Bitcoin: On our way to a Petahash

[ThePanCakeKid95]

Hey guys, I'm back doing videos and as you know earlier this year I made a prediction that we would hit a Petahash by December 31, 2013 but with quickly rising hashrate I am changing my prediction and I also give an analysis about why the hashrate is increasing so rapidly and give educated predictions on what is to come.

Video:  http://youtu.be/a1DHEBrPYVE

[jspielberg]

There is a good chance we will hit 1 PH by the end of September and 2 before November.

[ThePanCakeKid95]

wow that would be crazy wouldn't it

[Operatr]

This is exponential win.

Nothing like this has ever occurred in distributed computing in history, and before any complaining about mining ROI, think about what that means. Bitcoin's network is becoming a processing monster, and with dedicated ASIC core at it's heart is ready for a bigger world and a ton of transaction traffic. In the end, the network is quickly becoming larger, more distributed, and secure.

Really having a highly secure and robust network in place before Bitcoin takes its next giant leap forward is a really good thing. Not to equate GOX to mining directly, but we saw what it looks like when the infrastructure fails under the weight of a rapid push. Having a solid distributed net with multi-million dollar mining businesses involved should inspire further confidence in Bitcoin as a real currency.

What is more interesting however, is that despite soul-crushing ROI figures with the current crop of ASIC hardware, hashrate is still skyrocketing   Many mine operators aren't in it for the money, The Erupter USB's arn't selling because it will make anyone rich. I wonder if the simple need for economic change is driving it forward more than the profit incentive. Bitcoin mining would not be a multi-million dollar business right now (as a hardware maker or hardware user) without some level of belief in Bitcoin and what it means to the world.

So, being a miner at the moment sucks, but the end result is extremely good for Bitcoin.

[gmaxwell]

Bitcoin's network is becoming a processing monster, and with dedicated ASIC core at it's heart is ready for a bigger world and a ton of transaction traffic. In the end, the network is quickly becoming larger, more distributed, and secure.

Really having a highly secure and robust network in place before Bitcoin takes its next giant leap forward is a really good thing. Not to equate GOX to mining directly, but we saw what it looks like when the infrastructure fails under the weight of a rapid push. Having a solid distributed net with multi-million dollar mining businesses involved should inspire further confidence in Bitcoin as a real currency.

You do realize that all this mining stuff has nothing to do with processing transactions, and that instead of running Bitcoin nodes that can process transactions practically all of these miners are just getting paid by pools to do outsourced sha256 calculations, and they never see transactions or even the blockchain at all, and as a result don't contribute much to improving Bitcoin's security— ... and that we're even in a worst state than we were a year ago,  now kidnapping or hacking _two_ people is enough to freely perform transaction reversals...

You realize this, right?

[Syke]

You do realize that all this mining stuff has nothing to do with processing transactions, and that instead of running Bitcoin nodes that can process transactions practically all of these miners are just getting paid by pools to do outsourced sha256 calculations, and they never see transactions or even the blockchain at all, and as a result don't contribute much to improving Bitcoin's security— ... and that we're even in a worst state than we were a year ago,  now kidnapping or hacking _two_ people is enough to freely perform transaction reversals...

Make it _three_. BTCGuild, 50BTC, and Slush make up over 50% of the network. A year ago it took the top 5 pools to reach 50%.

And you don't need to kidnap them. Just use a $5 wrench.

[Operatr]

Bitcoin's network is becoming a processing monster, and with dedicated ASIC core at it's heart is ready for a bigger world and a ton of transaction traffic. In the end, the network is quickly becoming larger, more distributed, and secure.

Really having a highly secure and robust network in place before Bitcoin takes its next giant leap forward is a really good thing. Not to equate GOX to mining directly, but we saw what it looks like when the infrastructure fails under the weight of a rapid push. Having a solid distributed net with multi-million dollar mining businesses involved should inspire further confidence in Bitcoin as a real currency.

You do realize that all this mining stuff has nothing to do with processing transactions, and that instead of running Bitcoin nodes that can process transactions practically all of these miners are just getting paid by pools to do outsourced sha256 calculations, and they never see transactions or even the blockchain at all, and as a result don't contribute much to improving Bitcoin's security— ... and that we're even in a worst state than we were a year ago,  now kidnapping or hacking _two_ people is enough to freely perform transaction reversals...

