What are the limits of ASIC Mining?

[locksley]

My assumption is that ASIC miners optimize for the "hashes per joule" metric, as did GPU miners over CPU miners.

It would seem that after some time, the decision to mine Bitcoins would be based on if you could source energy cheaply enough.

When do we reach that point?

I'm not a hardware guy so I'm curious to understand what the upperbound limits are of ASIC mining from a technical perspective. How much more can we optimize?

Thanks!

[1715403610]

1715403610

[1715403610]

1715403610

[1715403610]

1715403610

[1715403610]

1715403610

[1715403610]

1715403610

[1715403610]

1715403610

[DeathAndTaxes]

It is hard to say exactly but most of the improvement (at least in the short term say next 2-3 years) in efficiency is already behind us. The ASIC designers are probably the only ones who have a good idea how much more efficiency they can squeeze out of a design but my guess is there is not that much less.  For a given process node maybe 3x or 4x current performance.   That brings the second factor.  When you shrink the process node (55nm -> 40mm -> 28nm -> 20nm -> 16nm) in general processors gets cheaper,faster, and cooler.  The compound effect of Moore's law.

Bitcoin kinda had a hyper Moore's law effect jumping almost a decade in process node improvements in less than a year.  The first ASICS were at 130nm we are now at 28nm with 20nm being worked on.    That is pretty much "state of the art".  So while slowly chips will move to smaller process nodes (usually a new node every 2 years) we aren't going to see another jump like we did from 130nm to 28nm in a year ever again.

GPU 300 to 500 J/GH
First gen ASICs 10 J/GH
Current best shipping ASICs 0.5 J/GH
Next gen ??  maybe 0.2 J/GH to 0.4 J/GH (improved design @ 20nm)

The gains are going to get smaller and smaller.   

[cp1]

I'm sure there are old Asics that aren't profitable now, depending on your electricity costs.  But I'm sure a lot of miners aren't concerned with profitability now, but how the price will go up later.

[helipotte]

There will eventually come a time when an ASIC (miner) cannot be manufactured for less than it should sell for to be profitable (for the buyer). At this time hardware makers will

stop selling hardware.  They might still make miners, but will probably mine with inventory themselves at a location that has cheap or free power.  Eventually only people with

free power will still mine for profit, and that profit will be very slim.

[JayRobo]

It is hard to say exactly but most of the improvement (at least in the short term say next 2-3 years) in efficiency is already behind us. The ASIC designers are probably the only ones who have a good idea how much more efficiency they can squeeze out of a design but my guess is there is not that much less.  For a given process node maybe 3x or 4x current performance.   That brings the second factor.  When you shrink the process node (55nm -> 40mm -> 28nm -> 20nm -> 16nm) in general processors gets cheaper,faster, and cooler.  The compound effect of Moore's law.

Bitcoin kinda had a hyper Moore's law effect jumping almost a decade in process node improvements in less than a year.  The first ASICS were at 130nm we are now at 28nm with 20nm being worked on.    That is pretty much "state of the art".  So while slowly chips will move to smaller process nodes (usually a new node every 2 years) we aren't going to see another jump like we did from 130nm to 28nm in a year ever again.

GPU 300 to 500 J/GH
First gen ASICs 10 J/GH
Current best shipping ASICs 0.5 J/GH
Next gen ??  maybe 0.2 J/GH to 0.4 J/GH (improved design @ 20nm)

The gains are going to get smaller and smaller.  

Hey,

Would you like to write a detailed article on this for my site BitcoinMiner.net?

let me know

[GigaBit]

Hey OP,

To me, your question itself isn't asked correctly.  Better question would include the SHA256 technology itself, since ASIC right now is what powers this technology.  In this case, my company's been doing extensive research to answer this very specific question.  So far we discovered that there is really no limit other than human imagination itself and its ability to push the envelope, in regards to the technology.  Whole networks and even websites could be built off of it and of course, miners are needed to power the SHA256 technology.  Lots more could be done, so if you worry about mining or the 56nm miners going obsolete... or a miner over saturation; chances are slim to none on all those accounts.  Obsolete miners can be used for other purposes.  I'll buy anything over 10GH/S of a killer price since we can resell the miner but marketed as a different product (company secret).

BitCoin is the currency of the SHA-256 technology.  So your miners eventually could power sophisticated security systems and you'll get paid in BitCoin, or another form of crypto coin.  This technology was technically born yesterday, it's still a baby.  No one knew some day that you could watch TV on the internet in the mid 90's.  Only people who had emails were nerds and businesses.  There's really nothing ground breaking that was built on the SHA-256 technology YET other than currencies but a lot more can be made; which in turn needs miners.

Technically, a whole new internet could be built using the technology.

Look at a tree like this: SHA256 technology is the tree trunk, BitCoin is the fruit, ASIC is the rainfall, the roots is the BlockChain.

-Only limits are the borders humans set for themselves.  Bitcoin CAN technically die any day, theoretically, it could be getting killed as we speak...  There's an unlimited amount of money on earth, if your goal in life is to get more money, you'll never be happy regardless.

[davejh]

My assumption is that ASIC miners optimize for the "hashes per joule" metric, as did GPU miners over CPU miners.

It would seem that after some time, the decision to mine Bitcoins would be based on if you could source energy cheaply enough.

When do we reach that point?

I'm not a hardware guy so I'm curious to understand what the upperbound limits are of ASIC mining from a technical perspective. How much more can we optimize?

Thanks!

ASICs managed to jump quite a few process geometries to get to 28 nm and 20 nm but from this point things get really hard. There's probably a 2x to 4x improvement every 2 years from now on, dictated largely by the semiconductor technology (http://hashingit.com/9-where-next-for-bitcoin-mining-asics, although I now suspect I was too optimistic on the 16nm change).

Power efficiency is certainly not going to improve faster than the hash rate now though unless someone identifies a brand new technology to use. The hash rate will continue to drive upwards (albeit at a much slower rate than it has) so it's inevitable that that will drive more mining towards lower cost energy (wherever in the world that may be found) and lower hardware margins (but more hardware burns more power). Each improvement, however, simply defers the time at which the majority of the mining reward goes towards energy costs.

[Syke]

My assumption is that ASIC miners optimize for the "hashes per joule" metric, as did GPU miners over CPU miners.

Most ASIC designs at this point have been quick and dirty designs. Only two seem to have had "hashes per joule" as a primary goal, Bitfury and Bitmain. Both of those designs are at a huge 55nm, yet are nearly as efficient as all the other 28nm designs.