old drives are nowhere to find (might be all grabbed), new enterprise drives like "Samsung PM983 NF1" aren't crazily expensive:

Though, probably lots of people won't be doing research and just keep buying consumer ones (not even buying those 1TB 1.8 PBW ones, but the 0.6 PBW Samsung junk) as they are cheaper short term.
 
You, guys, forgot one important thing: there is a big difference between plotting and farming.
Plotting is what kills your SSD, and plotting isn't as "lightweight" and "green" of a task as developers lead us to believe, but it only needs to be done once for each of your plots. Rich people use dedicated Xeon Platinum or Threadripper machines with tons of RAM for plotting, in addition to large arrays of NVME SSDs, though it's also possible to plot on HDD itself(only a loo-o-o--o-o-t slower).
Farming, on the other hand, only needs a whatever 4T CPU with lots of HDD storage, and consumes significantly less power (even HDDs are only spinning while you have a block to check against your plots, but 99% of the time remain dormant).
So, if someone decides to get into Chia thing, all he or she needs to do is open their eyes, turn on their brain(or at least a tiny portion of it), and RTFM. You can even guesstimate your TBW based on how much you want to plot, and figure out approximately when you'll need to get a spare drive. Plus, most high-tier SSDs can outlast their rated TBW by a large margin (SLC/2LC, or even enterprise TLC).

Just to put it in perspective, I'll use some numbers from my last experiment. At most I can generate around 2 plots simultaneously on a 1TB drive. I've only done sequential plotting due to space constraints, but if I can do ~2-3 plots per day sequentially, let's just assume that you can do 6 plots (~0.6TB) by doing a pair of plots at a time. So, if you have a 1TB TLC drive - you'll be able to exceed rated TBW in approximately 2.5-3 years (which is incidentally a most common warranty for low-to-mid-range TLC). For QLC it's theoretically the same if not more, since your write speeds are gonna fall so low, that you won't be able to generate more than 1/2 plot per day (which in terms of writes is even less than its rated DWPD). On a HDD it's even slower.
People have been actively plotting chia only in the past few months, so if someone's SSD died within this short of a timeframe(even QLC) - that's 99% manufacturer's fault and maybe 1% user fault due to some ridiculous operating conditions.

There are also some articles that use random garbage off reddit as a source of information, like quoting 1.5-2TBW per plot, which is f#$%ing ridiculous.
I've generated over 40 plots on my SSD, and here's what my S.M.A.R.T. looks like:

SX8200.PNG
This drive is more than 1.5y.o., and it's also my primary storage, and it suffered through a few dozen of agonizing 150+GB rsyncs from one of my work servers (lots of small- and medium-sized files) and gathered most of its TBWs thanks to bloated modern games, yet for some reason I don't see an extra 80TBW on top of what I had before, if that sensational number was actually true. Realistically your TBW per plot is the same or a little higher than the size of temporary files for each plot (e.g. 250-300GB). This means that my 40 plots have accumulated no more than 12TBW, and not that scary 70-80TBW according to reddit numbers (or whatever source those stupid idiots from tech news sites are quoting).
 
I was expecting something like this will be available, I hope it isn't too expensive lol

That would be a senseless waste of money...
Not a waste comparing to killing off multiple of consumer SSDs which likely would have plotted less; it's 21 PBW vs ~3-5 PBW of consumer SSDs of similar volume, at a similar price.
But then yeah you may have to think twice before investing into Chia at its current state.

You, guys, forgot one important thing: there is a big difference between plotting and farming.
Plotting is what kills your SSD, and plotting isn't as "lightweight" and "green" of a task as developers lead us to believe, but it only needs to be done once for each of your plots. Rich people use dedicated Xeon Platinum or Threadripper machines with tons of RAM for plotting, in addition to large arrays of NVME SSDs, though it's also possible to plot on HDD itself(only a loo-o-o--o-o-t slower).
Farming, on the other hand, only needs a whatever 4T CPU with lots of HDD storage, and consumes significantly less power (even HDDs are only spinning while you have a block to check against your plots, but 99% of the time remain dormant).
So, if someone decides to get into Chia thing, all he or she needs to do is open their eyes, turn on their brain(or at least a tiny portion of it), and RTFM. You can even guesstimate your TBW based on how much you want to plot, and figure out approximately when you'll need to get a spare drive. Plus, most high-tier SSDs can outlast their rated TBW by a large margin (SLC/2LC, or even enterprise TLC).

