Topic: ASICs - which should you choose? (Read 12980 times)

Another update:

http://organofcorti.blogspot.com/2013/03/912-asic-choices-asic-earnings-update.html

Updated:

http://organofcorti.blogspot.com/2013/03/911-asic-choices-avalon-and-bfl-earnings.html

Updated:

http://organofcorti.blogspot.com/2013/02/910-asic-choices-avalon-update-3.html

Sorry I didn't reply earlier, but I like MrTeal's idea (and by extension, yours) and I will be posting charts showing earnings for $0 running costs, and then a separate chart showing the cumulative running costs (for various scenarios).

ATM I'm working in a few posts about the "Weekend Dip" btc trading strategy; after I'm done with those (probably this weekend) I'll post upated charts for the ASIC choices series, so if you have other requests post them now and I'll consider them.

^THIS^ is an awesome idea! let people do some of their own calculations/estimations on their daily costs. maybe make it a new blog post, so as not to clutter up the excellent one already made.

I envision multiple price points but stick with a static hashrate, it could even be $/day/GH/s though I imagine $/day/66GH/s would make more sense to everyone, since 66 seems to be the consensus entry point.

and please, PLEASE, PLEASE! (with a cherry on top! i.e. a 1 btc tip from me ) include $0.00/day

Rather than separate charts for each unit and electricity cost, why not just make a set with cost to run per day? The cost for running a BFL Single at $0.30/kWh should be roughly the same (depending on actual power numbers) to running an Avalon with $0.05/kWh electricity, so they could both be put under a $0.75/day chart.

I've just make an order for one unit BFL single SC 2 days ago, I love the size and how does it looks and i prefer quality and power saving consumption as per offered. I don't care for shipping schedule as long i could keep mining later and contribute on bitcoin project.

Update for optential Avalon owners who pay ~ US$0.25 per kWh: your earnings have been estimated.

http://organofcorti.blogspot.com/2013/02/98-asic-choices-avalon-update.html

I've heard through the grapevine that they are going to be delayed one day, making the date of arrival even more special.

Seriously, though, this thread should be renamed. I choose not to choose for now.

Personally, I think the ASIC race - first to deliver a working production model -will be won by Goblin Logic ASIC.

In case you haven't heard, Goblin Logic have been working on ASICs for over a year now, they claim to be capable of
100GH/s with a power consumption of 1 Watt (Diagrams to follow)

Not bad for a bunch of goblins that live behind my shed (between Martins newsagents and the council bins)

I'm handling pre-orders for them - $150  (£95 real money) - will secure a Goblin ASIC when delivery arrives on 31st March.

Another quick post before the reviews are published:

9.7 ASIC choices: What will the first batch of Avalon ASICs earn?

A new update due to being schooled by Mr. Teal with some new assumptions and hashrate arrival time modelling.

9.6 ASIC choices: WWJE update

Share and enjoy!

A new update due to Avalon is shipping:

9.5 ASIC choices: Avalon ships, guerilla bitcoin mining and WWJE (what will Jeff earn?)

Could you explain your idea more clearly? I which part of the lined blog post was an advertisement and whom did it advertise? I'd like to know since I haven't been paid for any adverts.

I think you should choose something tangible...not some pixels ( to be read as advertisements )

I've published a belated update using the following specs;



All the pretty charts are here:

http://organofcorti.blogspot.com/2012/12/94-asic-choices-update-24th-december.html

Enjoy!

That was quick - thanks mate. I'll update with an extra $40 included in the cost to purchase.

newegg(.com) is the go-to computer hardware store for most of the US and Canada. Bastards just have to be located in California too, so I'm the only one that gets charged tax (well and 34million others but they don't count).

200W PSUs tend to be notoriously cost ineffecient (I wouldn't trust most of the sub $30 offerings with my enemies computer components) though.
I think the Antec EarthWatts 380D is probably one of the better low cost solutions out there (~$45 for 32A combined on 12V rails), though I will admit I've never really looked that hard for low power items.

