ASIC = The end of decentralized mining

[BFL-Engineer]

This is reasonable, and only time will tell.  I just can't help but feel that the GPU market (as a whole) will move faster than the FPGA market due to demand and market size.  Of course technology across the board will benefit from advancements, but I believe the larger and more profitable markets will see it first.

Very very very unlikely.  AMD just made a shift to 7000 series and performance (relative to 5000 series) was lackluster.   AMD won't be making a new generation for at least 18 months.   

Current 45nm FPGA compete well with GPU in all but the lowest cost areas.  28nm FPGA will eliminate any remaining advantage and ASICs will make GPU mining look as stupid as CPU mining is now.

GPU mining is dead.  Time will only worsen the relative comparison.

Note this comes from someone running a 15GH/s farm all GPU farm.

GPU being dead indicates that bitcoin is growing fast, and evolution/advancement will be inevitable.
I must say of course that a lot of faith was put in AMD 7000 series, but it didn't do well to meet
the Bitcoins market expectation in terms of hash processing power.


Regards,
BF Labs Inc.

[1714790407]

1714790407

[1714790407]

1714790407

[1714790407]

1714790407

[bombo999]

Do you know what time today we can expect the BitForce SC announcement from BFL?

[ice_chill]

Probably noon their time.

[rjk]

Probably noon their time.

In 4-6 weeks... sorry I had to.

[BCMan]

This is reasonable, and only time will tell.  I just can't help but feel that the GPU market (as a whole) will move faster than the FPGA market due to demand and market size.  Of course technology across the board will benefit from advancements, but I believe the larger and more profitable markets will see it first.

Very very very unlikely.  AMD just made a shift to 7000 series and performance (relative to 5000 series) was lackluster.   AMD won't be making a new generation for at least 18 months.  

Current 45nm FPGA compete well with GPU in all but the lowest cost areas.  28nm FPGA will eliminate any remaining advantage and ASICs will make GPU mining look as stupid as CPU mining is now.

GPU mining is dead.  Time will only worsen the relative comparison.

Note this comes from someone running a 15GH/s farm all GPU farm.

GPU being dead indicates that bitcoin is growing fast, and evolution/advancement will be inevitable.
I must say of course that a lot of faith was put in AMD 7000 series, but it didn't do well to meet
the Bitcoins market expectation in terms of hash processing power.


Regards,
BF Labs Inc.

Then FPGA is dead as well. And bitcoin market isn't growing, it's the same level like year ago, maybe a bit bigger. ASICs will make this field only for big players, which is not good for Bitcoin at all.

[rudrigorc2]

Probably noon their time.

In 4-6 weeks... sorry I had to.

lol'd
 

This is reasonable, and only time will tell.  I just can't help but feel that the GPU market (as a whole) will move faster than the FPGA market due to demand and market size.  Of course technology across the board will benefit from advancements, but I believe the larger and more profitable markets will see it first.

Very very very unlikely.  AMD just made a shift to 7000 series and performance (relative to 5000 series) was lackluster.   AMD won't be making a new generation for at least 18 months.  

Current 45nm FPGA compete well with GPU in all but the lowest cost areas.  28nm FPGA will eliminate any remaining advantage and ASICs will make GPU mining look as stupid as CPU mining is now.

GPU mining is dead.  Time will only worsen the relative comparison.

Note this comes from someone running a 15GH/s farm all GPU farm.

GPU being dead indicates that bitcoin is growing fast, and evolution/advancement will be inevitable.
I must say of course that a lot of faith was put in AMD 7000 series, but it didn't do well to meet
the Bitcoins market expectation in terms of hash processing power.


Regards,
BF Labs Inc.

Then FPGA is dead as well. And bitcoin market isn't growing, it's the same level like year ago, maybe a bit bigger. ASICs will make this field only for big players, which is not good for Bitcoin at all.

+1

[lemonz]

Probably noon their time.

In 4-6 weeks... sorry I had to.

Haha... that was awesome rjk.

Meanwhile, patiently waiting an announcement...

[P4man]

You are delusional about the capability of ASICs
[..]
Bottom line, ASICs are not an "exponential leap in hash rate, power consumption AND eventually cost".

Care to revise your opinion?
http://news.yahoo.com/butterfly-labs-announces-next-generation-asic-lineup-054626776.html

Mind you, those are initial prices. Expect them to tumble over time.