You realize this, right?

Why don't you try again with a less dickish tone, and I'll reply



 

[thy]

If KnC holds there schedule and starts delivery 25 sept or so we will see 1 PH about the second day, 26 sept and when there first batch is over around 21 days later, the 15 nov we will probably be at 4-5 PH at least.

[jamesc760]

knc stated "delivery at end of september" not september 25th.

[mgio]

knc stated "delivery at end of september" not september 25th.

that's why he said "or so"

[Puppet]

Actually we're heading towards a exahash.

Reasoning: those upcoming  28nm chips produce ~500GH and my guess is ~$30 is roughly the price at which they can still be sold profitably (that includes basic PCB,  housing, minimal markup; variable cost of the chip alone is probably closer to $10 in large volume). So as long as there is demand at or above that price point, it makes sense for asic vendors to sell (or produce and mine themselves). If you do the math at what point most miners would no longer be interested in buying 500GH (250W) for ~$30, depending on electricity cost and BTC exchange rate, I get at a difficulty on the order of ~100 - 200,000,000,000. Which means a network speed of roughly a exahash. Thats where we are headed IMO, though Im not taking any bets how long it will take. Could be as little as 2 years from now  to get close to that.

[FeedbackLoop]

Actually we're heading towards a exahash.

Reasoning: those upcoming  28nm chips produce ~500GH and my guess is ~$30 is roughly the price at which they can still be sold profitably (that includes basic PCB,  housing, minimal markup; variable cost of the chip alone is probably closer to $10 in large volume). So as long as there is demand at or above that price point, it makes sense for asic vendors to sell (or produce and mine themselves). If you do the math at what point most miners would no longer be interested in buying 500GH (250W) for ~$30, depending on electricity cost and BTC exchange rate, I get at a difficulty on the order of ~100 - 200,000,000,000. Which means a network speed of roughly a exahash. Thats where we are headed IMO, though Im not taking any bets how long it will take. Could be as little as 2 years from now  to get close to that.

Excellent reasoning. I think you are on target. It would be very interesting to get a leak on how much chips actually cost at manufacturing (28nm for example) to get a base seed value for your estimate.

[Puppet]

Actually we're heading towards a exahash.

Reasoning: those upcoming  28nm chips produce ~500GH and my guess is ~$30 is roughly the price at which they can still be sold profitably (that includes basic PCB,  housing, minimal markup; variable cost of the chip alone is probably closer to $10 in large volume). So as long as there is demand at or above that price point, it makes sense for asic vendors to sell (or produce and mine themselves). If you do the math at what point most miners would no longer be interested in buying 500GH (250W) for ~$30, depending on electricity cost and BTC exchange rate, I get at a difficulty on the order of ~100 - 200,000,000,000. Which means a network speed of roughly a exahash. Thats where we are headed IMO, though Im not taking any bets how long it will take. Could be as little as 2 years from now  to get close to that.

Excellent reasoning. I think you are on target. It would be very interesting to get a leak on how much chips actually cost at manufacturing (28nm for example) to get a base seed value for your estimate.

Im basing my guestimate on the embedded SoC market. A typical highend phone SoC sells for about $15 in large volumes. That is when you buy them from nVidia (qualcomm, samsung etc)., so these companies already have a margin on that, fabbing cost will be lower. OTOH  28nm mining chips may be a fair bit larger than mobile phone SoCs, I dont have any size data (does anyone?), but they should also be a lot more simple to produce and validate, contain less layers, far less IO pins, etc.

My guesstimate may well be off by a factor 2 or 3, but not likely more than that, and so should still give you a ballpark idea.

[bcp19]

Actually we're heading towards a exahash.