EDIT: also a sweet seasonic 300W on sale for $39.99, 80% Bronze and all that.

I was going to do an update for the new bASIC specs, but then I read that they won't be shipped with PSUs, which I'll need to include in the ROI costs. What sort of wattage would the average buyer use - how many are buying multiples? Or should I just assume a 200W PSU is being used for one bASIC?

Also, I know how much PSUs cost where I live and I have no idea which places are preferable for purchasing such products. So if someone has an idea how much people in the US pay for say a 200W PSU, it would save me some google time.

Yeah, it all depends on what hardware you buy. But reducing running costs is key to remain profitable in the long run.

Many may not want extra features if they use too much power or are redundant after more than one device! (Of course there are a few celebs I'd like to have dinner with)

I would also worry. I can't see US$/kWh decreasing, and as D increases low energy consumption will be key. In fact I predict that in a year miners will worry more about electricity costs than anything else. I also predict that within a year all ASIC manufacturers will make efficiency their top priority so that ASICs will have very similar hashes/joule and hashrate/$, and will have differential pricing based mostly on usability features (standalone wifi vs connected to a computer, flashy paint job, dinner with some celeb etc).

I'm not sure about this. Here's an example: Miners don't switch off as long as it doesn't cost them to mine. If, when all the pre-purchased ASICs come on line D,  exchange rate, and electricity costs are all at a level that means the break even point is 2 years instead of one, then as long as they are still mining they still earn and difficulty won't drop. It may even increase as people finding out about bitcoin mining try it for the first time.

Your suggestion is simple, but that simplicity is obscuring many assumptions about how and why people mine. Not all miners are as logical as you.

I would absolutely worry about power usage. It will mean the difference between and profitable and non-profitable miner in the future. You may only be looking at the price or the hash rate now, but later, power usage will be the only thing keeping your head above water.

It won't play a big difference for the first few months, but going from 2 Singles to 2 Avalons can mean the difference between $15/mon and $90/mon in costs.

True that

I wouldn't worry about power usage if you have one or two Asics, they use as mutch as some lightbulbs XD.

Working on some difficulty curves along those lines. But after the long weekend...

What about a different approach to the difficulty prediction: Calculating how much of ASIC hardware is added to the network reach an equilibrium for the time to reach  ROI for the ASIC.

 My prediction is that the network will find an equilibrium where the time frame to reach ROI for an ASIC device is somewhere comparable to current GPU mining. With the 25 btc block award when ASICs hit the market, it can be argued that the ROI time frame will double from current GPU mining.
If current ROI predictions for a GPU has been about 6 months, a reasonable estimate would be that the network will reach an equilibrium difficulty with ASIC ROI of 1 year. (If ASIC ROI is over 1 year, less people are willing to invest in mining, thus reaching an equilibrium at 1 year) What is the difficulty with the current generation ASIC devices and prices?
If the difficulty will jump to a value where buying an asic would pay itself in 1 year, then it would be about 20x of the current value (assuming 60gh/s, 100w, price 1300 dollars, btc price at current level using bitcoinX.com calculator).
At the current ASIC prices of 50 Mh/dollar the network total of about 500 TH/s would cost about 10 million dollars. This is probably more than the price of the current GPU network.

I was getting the info from this post:

That's from the BTCFPGA Forum

The quote was along the lines of "a 1000W PSU will power 8 to 10 of the 54 Ghps bASICs". I used the more conservative estimate of 1000/8 = 125W.

Tom has said the 54GH/s unit will use 100-120W, and will take either a 4pin molex, or a DC barrel jack.

This is what I based my last post on:



It's lower than 200W. Maybe you were looking at earlier posts?

As far as the rest, I don't know. I play with numbers, then get other people to explain to me how to deal with the hardware

Edit: I'd forgotten to link to the lastest post in the OP. Sorry about that.

