Topic: Bitcoin mining energy efficiency over time (Read 2367 times)

Which when you look at the curve of others, is absolutely astounding. Who would have thought in 2013 that one could keep the same set of hardware for 18 months.

Its not that GPU mining 'cant' sustain the network - its that ASICs are more profitable, and as a result pushed up difficulty.

dfficulty =/= capability  (and is driven almost entirety by price

I believe that the only way to make a profit using a GPU to mine actual Bitcoins (not an alt-con converted to Bitcoin), is to have a "free" electricity. By free I mean somebody else is paying the bill. If I had free electricity, I could run an operating profit using a 333 MHz Block Erupter.

In the long run, Bitcoin would need something more efficient than a GPU to mine. What would it cost to process Bitcoin transactions using GPU's? It just wouldn't fly at scale, as fun as it might have been 3 years ago.

The graph of log change in mining efficiency has the shape of 1/(log difficulty) ..
*Please note the axes are mislabeled. The left y-axis is 1/log difficulty and the right y-axis is log J/GH

what is interesting is that difficulty is outpacing technological change. Economics would predict that they will eventually converge to an equilibrium, and we see that during much of 2014 as ASIC tech kept pace with network growth-

Whenever the orange line (energy efficiency) is above the blue line (difficulty), changes in difficulty are exceeding changes in efficiency and vice versa. So right now, difficulty is outpacing tech. progress. When ASIC first introduced, technology outpaced network growth (difficulty is a direct measure of network size)

So put one last way - the purple shaded area exists when the network growth is outpacing technological change. The data on GPU mining is admittedly not complete having just one representative data point, but I think that it nonetheless tells the same story.
Green areas exist when technological change is outpacing network growth. When they line up, the "supply and demand" is in equilibrium..

IN GREEN AREAS, ONE WOULD EXPECT TO EARN OUTSIZED RETURNS TO MINING WITH THE NEWEST TECH.

And this also tells a nice economic story: In the beginning, people mined with CPUs and all was well. GPU mining came along and only served to crowd out CPU miners by growing the network exponentially. But since GPU cards are not developed with the express purpose of mining - they are made mainly for computer graphics - they were not induced to improve. GPUs just happened to be better than CPUs. It's like mining for gold with a shoe because it happens to be better than a sock. Also, since CPUs and GPUs run inside a PC which is normally left powered on anyhow, the electricity usage extracted for mining was less obvious than it is now. And GPU mining also started to crowd out lesser GPUs. We see this as the difficulty chart starts to level out a bit in 2012.

This all changed with ASICs - specifically designed to mine. The shoe was replaced by a shovel. The shovel then turned into a steam shovel and then a modern-day mining operation. It is only with ASICs that the economics begin to line up because the induced technological change is making for better shovels and not finding something better than shoe which also not a shovel.

Yes can do something like error bars..

I don't want to take hazenyc's thread any further off-topic than it already has, and there's existing threads that cover the ASICs killed GPUs (while CPU miners chime in with a now-you-know-how-we-felt) discussion much more in-depth anyway.

---

I knew cscape's post was buried in the bitfury thread somewhere.
BitFury S-HASH board (overclocked)
July 24th 2013
35W@40Gh/s (0.875J/Gh)
https://bitcointalksearch.org/topic/m.2796663

There's also some figures for the H-cards, but those don't function stand-alone, so efficiency numbers for the card alone aren't quite right.

Can skip this one, as efficiency was much lower at a later date:
Bitburner Fury
September 20th 2013 (date of images of final board posted)
85W @ 42.59Gh/s (~2J/Gh)
132W @ 50.22Gh/s (~2.63J/Gh)
https://bitcointalksearch.org/topic/m.3302316

i'm not saying this(but actually Wolfo is still making profit with them, so perhaps there are still some profit to be made, if you have his miner), what i was trying to say is that, if asic never existed, and bitcoin was still done with gpu,

casual miners would have more chances to join because it would have been much harder for big company to clump gpu instead of asic, because of efficiency, heat, encumbrance ecc...

Are you saying that GPU mining is still profitable?  I parted mine out long ago as I was not pulling much profit.

I might be wrong.  But I just don't see it being able to ROI.  I have one rig just setting there I kept in case but I don't even plug in to mine at this point.

it's not the same, because gpu are not produced by angry miners nerd, that release a new asic every month, they are produced by company that need 1 year for the next gpu to be ready

this result in low hash, and allowing my 1 gpu to stay compotitive, even if you buy 1k gpu, because the total diff would be much lower compared to asic

Yup, my Bitfury ran at .9 J/GH, delivered around Nov '13.

Another option for your graph would be to have a small "bar" for each current point. Some bars would much like a point, others would be much taller. The bar would be vertical since it happens on a date, and it's height is determined by how flexible it is in terms of efficiency. Just a thought.

I can see your point on it.   I think adding maybe a second color with it's best efficiency might be nice.  As over time this is what most of the gear will run at and end it's life running.

Or a second chart might be more proper a underclocked chart.   I'm not sure if second color on original would look messy as I like the clean look.  So maybe a second underclocked chart.

you make a good point. While my first chart is illustrative in showing technological progress over time, it doesn't represent the evolution of the network as a whole since there will always be people mining with not the best tech.

What would be kind of awesome is to guesstimate how much current hashing power is coming from say SP20s and what %age comes from S5s and what %age from A1s etc. And then the network average efficiency can be interpolated, assuming some constant average world electricity price per kWh.