Reasoning: those upcoming  28nm chips produce ~500GH and my guess is ~$30 is roughly the price at which they can still be sold profitably (that includes basic PCB,  housing, minimal markup; variable cost of the chip alone is probably closer to $10 in large volume). So as long as there is demand at or above that price point, it makes sense for asic vendors to sell (or produce and mine themselves). If you do the math at what point most miners would no longer be interested in buying 500GH (250W) for ~$30, depending on electricity cost and BTC exchange rate, I get at a difficulty on the order of ~100 - 200,000,000,000. Which means a network speed of roughly a exahash. Thats where we are headed IMO, though Im not taking any bets how long it will take. Could be as little as 2 years from now  to get close to that.

Excellent reasoning. I think you are on target. It would be very interesting to get a leak on how much chips actually cost at manufacturing (28nm for example) to get a base seed value for your estimate.

Im basing my guestimate on the embedded SoC market. A typical highend phone SoC sells for about $15 in large volumes. That is when you buy them from nVidia (qualcomm, samsung etc)., so these companies already have a margin on that, fabbing cost will be lower. OTOH  28nm mining chips may be a fair bit larger than mobile phone SoCs, I dont have any size data (does anyone?), but they should also be a lot more simple to produce and validate, contain less layers, far less IO pins, etc.

My guesstimate may well be off by a factor 2 or 3, but not likely more than that, and so should still give you a ballpark idea.

How does that $15 phone compare to an $1100 i7 4930 processor?  It doesn't.  Neither does it compare to 28nm mining chips.  You're talking a drop in pricing by over a factor of 100 in less than a year.  AM's overpriced toys have only come down by a factor of 20, but he started out so overpriced PT Barnum is bowing in his grave.

1 Exahash = 2 million 500GH chips.  That's only 2x what Avalon had 'on order'.  780BTC for 10,000 chips = .078BTC per chip... and these were 110nm chips.  I'm sorry, but I cannot see a more than 4x reduction in size for ~3x cost in that short a time. 

[Puppet]

How does that $15 phone compare to an $1100 i7 4930 processor?  It doesn't.

It does. Intels variable production cost for an i7 is in that ball bark too.

Now an i7 is about 2 or 3x the size of a mobile phone soc,  silicon cost scales accordingly, and  a bit more to account for lower yield, but dont be fooled thinking it actually costs intel $100, let alone $1000, because it doesnt. But there is a huge NRE to write off, developing an i7 is in no way comparable to a mining chip and may cost billions of dollars, not too mention the process tech, since intel does everything in house.

Its rather simple, a fully processed wafer on state of the art process tech from a foundry typically costs ~$5000  (and drops off to ~$2000 over time as the process matures).
IM not sure what 28nm wafer go for today, but last year they were $5000, today is probably quite a bit less. Lets take §4000 .

A 300mm wafer holds about 600 chip candidates if they are 10x10mm2
http://www.silicon-edge.co.uk/j/index.php?option=com_content&view=article&id=68

$4000/600=$6.6 silicon cost per chip candidate.

If the chip is huge, say 18mx18mm=324mm2 you get 177 candidatres

$4000/177=$22.5 per chip candidate

Add a few percent to accommodate imperfect yields and add a few dollar per chip for cutting, testing and packaging and and you are in the ballpark.

1 Exahash = 2 million 500GH chips.  That's only 2x what Avalon had 'on order'.  780BTC for 10,000 chips = .078BTC per chip... and these were 110nm chips.

wafer processing costs for 110mm are not dramatically lower than for modern processes, the difference is probably only about a factor 2x or 3x at most. So 110nm chips are actually MUCH MUCH more expensive per transistor (/hashing engine/GH) than 28nm ones.  The big difference is that  the NRE (design and maskset) is dramatically cheaper for older processes (could easily be a factor 10 or more), and its much easier to book capacity without having to wait many months since its not like AMD and nVidia are putting in a lot of orders on 110nm anymore.

[Puppet]

BTW, regarding avalon. I read elsewhere their chip is only 16.13 mm2. Lets assume its square, that gives 4.65x4.65mm.
Im also assuming this is a 200mm wafer, not many if any 300mm fabs use such an old process. That gives up to 1188 candidates.

A processed 200mm 110nm wafer should cost no more than $2000 in volume. There is a fixed tooling/upstart cost that will weigh in on small orders, but still:
$2000/1188=$1.6 per candidate.