[mrb]

P4man: my latest estimations ("probably ~1000 Mhash/J at 40nm") are in line with BFL's press release claiming "3.5Ghash/s from a USB-power device (ie. 2.5W)" which would translate to 1400 Mhash/J... (What surprises me though is that it would mean they are going for 40nm instead of an older node like 90nm or 130nm.)

However, your price estimations were incorrect by a factor of 30x-40x: you said an ASIC would be "less than $1 per GH", but their press release announced $30-40 per Ghash/s.

[Vladimir]

You are delusional about the capability of ASICs
[..]
Bottom line, ASICs are not an "exponential leap in hash rate, power consumption AND eventually cost".

Care to revise your opinion?
http://news.yahoo.com/butterfly-labs-announces-next-generation-asic-lineup-054626776.html

Mind you, those are initial prices. Expect them to tumble over time.

awesome!

[Garr255]

However, your price estimations were incorrect by a factor of 30x-40x: you said an ASIC would be priced at "less than $1 per GH", but the pricing they announced is $30-40 per Ghash/s.

I can see ASICs being sold for that in the near future.

[P4man]

However, your price estimations were incorrect by a factor of 30x-40x: you said an ASIC would be "less than $1 per GH", but their press release announced $30-40 per Ghash/s.

I said it would cost BFL far less than 1$ per GH (I actually calculated somewhere between $0.1 and $1 per GH on 130nm). Of course they are taking a fat margin on that, no one expected them to release these anywhere near marginal cost when they can charge 100x more. Over time however, their price will approach marginal cost.

P4man: my latest estimations ("probably ~1000 Mhash/J at 40nm") are in line with BFL's press release claiming "3.5Ghash/s from a USB-power device (ie. 2.5W)" which would translate to 1400 Mhash/J...

Changing goalposts . I responded to this:

"40nm ASICs will do about 200 Mhash/Joule."

And I have no reason to believe this asic is 40nm. 65nm would surprise me, 90 or 130nm seems much more likely.

[mrb]

Why reply to my old incorrect post? I already showed you how I computed a revised estimate of ~1000 Mhash/J at 40nm... Our difference of opinion is that you think they do it at 90nm/130nm, whereas I maintain this should only be possible at 40nm or better. Knowing BFL, they will never disclose this detail. We will not know until someone decap the chip.

About cost vs price: same thing, we will never know their true cost. But the fact they price the hashing speed "only" ~10x better than FPGAs (mini rig = 1.65 Mhash/s/$) is one of the reasons why I said ASICs won't be an exponential leap. (I was talking about price, you about cost.) ASICs manufacturers will initially price their product to be "better" than FPGAs, but not "exponentially better".

[P4man]

About cost vs price: same thing, we will never know their true cost. But the fact they price the hashing speed "only" ~10x better than FPGAs (mini rig = $1.65 Mhash/s/$) is one of the reasons why I said ASICs won't be an exponential leap.

Its rather the opposite. The fact they intend to launch at 10x better MH/$, while they could easily charge 5 or even 10x more considering the power efficiency gains, shows what kind of tectonic shift this really represents.  I expected something much close to FPGA prices initially. But these prices will only last until price/difficulty catches up, therefore asic sales would dry up, so BFL will slash them over and over until they are >100x lower than today and finally approach somewhere closer to marginal costs which is damn close to zero $ per GH.

That is the 'exponential' leap. Want to take bets that in 24 months, bitcoin difficulty wont be 100x higher than today?

[Mageant]

ASIC mining is a good thing IMHO.

There has often been the argument against Bitcoin that it "wastes" a lot of electricity.

With widespread ASCI (and FPGA) mining with their low electricity consumptions this takes away this argument against Bitcoin.

Also, it gives people who live in areas with high electricity costs the chance to take up mining profitably again.

[mrb]

Its rather the opposite. The fact they intend to launch at 10x better MH/$, while they could easily charge 5 or even 10x more considering the power efficiency gains, shows what kind of tectonic shift this really represents.

You keep trying to make it sound like ASICs are different, but 10x better Mhash/s/$ has already been seen in the past: when we moved from CPU to GPU mining.

 I expected something much close to FPGA prices initially. But these prices will only last until price/difficulty catches up, therefore asic sales would dry up, so BFL will slash them over and over until they are >100x lower than today and finally approach somewhere closer to marginal costs which is damn close to zero $ per GH.
That is the 'exponential' leap. Want to take bets that in 24 months, bitcoin difficulty wont be 100x higher than today?

