What'll happen to ASICs when they're no longer profitable to run?

[odolvlobo]

The idea of using low-powered unprofitable equipment to "support" or "protect" the network is flawed. Hardware that is no longer profitable to use for mining will probably not have a significant contribution to the total hash power of the network.

Imagine if all the GPUs started hashing again. They would contribute at most 25 TH/s, or less than 5% of the current total hash power.

[1714924022]

1714924022

[1714924022]

1714924022

[DeathAndTaxes]

Who would want to buy MORE expensive chips?

People who would not make any money mining 28nm devices.

Just because you bought 28nm pre-order, it does not mean every ASIC designer out there will not try anything that is more efficient than your pre-order.  Looks like you are trying to convince yourself that your purchase of hashfast will not become obsolete for a very long time.

The ASIC train is moving faster than anyone can say: WTF just happened...

Avalon already said they will be working on Gen 2, 3 and 4 (at the same time?).  Are you ready for your WTF moment?

More expensive means higher MH/$.  Why would anyone want that.  I even pointed out it is likely there will be more efficient 28nm chips but given the higher cost of 20nm chips a given design (any design of any efficiency) would cost MORE not LESS at 20nm.  Once again why would anyone want MORE expensive chips?

[erk]

The idea of using low-powered unprofitable equipment to "support" or "protect" the network is flawed. Hardware that is no longer profitable to use for mining will probably not have a significant contribution to the total hash power of the network.

Imagine if all the GPUs started hashing again. They would contribute at most 25 TH/s, or less than 5% of the current total hash power.

It would be closer to 50TH/s I think. If you add up the hash rates of the scrypt coins on Coinchoose, and then multiply it by 1000 for the approx scrypt to SHA256 conversion rate, you will see what I mean. That's still only 10% of the net hash.

GPU's are not a good example, as they chew a lot of power, it's probably more relevant to talk about the 110nm Avalon and ASICMiner chips out there. No ROI on batch #3, Block Erupters etc. but 110nm represents most of the current net hash. There is nothing wrong with the 110nm ASIC tech except it was sold at a ripoff price to early adopters. That stuff could easily be pumped out below $10 per GH/s. to make it competitive again.

[SaltySpitoon]

Someone who is heavily staked in low GHash ASICs will try and create a new SHA Alt coin, to attract all of the people who were left behind by the next gen asics. Same thing happened with GPUs and 1st Gen asics. The Scrypt Alt coin craze began to draw in all of those BTC Miners who didn't want to sell off their GPUs at next to nothing to buy promises of ASICs that would come in a year.

[erk]

Someone who is heavily staked in low GHash ASICs will try and create a new SHA Alt coin, to attract all of the people who were left behind by the next gen asics. Same thing happened with GPUs and 1st Gen asics. The Scrypt Alt coin craze began to draw in all of those BTC Miners who didn't want to sell off their GPUs at next to nothing to buy promises of ASICs that would come in a year.

That will probably work to some extent, if you look at the profitability charts, you will see a LOT of scrypt based alts that will get you more than twice the BTC per day that just mining BTC will. The only main offerings on SHA256 alts have been PPC and TRC, not really successful, but ZET is looking ok, it 143.54% BTC on Coinchoose atm.

ZET just listed on Cryptsy a few hours ago, so it's still finding it's market price. If you are going to mine it don't dump insane amounts of ASIC onto it and screw the block chain like TRC, bring it up gradually if you want to preserve the price of the coin.



   

[DeathAndTaxes]

There is nothing wrong with the 110nm ASIC tech except it was sold at a ripoff price to early adopters. That stuff could easily be pumped out below $10 per GH/s. to make it competitive again.

However the smaller stuff can be pumped out at an even lower price point.  Maybe $2 per GH/s.  The raw silicon costs have the potential to be the minority of the overall system. Take HF chip. 19mm x19mm @ 28nm.  If you assume $10K per 300mm wafer it works out to ~$0.15 per GH/s.  Granted you still have testing, cutting, packaging, etc so lets say $0.25 per GH/s. However say HF (or Bitfury or KNC) decides to just sell raw chips in bulk @ $2 per GH/s and system builder builds a rig for another $1 per GH/s. and marks it up a buck for say $4 per GH/s.  That is getting close to the build cost @ 110nm.  Now consider that even if you could buy a rig from either process for $4 per GH/s.  Why would you buy then one which uses almost 12x the power for the same hashrate.