Just to put it in perspective, I'll use some numbers from my last experiment. At most I can generate around 2 plots simultaneously on a 1TB drive. I've only done sequential plotting due to space constraints, but if I can do ~2-3 plots per day sequentially, let's just assume that you can do 6 plots (~0.6TB) by doing a pair of plots at a time. So, if you have a 1TB TLC drive - you'll be able to exceed rated TBW in approximately 2.5-3 years (which is incidentally a most common warranty for low-to-mid-range TLC). For QLC it's theoretically the same if not more, since your write speeds are gonna fall so low, that you won't be able to generate more than 1/2 plot per day (which in terms of writes is even less than its rated DWPD). On a HDD it's even slower.
People have been actively plotting chia only in the past few months, so if someone's SSD died within this short of a timeframe(even QLC) - that's 99% manufacturer's fault and maybe 1% user fault due to some ridiculous operating conditions.

There are also some articles that use random garbage off reddit as a source of information, like quoting 1.5-2TBW per plot, which is f#$%ing ridiculous.
I've generated over 40 plots on my SSD, and here's what my S.M.A.R.T. looks like:

View attachment 201988
This drive is more than 1.5y.o., and it's also my primary storage, and it suffered through a few dozen of agonizing 150+GB rsyncs from one of my work servers (lots of small- and medium-sized files) and gathered most of its TBWs thanks to bloated modern games, yet for some reason I don't see an extra 80TBW on top of what I had before, if that sensational number was actually true. Realistically your TBW per plot is the same or a little higher than the size of temporary files for each plot (e.g. 250-300GB). This means that my 40 plots have accumulated no more than 12TBW, and not that scary 70-80TBW according to reddit numbers (or whatever source those stupid idiots from tech news sites are quoting).
At most I can generate around 2 plots simultaneously on a 1TB drive
Quite certain you can do 3, it takes at most 260 GB (239 GiB) for one plot.
Realistically your TBW per plot is the same or a little higher than the size of temporary files for each plot (e.g. 250-300GB).
I believe this isn't true, generating a plot it needs to do 1.7 TB write if memory serves.
PS: unless this has changed, otherwise
One k32 writes 1.8TiB in non-bitfield mode and 1.6 TiB with bitfield enabled.
 
Last edited:
Not a waste comparing to killing off multiple of consumer SSDs which likely would have plotted less; it's 21 PBW vs ~3-5 PBW of consumer SSDs of similar volume, at a similar price.
But then yeah you may have to think twice before investing into Chia at its current state.
No, it's still a waste...
 
Quite certain you can do 3, it takes at most 260 GB (239 GiB) for one plot.
I think on several of my attempts to do 2-3 plots simultaneously I ran out of space. Maybe I misconfigured my delays and completed plots weren't clearing up fast enough for new ones to start.

One k32 writes 1.8TiB in non-bitfield mode and 1.6 TiB with bitfield enabled.
There's conflicting info, cause in one of their other articles they've mentioned something along the lines of 360GB per plot, while in the beginning of this article they start with 1.3TBW per plot etc. etc. etc.
Otherwise math doesn't add up: 40 plots x 1.8TBW = 72TBW, which is approximately the same as my grand-total TBW, including 1.5 years of excessively-active usage. Pretty sure I had at least 40-50TB clocked even before CHIA. If I manage to get another HDD, I'll do some tests just to be sure.
 
I think on several of my attempts to do 2-3 plots simultaneously I ran out of space. Maybe I misconfigured my delays and completed plots weren't clearing up fast enough for new ones to start.


There's conflicting info, cause in one of their other articles they've mentioned something along the lines of 360GB per plot, while in the beginning of this article they start with 1.3TBW per plot etc. etc. etc.
Otherwise math doesn't add up: 40 plots x 1.8TBW = 72TBW, which is approximately the same as my grand-total TBW, including 1.5 years of excessively-active usage. Pretty sure I had at least 40-50TB clocked even before CHIA. If I manage to get another HDD, I'll do some tests just to be sure.
It will be interesting to actually test, so far all my calculations are based on the 1.7 TB per plot assumption; I will take it as a worst-case scenario for now.
 
I generally hate crypto but chia is another level of stupidity. You make plots and hope that one day you will get chosen and paid. That's like digging all around the earth in plots hoping to find a plot with hidden gold. Plus the fact that is destroys hard drives at record pace thus creating a lot of e waste.
 
Last edited:
Back
Top