Thanks for your insightful thread organofcorti. It leaves me curious, though.. I have some bASIC devices incoming and your estimates are much higher than I was expecting. Is it not safe to say that the 54gh/s units will top at around 200w? Also, what kind of power connector will be required? others have speculated it will run on standard 6-pin pci-e connector, but wouldn't 8-pin be necessary if they really draw this much juice?

I did a 3d version, not as easy to understand as the tile plot and a bit uglier. I can post it though if you want.

I want to make that chart 3d... mmm 3D....

I follow you - it's about buy and hold vs invest in mining equipment, yes? I'll try to think of a way to show that not involving using a btc denominated chart.


I'd like to point out that my sexual fantasies rarely involve bitcoin.

That said, there is a very good reason I increased the x-axis to US$40/btc in the chart you refer to, and that is to enable comparisons. If one device at one locale can get you to the break even point at 12 months at starting difficulty = 30 million and an exchange rate of US$10/btc, and the same device will reach break even at 12 months at starting difficulty = 30 million  and an exchange rate of US$30/btc, then you know it's time to move to a state where electricity is cheaper. I'm not expecting btc to reach maximum of the x-axis, but if I didn't include up to US$40/btc then some of the chart would have been blank.

With those points in mind, and with the aim of producing an ROI like chart, I made the following:







More charts and more detail at http://organofcorti.blogspot.com.au/2012/11/93-more-on-asic-choices.html

Good effort but putting such a large range on BTC price only hurts your presentation.  You should be tackling mining profitability with the mindset of "Hey, let's see how profitable this is given that BTC stays at $11 or thereabouts."  There is absolutely no point in even peeking at mining return over ~$10/BTC (A crueler man than I would call it masturbatory).  I recommend a range of $11/BTC and down if you insist on having exchange rate as a variable.  Otherwise, I'd just stick to a constant $11/BTC.  It is much more useful to show a wide range of difficulties and difficulty prediction functions as that is the true unknown variable here.

Interesting. Thanks for posting.

Very nice graph and good article, keep up the good job 

Mmm, I am a bit confused now. Would you please restate your point?

AFAIK it's simple: assume you have an initial amount of $1000. You decide to buy mining equipment which generates 100 BTC before it blows up in your face (we neglect the electricity cost for simplicity). Alternatively you could have bought 100 BTC from your initial $1000 at $10 per BTC. Assume in 1 year BTC is $1000 a piece. Now your mining equipment generated $99000 profit.
Correct? No - because the profit is an outcome of the value appreciation of BTC. Your device just made it barely to break even.

I'm not sure about the opportunity cost as it applies to mining since converting either from local currency to btc or btc to local currency means you have to hold on to one or the other. So, assuming you're correct - and I guessing here - but is there no opportunity cost for local currency because you're spending it and not earning it?

nothing?

Technically you can easily renormalize the denomination at the break even point with the above assumption. Personally, I prefer the BTC denomination, because it provides information on how much BTC can be "generated" with the device (production-cost). If you denominate in $, you should always have two graphs: the value increase of your mining investment, and the missed opportunity of holding on to your BTC. The difference provides true information on the ROI.

Sorry. Ummm...sharing is caring?

Looking forward to your updates.

Ah crap. Another update is due, I believe. Once I've hashed out the ROI curve details with Jutarul, I'll update.

That's pretty much what I thought. What have I misunderstood?


Since I'm not attempted to create an accurate model for mining but a comparison between devices, I'd thought that evaluating a1(x) - a2(x) at the point of evaluation only (ie as if no electricity bills are paid until the evaluation point, and no btc are spent until then). In this case isn't the denomination of the graph immaterial? Whether it's denominated in btc, local currency or % ROI the results should be equally valid. Is this right?

I'm just asking this because I'm working on an ROI chart like yours, but as a tile-plot, the third dimension as the exchange rate at evaluation point. I don't want to post it if there's a glaringly unforgivable error in doing so.

You're welcome. Since you took the time to create a nice post, I thought I'll take the time to give you some feedback. Also gave me an opportunity to re-evaluate some aspects about mining profitability.

