Difficulty curve

[byteminr]

It's an exponential right now, but in a few months this curve will look much more like an 'S' curve.

The question is when will it start to tail off?

My guess is December, what does everyone else think?

[FeedbackLoop]

Cointerra is set to deliver December/January and they look serious so I would say it will take longer than that to leave the exponential (on average).

[byteminr]

Cointerra is set to deliver December/January and they look serious so I would say it will take longer than that to leave the exponential (on average).

Yes, but say, for example (and I do not know how much they have sold) KnC deliver 2 PH in October, to keep the exponential BFL/HF/Cointerra etc would need to deliver what, 8PH before Jan to keep up the rate? I can't see that happening can you?

[FeedbackLoop]

Cointerra is set to deliver December/January and they look serious so I would say it will take longer than that to leave the exponential (on average).

Yes, but say, for example (and I do not know how much they have sold) KnC deliver 2 PH in October, to keep the exponential BFL/HF/Cointerra etc would need to deliver what, 8PH before Jan to keep up the rate? I can't see that happening can you?

Good point, indeed I don't see 8 PH/s but i do see 5 from all others minus KNC.

I agree that KNC will be the main contributor of hashrate in the coming months (hopefully starting September since I have an order with them). Assuming your estimate of +2 PH/s, that would be about triple the network (~1PH/s). At +64% per month exponential that's 2.25 months (3 = 1 * 1.64^2.25) so KNC alone keeps the exponential till December (fitting the October and November deliveries). Then the other guys need to deliver December plus January at +64%. From now (~1PH/s) that's 1*1.64^4.25 = 8.18 PH/s which, removing the 3 PH/s is aprox 5 PH/s. So they only need to deliver 5PH/s in December/ January.  I think 5 PH/s is not unreasonable for January deliveries from all the other guys (in which I would not include BFL). I actually think you will need to take a 60 day running average to keep the exponential but, like I said, on average (60d ) I expect it to last till end of January.  

So I would just put the exponential one month further than you and with a larger running average. But after that, of course, there's next gen...

I actually feared in the beginning KNC would deliver closer to 3PH/s but been taming those expectations a lot from feedback from other forum members. You in particular saying 2 PH/s pushes that fear back up a little bit again since you should have better estimates than most ;P.  2 PH/s is curiously 2*2500 which would be the November Jupiter available stock of 2500 (if one believes the hardcoded website limit like someone posted) times 400 GH/s times 2 for the September/October delivery.

edit:  8 PH/s by end of January matches this:
 https://bitcointalk.org/index.php?topic=283820.0
with one month delay

[byteminr]

Yes, but say, for example (and I do not know how much they have sold) KnC deliver 2 PH in October, to keep the exponential BFL/HF/Cointerra etc would need to deliver what, 8PH before Jan to keep up the rate? I can't see that happening can you?


Good point, indeed I don't see 8 PH/s but i do see 5 from all others minus KNC.

I agree that KNC will be the main contributor of hashrate in the coming months (hopefully starting September since I have an order with them). Assuming your estimate of +2 PH/s, that would be about triple the network (~1PH/s). At +64% per month exponential that's 2.25 months (3 = 1 * 1.64^2.25) so KNC alone keeps the exponential till December (fitting the October and November deliveries). Then the other guys need to deliver December plus January at +64%. From now (~1PH/s) that's 1*1.64^4.25 = 8.18 PH/s which, removing the 3 PH/s is aprox 5 PH/s. So they only need to deliver 5PH/s in December/ January.  I think 5 PH/s is not unreasonable for January deliveries from all the other guys (in which I would not include BFL). I actually think you will need to take a 60 day running average to keep the exponential but, like I said, on average (60d ) I expect it to last till end of January.  

So I would just put the exponential one month further than you and with a larger running average. But after that, of course, there's next gen...

I actually feared in the beginning KNC would deliver closer to 3PH/s but been taming those expectations a lot from feedback from other forum members. You in particular saying 2 PH/s pushes that fear back up a little bit again since you should have better estimates than most ;P.  2 PH/s is curiously 2*2500 which would be the November Jupiter available stock of 2500 (if one believes the hardcoded website limit like someone posted) times 400 GH/s times 2 for the September/October delivery.

edit:  8 PH/s by end of January matches this:
 https://bitcointalk.org/index.php?topic=283820.0
with one month delay

One other thing to consider - before we believe all these sales are real and will materialise. When KnC deliver their first batch, how many people will be cancelling HF / BFL/ CT orders and asking for refunds? There has to be some collateral damage here if they deliver everything we expect.

Additionally:
5 PH, even at $15/GH (which is isn't yet, it's still higher than that) would cost $75M USD. Is there really 75Mil USD being piled into mining in the next 4 months? I'm not so sure. That's without hosting, power operations etc. etc. etc..

It's going to be an interesting time for miners, thats for sure!

[Tsukene]

It's an exponential right now, but in a few months this curve will look much more like an 'S' curve.

The question is when will it start to tail off?

My guess is December, what does everyone else think?

I am thinking early 2014

[ModusPwnd]

Just because it rises fast does not make it exponential!  This is a common misunderstanding of how an exponential function works.  Things that increase or decrease exponentially do so because their rate of change is a function of their value.  Some common examples are interest and radioactive decay.  In each of these cases the rates of change are functions of the values.  This is not the case for bitcoin mining.  How fast new miners turn on is not a function of how many miners there are now.  Rather, bitcoin mining rates increase as a function of other external factors.  In other words, if I turn my miners back on right now that will not increase the rate of new miners turning on (quite the opposite if you consider the effect of higher difficulty).