The good news is that ASICMiner and Avalon should have plenty of cash to tape out a smaller process chip, something in the 28 to 55 nm range.

[Takeshi_Kovacs]

For sure we will see some 20nm next year

 IMO, while it should be technically possible, gut-check says I don't see it happening for Bitcoin ASICs in 2014 due to excessive NRE costs. If we're extremely lucky, we might hear about something near Christmas 2014, but more than likely 2015 before being able to own physical 20nm ASICs.

Exactly.  I am not saying 20nm is impossible but for it to make sense with Bitcoin it would have to have a lower cost per transistor than 28nm that isn't projected to happen until 2015 or later.  Even when the cost per transistor reaches parity with 28nm you still have the issue or a brand new NRE.  The cost will need to decline substantially so that even with NRE 20nm is cheaper to make than 28nm.   That isn't happening in 2014, maybe 2016.

20nm being "available" means nothing if it can't be priced under 28nm miner tech.  Nobody is going to pay a premium on new miners when ROI% are stretched out to 2+ years due to massive over capacity just to be cutting edge.  Proven, highly available, "good enough", and cheap 28nm will be the name of the game for 2014.

You have missed something rather important there.

A smaller transistor will switch faster and use less energy to switch than a larger transistor. So, a straight reduction of an existing design from 28nm to 20nm ought to give more hashes for fewer watts and therefore it would be a preferred technology even if the transistor prices were higher.

[DeathAndTaxes]

Except in reality probably not.  Miners have generally not preferred technology which has a higher capital cost and marginally lower operating cost.  I don't see that changing anytime soon.   A company dumping millions (plural) into 20nm NRE knowing their marginal cost of production is higher than 28nm counterparts would be taking a massive risk.  The only way to put that risk on consumers would be through pre-orders.  I don't really see pre-orders that cost MORE (per MH) than available gear being popular.

If today you could get a miner from one company at $10 and 1 W per GHash with delivery tomorrow OR you could pre-order one from another company for $15 and 0.5W per GHash with delivery in 60 to 75 days would you really pick the latter?

[Stinky_Pete]

If today you could get a miner from one company at $10 and 1 W per GHash with delivery tomorrow OR you could pre-order one from another company for $15 and 0.5W per GHash with delivery in 60 to 75 days would you really pick the latter?

Maybe. My electricity is expensive. New tech is cool.

[Ente]

They will be replaced by more energy efficient ASICs or will be run in areas where electricity is dirt cheap.

Once we reach technological brick wall (10nm), the network hash rate will level off and difficulty might actually stay flat or go down.
At that level, mining=electricity & hosting costs + a tiny profit.

The economic brick wall is 28nm.  20nm is beyond rediculously expensive for anyone who doesn't know their own fab (Intel, Samsung).  Nobody is predicting high availability or decent prices of 20nm wagers from any foundry in 2014 either. Nobody (not even Intel) is using 14nm yet, the technical challenges of 10nm haven't even been addresses.   Still even if 20/14/10nm is techincally available what matters in mining is the cost per hash and Bitcoin mining is almost perfectly parallel.  This means a 20nm chip is going to cost MORE than a 28nm one (everything else being the same) until the cost per transistor is lower on 20nm then 28nm.   Usually that takes 2-4 years from process start and it is taking longer and the end benefits are getting smaller (higher NRE, more delays, more complexity, more yield issues, and overall a shrinking gain on the prior process node).  So 20nm SHA-2 ASICs probably won't make sense even in 2015 or 2016 and 14nm isn't even on the map yet (2020?+).



That means for the convieable future 28nm is as good as it gets.  There may be more efficient 28nm designs but that would mean a brand new NRE cost and it would have to be a lot better to make sense (i.e. a company's new chip would need to be so much cheaper (MH/$ and MH/W) that even after a share of the NRE it added it is still a better deal that then existing design).  It remains to be seen if that will happen.  Regardless once everyone is on 28nm and shipping in volume, the market will become highly saturated and mining margins will be a small percentage over electrical cost of the most efficient miners.  Have a less efficient miner, have higher than normal electrical costs, paying for expensive datacenter space expect it to only be a question of how long before your net operating margin goes negative.  Really no different than GPU mining.  Those with excessive electrical costs or poorly though out rigs could only profit marginally when the price/difficulty was high and when it tanked they had to idle or mine at a loss.

Highest quality post in a good while.
Thank you, DnT.

Ente