Yes. a2(x) is a bitch. And in reality it's not a constant and does not resemble an average.

Rational miners should mine as long as a1(x)>a2(x). However, that disregards other factors, like resale value of the hardware, protection of the network, etc.... That's why I used the term "upkeep". It's the expense you have for generating new coins and "protecting" your investment.

The thing about a1(x)>a2(x) is also that it may trigger some people to hold on to their hardware for latter use, when the exchange rate becomes high enough again... It's a very dynamic system.

Hot off the presses


I believe that should send that debate to it's final resting place.

https://bitcointalk.org/index.php?topic=79637.2060;topicseen

Of  course, go ahead and use what you need. It's why I wrote the blog post.


The challenge for you will be modeling the initial change in difficulty until it reaches steady state. After steady state it's likely to increase by some arbitrary average percentage, but until steady state is reached the increase in hashrate will be a function of the hashrate and number of sales of the devices over time. At a very uninformed guess, I would expect the cumulative increase in hashrate over time to be something like an exponential or pareto CDF.

Once you have your model for the hashrate increase, you need to model the increase in difficulty. To do this properly you'll need to include difficulty increase by a maximum of 400% per difficulty period, and also that the greater the increase in hashrate, the shorter the difficulty period will be (in terms of time elapsed).

You've probably already thought of all this, but if not I hope it helps. I'll be interested to see the results you get.


It was based on the average percent difficulty change since the start of the year, and assumes that difficulty has reached a steady state after the bulk of the ASICs have been introduced. It is an assumption that may not be at all warranted, but it's likely closer to be more accurate than trying to model the exchange rate.

Thanks for taking the trouble to write a helpful, clear and thoughtful post, Jutarul. I followed your explanation well enough to have a few questions which I'll post one at a time.


a2(x) what what I was trying to avoid. I think btc exchange rates will continue to be much more volatile than I'm happy to attempt to model. So by calculating the fiat to btc conversion once and using the the exchange rate as an independent variable, I don't have to be concerned with long term exchange rate volatility affecting a model's outcome.

Have I misunderstood any of this so far?

TIA

Thanks a lot organofcorti for all this info, charts, blog.. very nice

Def donation worthy.

I've learned lots and I hope you don't mind me using some of this.
I'd like to project some possible difficulty curve increases. Yes, it will be full of assumptions.

Was there any reason why you chose 5% per diff change or was it just a reasonable assumption used to evaluate your original analisys of "Which ASIC to choose?"

Gladly.
it's a 2d graph with follows: y = a*x+b. It should be denominated in BTC since this compensates for opportunity cost. In that case we assume the miner holds on the the mined BTC till the break even point (at which point the exchange rate matters).

y = profit per day in BTC
x = day
a = mined BTC per day
b = (negative) initial hardware costs in BTC

a itself is a function of x, since the efficiency of the mining changes with time, thus: a=a(x)
a(x) can become negative since mining operation include an upkeep (electricity cost), thus the price of electricity matters:

a(x)=a1(x)-a2(x)

a1(x) = generated coins
a2(x) = upkeep

a1(x) depends on difficulty (D), hashing power (H) and reward (R)
a2(x) depends on electricity costs (EC), consumption (W) and exchange rate (ER) (since the graph is denominated in BTC)

a1(x) converges against zero with time since difficulty is supposed to increase and rewards drop according to the schedule.
a2(x) may fluctuate a lot around an average value. It is supposed to decrease in short term (increasing exchange rate) but increase in long-term (higher electricity costs).
The lifetime of the hardware is characterized by the time (x) at which a1(x) intersects a2(x).