Cutting, testing, packaging and handling likely cost more per chip than the fabbing in this case, but even so, and after applying a yield %  I seriously doubt it adds up to more than $4 per chip.
The rest is margin. Unless they used 150mm wafers in Fab2: http://en.wikipedia.org/wiki/TSMC

Then you get 630 candidates per wafer, and you may get in the ballbark of the numbers you quoted assuming a substantial fixed fee.


One last point of reference. AMD Llano. Granted, its on 32nm, but that shouldnt make all that much difference, especially since it uses SOI wafers which are substantially more expensive than bulk. It has a diesize of 228mm2, and it retails for as low as $39.99 on newegg, so consumer price,  boxed with fan and cooler, warranty etc.  Do you really believe it costs AMD more than $20 to fab the chip?  I dont. I actually think its closer to $10, but lets use $20

srqt(228mm2)=15.1mm
That gives 256 candidates per 300mm wafer.
256x$20 (tops!) = $5120 per wafer.

Thats includes  yield, cutting, testing, packaging.

Still think Im far off?

[bcp19]

Where did you get the $5k price on 28nm wafers?  Does that 5K include the design cost of the masks?  What about the bumping and binning processes?  The boards? Etc?

Sorry, I just don't see it.

[crazyates]

BTCGuild, 50BTC, and Slush make up over 50% of the network. A year ago it took the top 5 pools to reach 50%.

I'm sorry. Did you not remember when Deepbit was 35% of the network back a year or two ago? Or when BTCGuild was like 50% of the network by themselves?

[Puppet]

Where did you get the $5k price on 28nm wafers?  

Experience and industry contacts

edit, found a public source for you:
TSMC decided to increase prices for 28nm chips "by somewhere between 15% - 25%. Given that a wafer processed using latest process technologies can cost $4000 - $5000, ..
http://www.xbitlabs.com/news/other/display/20110912192619_TSMC_Reportedly_Hikes_Pricing_on_28nm_Wafers_Due_to_Increased_Demand.html

Please note the date on that, 2011, when 28nm was brand new, very low volume and very few fabs offered it. 28nm is now mainstream, offered in volume by all major fabs. So If anything,  $4000 today would be a high estimate, but Im okay using it since Cointerra is not an AMD or nViida and pretty low volume for a company like TSMC.

Does that 5K include the design cost of the masks?  

Absolutely not. Masksets, especially for 28nm, are orders of magnitude more expensive, but its a one time, sunk cost. Once you have a maskset, to determine production volume all that matters is the variable production cost vs market value. Sunk costs will determine whether or not you end up making a profit, but not whether or not you should produce more batches. Its like when buying a miner, once you bought it, purchase price (sunk cost) no longer matters to decide to run it or not, you will run it until it earns you less than it costs in electricity, but the purchase price will largely determine if you end up making a profit or not.

What about the bumping and binning processes?  The boards? Etc?

"Bumping and binning" is part of testing and packaging. Its not included in the $5K/wafer price, but it is included in a 'a few dollar per chip", since the cost is typically closer related to the number of die's than wafers.
If you want a public reference for that too, this is all I can find quickly:
http://electroiq.com/blog/articles/2001/03/technology-comparisons-and-the-economics-of-flip-chip-packaging/

(scroll down to the last charts as the others are unreadable, and you may want to note the numbers they picked for average die cost too: $5. Not $50 much less $500).
Please note that article is more than 10 years old. Gives you an idea what BGA packaging costed back then, when it was state of the art technology. The numbers have dropped since, but are still within the same ballpark of a few dollar per chip.

For PCBs, you can get an idea from the prices of infinitely more complicated PC motherboards (that contain dozens of other asics) and still sell for $20 or so in bulk, or from a raspberry Pi that retails for $40 including the  SoC, RAM, IO etc.  
PCB cost doesnt amount to much in the big picture.

Sorry, I just don't see it.

You can lead a horse to water...

[Puppet]

Speed (TH/s)   1003.00
http://mining.thegenesisblock.com/


One petahash is a fact now.
Im betting on 3 PH before the end of october, just 6 weeks from now.