You might think about putting A1's/Dragon's on your chart.  I know they are pretty inefficient at this point, but there is a heck of a lot of them out there.

The breakeven efficiency right now with the current market price and difficulty is around 0.80 J/GH (W/GH/s)
It is interesting to see  that amount of hashing power (measured by mining difficulty) has so far been outpacing increases in hardware efficiency by quite a bit.
If everybody in the network were running S5's today, the price of a bitcoin would be $156. It is because there are still people mining with marginal hardware - or in your case mining at a marginal loss that keeps the price of bitcoin higher.. for the moment.

IIRC the bitfury gear was actually more in the 0.9-1.3w/GH range depending on overvolting. I still have my 6-card rig from the september batch hashing away, but i think its drawing >1.1w/GH and is time to be unplugged any day now if the price doesnt break $250 upwards

just fyi the plotted points ought to be the "stock" setting and not its efficiency when under- or overclocked.

Phil has got the Avalon 4.1 at .453 -  https://bitcointalksearch.org/topic/m.10827180

I personally have not tired yet on that low I still am in .5X area as I was very happy with that currently - https://bitcointalksearch.org/topic/review-avalon-41-miner-summer-mining-notlist3d-1008726

Newest ASICMiner is prisma and I don't think it will be as good as the SP20, S5, Avalon 4.1 etc.

Very nice work.  I would like to see the Avalon 4 and newer ASICMiner on the chart just to see where they fit in.

I would agree with other's on taking it off look in the hashcoins thread.  But they still have yet to show a Zeus which is not near as effective as the uranus.

All they have shown is a scrypt that was re branded, and a miner that is not much better then the dragon A1's in specs.

Can't really conclude that.  There's only one data point for CPU and GPU.  There were certainly more efficient CPUs and GPUs as well, though solid data on those is sparse because 1. ancillary hardware and 2. nobody really cared about efficiency back then, hashing faster and faster was the name of the game.

Not much has really changed anyway.  Bitcoin's PoW provides something of a positive feedback loop.  If B has greater hashing power than A, then B effectively gets more funds than A, allowing them to invest in even greater hashing power than A.  Laws of economy mean that this is not proportional between the two.  E.g. if I have the startup capital to buy 100 GPUs, and you can only buy 1, I can already buy those 100 GPUs at a discount over your single purchase.  I also get more reward, so by the time you can buy a 2nd card, I can buy - say - 110.  By the time you get to your 10th card and have to deal with the heat, I've got a thousand and with a little extra investment they're churning away in professional racks.  ASICs may have expedited things (and one can argue whether that's a good or a bad thing), but it was always going to go this way.  I don't know if Satoshi actually hinted at this, or foresaw this, as it truly is now - but you're absolutely right that at least he didn't see supporting the blockchain as something that everybody would be doing.

Anything based on BitFury or KnC's new stuff (if existent), or theoretically CoinTerra and HashFast's designs, or Coinbau et al, for example, may end up somewhere continuing that curve.  The danger in including the Uranus is that some people will point to it, go "it's the most efficient!", and run off to buy one...when there's as of yet no evidence one exists and its manufacturer is unwilling or unable to cooperate in changing that status.  The difference between that and the others mentioned is that you can't buy those.  Unless you show up on their doorstep with some serious investment money, at least. ( Though I guess HashFast's chip design is technically still for sale, if you can convince DXCorr to part with it and then feel like fronting the money for fabrication and a miner design around it anyway. )

Back a bit more on-topic.. true, BitFury hardware especially can be difficult to figure out given that a bunch of it was manufactured for large mines (some of the hardware popped up in threads here a few days ago, bought by private individuals off of mine auctions) early on and not so much for retail sales.  Still, you should be able to goog some of the more popular completed miners and find a date-ish   If not, hopefully some others will be able to get you some values there.

The hard part isn't getting the specs as much as getting the launch dates..  And again, I am not including any "backward steps" for this particular charting effort.

This is capitalist induced technological progress at its best !

As for the Uranus, even if that is not the 0.26 J/GH chipset, there is bound to be one debuted around that time based on the trajectory of the chart - that is until a technological wall is reached..

Agreed.

It was, and those little H-boards were well ahead of the curve. Some people are only just now replacing them from their farms.

this is what i was searching, notice how the production cycle is much faster with asic, gpu are far slower, if we stayed with gpu, the diff now would have been much lower, allowing many other people to mine

instead we have a centralized pow(in the end is what satoshi wanted, just few big farms, but i would still prefer that casual miners could join the network)

I wouldn't include the Uranus unless and until it becomes provably available.

I think the gap between the BFL and the Antminer is probably filled with some of the BitFury hardware?
https://en.bitcoin.it/wiki/Mining_hardware_comparison
( not an exhaustive list - try trawling some of the mining calculator sites, they've got wattage numbers for miners at least )

It's a bit difficult to gauge, as J/Gh can be lowered for most miners by running them slower.

 I couldn't find a better visual resource so I made my own..

Anybody care to help fill in the blanks? I tried finding on forum threads the official launch dates or available for sale dates.

I tried to include only the most energy efficient for any time frame, so if a new device came out later but was less energy efficient, it was omitted..

For GPU & CPU I just used one standard average, there were probably too many different configurations of chips & GPU cards all running at once in that time period..

*Please note the y-axis is a base10 log scale.