To provide an example I use your data from your post for BFL (some simplifications used):
b= -100 BTC ($1299 with $12.99 per BTC)
a1(x)= [ 60 Ghps / ( 90 Thps * (1 + 0.05 * (x/14) ) ) ] * ( 25BTC * 6 * 24 ) = [ 60 / 90000 ] * [ 1 / (1 + 0.05 * (x/14)) ] * 3600BTC = 2.4BTC / (1+0.05*(x/14))

(1.05 derived for your 5% increase in total hashing power per difficulty period, 14 days per difficulty period, neglecting compounding)

a2(x)= [ (60 * 24 Wh)*(0.1 USD/kWh) ] * [ 0.1 BTC/USD  ] = 0.0144 BTC

The graphs look something like:

Good work organofcorti and nice blog

I've written an updated post with the suggested specifications:

http://organofcorti.blogspot.com/2012/11/92-asic-choices-update-2nd-november.html

I've attempted to stick to consensus, so I wont update again until there's new published information about device specs or I've made another transcription error.

Avalon is 66Gh/s at $1,299, currently topping out at 400Ws. the reason is we are still finding the good balance between power consumption and speed. which means speed can still go up and power can still go down.

Here are my estimation on the competition's power consumption based on simulation using tools we have at Avalon - do make yourself clear of this fact.

With that said,
from the look of things and information released by Tom, they will be running 100W per module, each module at 27Gh/s, their 54Gh/s rig will be around ~200Ws. ( Yes, I also believe 300 is too high. )

BFL, I expect to reach ~1.5W or more per Gh/s with their 65nm chip. whether they can power their 8 slot board at 7.5Gh/s/7.5W chips each to met their announced specs is a totally another story.

p.s.
been a long time reader of your blog, keep up the good work, some donation has been sent your way.

You misread me. My suggestion was to plot starting diff vs exchange rate at break even vs number of days to break even.


I have no idea how you could possibly include these things without making lots more assumptions. As I mentioned in the blog post I didn't want to make any assumptions about what the exchange rate would be at a given point in time, so it is the independent variable. In order to make sense of this, you have to assume a miner is holding on to all mined coins. Hence the importance of exchange rate at break even point.

As they stand, I think the calculations provide the average miner with a simple method of comparing several different ASICS based only on hashrate, power consumption, assumed starting difficulty and assumed change in difficulty per difficulty period. We can make some educated guesses about these last two assumptions, but not about the exchange rate at a given point in time.



Post a link to an example and an explanation of the calculations you're talking about and I'll try to make one. Otherwise I've provided a method to calculate the data - go for it!

contour sounds interesting. plotting 'starting diff vs exchange rate at break even' may not give you a lot of information since it's strongly correlated. If you want to do a three dimensional graph (which is a contour plot) you'd preferably have 3 independent variables of significance.

Actually I don't really understand why you'd want to use the exchange rate at break even point. AFAIK you should rather use the exchange rate at the time of buying the hardware / pre-order since this factors in opportunity cost. What matters is the time required to recoup the equivalent amount of bitcoins through mining, and this is rather independent of exchange rate (of course low exchange rates at ordering time discourage hardware investments, because more BTC have to be mined...) with the exception of having to sell some BTC for covering for electricity. (actually your post makes no reference to opportunity cost, why?).

A more intuitive graph would be a ROI graph, with the x-axis being the time of operation. the y offset is the initial hardware cost, and the xoffset is the break even point... I didn't see one.

I agree these would be good measures, and I did do a few charts along those lines (unpublished). But I think they might mystify most miners, and the idea here wasn't to present charts but to present the math that will allow people to make their own decisions, and to illustrate that with charts.

That being said, I love a pretty contour plot, and I generally prefer them rather than use a synthetic variable (eg I'd prefer a tile or contour plot of starting diff vs exchange rate at break even, colour indicating number of days to break even).

Any preferences? Get them in now and I'll try to do it when I fix my errors tonight.

nice post. Given that the BTC-USC exchange rate at break even and the starting difficulty are perfectly correlated it would make sense to normalize the graphs against that. Which means a plot: Days to break even VS. ($breakeven / average or starting difficulty)

Actually what would be even nicer is to use some combined efficiency measure on the x-axis. Right now the efficiency factors are distributed across different graphs.