Remember, a fast rising curve is not necessarily an exponential curve and moreso an exponential looking curve is not necessarily representative of any underlying process operating in an exponential way.  Furthermore, a very slight growth rate can be exponential too, if the value is slight the growth rate is slight.  This is how exponential relationships work.

Sorry if this rant is too off topic, its a pet peeve of mine.

[udet4food]

It's an exponential right now, but in a few months this curve will look much more like an 'S' curve.

The question is when will it start to tail off?

My guess is December, what does everyone else think?

I am thinking early 2014

I believe it will start to tail off in 2nd quarter 2014 and in late 2014 difficulty will be stable

[Syke]

Just because it rises fast does not make it exponential!

If you look at the difficulty adjustments for the past 3 months you'll find the network has grown at an average pace of 25% per adjustment. So we can say that the difficulty is roughly following an exponential curve.

[Puppet]

It's an exponential right now, but in a few months this curve will look much more like an 'S' curve.

The question is when will it start to tail off?

My guess is December, what does everyone else think?

Most certainly not december, December and January are the months you will probably see the biggest difficulty explosion ever, assuming at least some of the promised 28nm products make it to the market roughly on time.

When is it going to taper off? When marginal production cost meets mining profitability. Until then, ASIC manufacturers will just drop prices to maintain sales (causing difficulty to increase further, and requiring further price drops).
Cointerra is now on preorder for ~$2800 per chip. Im guessing variable production cost of the chip alone is closer to $10 per chip. Do the math were difficulty would need to be for 500GH at ~$10 (+PCB+ case+margin, say $25-$30 ) to be only marginally profitable. That gives you an idea. Then double it when 20nm becomes affordable.  The short answer is: not anytime soon.

edit; pallpark figure, when difficulty reaches ~100,000,000,000 is where 500GH for $30 is only marginally profitable, depending on electricity cost, BTC rate etc,  Take that figure with a grain of salt, but it gives you an idea. It may also taper off sooner, but OTOH I also expect we will overshoot this value due to inertia, free electricity and miner myopia.  Also, this is a race, until we get close to this point, the only brake on network growth will be the ability of the various vendors to produce and ship. Unless they are all as inept as BFL and avalon in getting their goods out of the door, that may go incredibly fast. If you thought 2013 was crazy, you havent seen anything yet.

[AussieHash]

Just because it rises fast does not make it exponential!

TL;DR

[Puppet]

heh, you expect them to be abe to read a logarithmic chart?
 
“The greatest shortcoming of the human race is our inability to understand the exponential function.”
Albert Bartlett

Maybe this will help though:



2.5% growth per day is an exponential function (that happens to work out almost exactly to doubling every month).
And if that is not scary enough, the growth rate is still accelerating, and thats well before even a single 28nm asic was shipped.

[ModusPwnd]

Its also a power function.  Its also a set of piecewise linear functions.  Its also a series of step functions which is physically what actually happens to the difficulty.

If you look for a pattern you can often find it.  That doesn't mean the pattern is representative of physical processes and useful for predictions.

[smoothie]

I think it will be about 6-9 months before it start to truly level off. We may see a brief pause as new asics get shipped and out into the hands of users.

But it's on...like donkey Kong.

[FeedbackLoop]

“The greatest shortcoming of the human race is our inability to understand the exponential function.”
Albert Bartlett

Awesome quote!

Also exponential functions are present in very much everything self organised. And humans are bloody good at self organising... It is not surprising that the growth is indeed exponential and I won't be surprised to see it continue well into mid next year with all the manufacturers delivering second gen, third gen, bulk chips of all gens and all the hype overshooting well past the point any retail customer can ROI. Along the way people without end will keep saying that it's about to level or go S shape or linear or that people will stop pouring money at it and what not...

Disclaimer: I am still one of those "pouring money at it" (via one purchase at KNC)... still...

[ajw7989]

I also feel we have a long wait before the difficulty rise starts to slow down. We will have to wait until technology has peaked and more efficient miners are not being made.

[jmutch]

hi guys,
Now that we are at 1 peta, this is my guesstimate:

[AussieHash]

hi guys,
Now that we are at 1 peta, this is my guesstimate:

That would be nice, but NO !
https://bitcointalk.org/index.php?topic=283820.0

[Puppet]

hi guys,
Now that we are at 1 peta, this is my guesstimate:

LOL.. I was going to make a similar one. Only its seems much more likely that the curve will look like a bell curve and next year difficulty will go down. You know it makes sense.

[gmaxwell]

Sorry if this rant is too off topic, its a pet peeve of mine.

It's also a peeve of mine, and it very much is ontopic. The people just fitting random formulas to the hashrate and making predictions are producing BS numbers that obscure better analysis (e.g. ones factoring in published shipping schedules or profitability over power costs) and make everyone dumber.

Also exponential functions are present in very much everything self organised. And humans are bloody good at self organising...

All living men are mortal.
Socrates is mortal.
Therefore, Socrates is alive!

Sorry, you can't just wave some hands and some things that have some vague commonality. Good reasoning requires at least arguing for a causal relationship. I can argue how increases in hashpower encourage decreases in the rate of increase— higher hashpower makes every bit of marginal increase in hashpower less profitable— but not the opposite.