Going by that logic I would remove bASIC completely! Both 200 and 100W are the same rough numbers.
I'm suggesting 100W is FAR more likely a number since that is how the detail was presented.

I understand what you're saying, but facts leave no room for error. Estimates do. We have no official statement about the final power draw. Rough numbers thrown up and quickly taken down do not count.

I would just line up everything fairly.
Take what the companies have let out, and use those figures, rather than only doubling Tom's.

I never actually saw Tom's post that gave rough power numbers, so I'm going off everyone else's claims. As far as why I might come across as a pessimistic, what's the saying: Plan for the Worst and Hope for the Best? I'm going to run my own personal numbers based on the worst case scenarios, but I am more than willing to be proven wrong.

Why would you guess the worst?

He accidentally said 54G was 100W. Deleted reference to it.
Accidentally posted 10 bASIC units would run on a 1000W PSU.

I think it's pretty safe to assume 100W is the aim for bASIC.
60W is the aim for BFL SC 60G.
400W AVALON.

Thanks for your comments, everyone.


Yes, I think that somewhere in transcribing data I swapped bASIC for Avalon. I also didn't know that Avalon was stand-alone, or that it's hashrate would be increased. Since the first two of these are errors on my part I'll update the figures and post an update tonight.

Please keep posting corrections and any updates to device specs - I don't have much time to spend reading relevant threads.

First of all: Awesome chart! A few things that need to be updated, tho, which all have the potential to change the general outlook of the chart:

1) The Avalon team has confirmed it will run at at least 66GH/s, rather than 60GH/s. 10% speed boost? Hells ya!

2) The Avalon will retain it's $1299 price even after the pre-orders have finished. They will NOT be raising their prices to $1999 like they originally said.

3) The Avalon has confirmed a 400W draw, but they said that that could go down further by the time they ship.

All of those facts can be confirmed in the first post here: https://bitcointalksearch.org/topic/announcement-avalon-asic-development-status-batch-1-120184

3) We dont have exact power numbers for the bASIC, but 300W is a little high. That number about the 1000W PSU being able to drive 10 bASICs doesn't clarify whether that's the 27 or 54GH/s units, but I'm inclined to think the second. I'm guessing 1000W will power either 10 27GH/s units, or 5 54GH/s units.


End result: Increase the Avalon hash rate, decrease the Avalon price, and swap the Avalon and bASIC power numbers.

General consensus is basic 54g is 100W.
You missed another post where the flickr account said it and was then deleted. It may have been wrong, but it's a better confirmation than the psu and assuming the worst for your graphs.

Also, the price and hashrate on the Avalon is wrong, it's $1300 for 66GH/s, with a maximum power draw of 400W.

Great work! Life is full of surprises, though - and in this case surprises will turn out to be critical to the outcome. It will all make sense in retrospect.

On the site it says this 1000 watt psu will power 10 bASIC units. What it doesn't say is if that is the 27 or 54 Gh/s unit.

https://www.bitcoinasic.net/index.php?route=product/product&product_id=54

Tom hasn't released his power usage yet unless I missed it. Where did that 300 watts come from?

Excellent work! 

You used two of my favorite things when writing up an article; Math and Proof of Work!

I actually didn't know there was such a disparity between device wattage of BFL, Avalon, etc. I'm sure I've seen it around, but it just never stuck with me. I'm glad to see everything compiled into one doc! 

I've written a short post detailing how to estimate an ASIC device's yearly earnings, the time it will take to recoup the initial cost, how long you can can expect to mine profitably with a given device and more. I won't repeat the entire post here, but I've posted a teaser chart below.  Post your own analyses and chart, and help fellow miners figure out which device they should purchase. I look forward to your comments and suggestions either here or at the blog.

http://organofcorti.blogspot.com/2012/11/91-asic-choices.html
http://organofcorti.blogspot.com/2012/11/92-asic-choices-update-2nd-november.html
http://organofcorti.blogspot.com/2012/11/93-more-on-asic-choices.html