BFL will never be able to slash their prices by 100x because the other assembly/component costs would make up most of the cost. Think about it, if they did, the BitForce SC Single would sell for $12.99. But the case, PCB, fan, power adapter, etc alone cost at least $20-30.

Also, assuming that the capital invested in GPUs by miners (as of today) is fully reinvested in ASIC mining products, it would be sufficient to increase the difficulty by 10x-25x, because ASICs achieve 10x-25x more Mhash/s/$ than GPUs. And I think the invested capital is likely to double in the next 24 months, so difficulty could increase to about 50x. And that's without BFL having to slash their ASIC prices at all. (50x is too close to your 100x, so no I won't take the bet.)

However, I would be ready to bet on the fact that BFL will not slash their prices by more than, say, 30% before June 2014. For once, I expect them to take at least 1 year to develop and ship the first ASIC product.

[P4man]

You keep trying to make it sound like ASICs are different, but 10x better Mhash/s/$ has already been seen in the past: when we moved from CPU to GPU mining.

You keep failing to see the massive difference. GPU or FPGA pricing is not dependent on bitcoin price/difficulty.  If xilinx or AMD would somehow double or 10 fold their price/performance, you would expect a tenfold decrease in P/D. The other way around, there is zero impact. P/D skyrocketing for whatever reason will do next to nothing to AMD or Xilinx prices. IOW, GPU or FPGA performance/$ is a given and relatively constant, and it results in a relatively predictable P/D, not the other way around.

ASIC performance/$ is completely dependent on bitcoin P/D AND will completely determine it. This feedback loop combined with the enormous potential for price drops is what makes it completely and utterly different. 10x better MH/$ today will become 100x better tomorrow. And as a result, your ROI time will increase 10x too.

BFL will never be able to slash their prices by 100x because the other assembly/component costs would make up most of the cost. Think about it, if they did, the BitForce SC Single would sell for $12.99. But the case, PCB, fan, power adapter, etc alone cost at least $20-30.

Im sure you can think of other ways to increase MH/$ than a $13 device.

However, I would be ready to bet on the fact that BFL will not slash their prices by more than, say, 30% before June 2014. For once, I expect them to take at least 1 year to develop and ship the first ASIC product.

Thats actually why I used a 24 month period, because Im also not expecting them to ship any asics this year. But 30%? Hey, cant let that pass by. Lets try this: 12 months after BFL shipped their first SC mini rig, the price/MH will be lowered by 50% or more per MH.

[kibblesnbits]

You are delusional about the capability of ASICs
[..]
Bottom line, ASICs are not an "exponential leap in hash rate, power consumption AND eventually cost".

Care to revise your opinion?
http://news.yahoo.com/butterfly-labs-announces-next-generation-asic-lineup-054626776.html

Mind you, those are initial prices. Expect them to tumble over time.

No last names in the PR release? 

[mrb]

However I do agree that Mhash/dollar will be a more interesting metric to watch than Mhash/J. I wonder why you think ASIC will contribute a 1000x improvement in this area (going from $1 per Mh/s to $1 per Gh/s)?

Basing on the square mm and a clock speed of 1 Ghz the raw manufacturing cost would be closer to $0.10 per GH/s.  Now granted you have the NRE, the capital cost, the profit magins, yield losses, salaries, etc but even with 1000% markup <$1 per GH/s would be possible.

One way to look at it is the SHA-256 hasher only took 20kGE.    Lets say scaling it to 1 Ghz required twice as many GE and you want to make it perform a double hash; so 80 kGE. Obviously you wouldn't make a chip that small.  But hashing is perfectly parallel.  Instead of 1 single hashing engine running 1Ghz you could lay down 20 parallel engines.  So that on each clock 20 nonces are calculated simultaneously (20 GH/s @ 1Ghz).  Even that would only be ~ 1.6M GE.  Tiny small by modern chip standards (which have transistor counts in the billions).   The $20 CPU in your smartphone likely has a higher transistor count.

Your assumption that running at a higher frequency is better is wrong. It is better to have many tiny 20kGE SHA-256 units running at a slower clock, than fewer bigger ones, for energy efficiency. But this does not change the conclusion...

I changed my mind. I don't know why I did not run this math myself, but I now agree that a 100x better Mhash/s per dollar than BFL's ASIC announcement is achievable with current tech 40-60nm ASICs.

However, this would be doable only at sufficiently large production volumes. With thousands of wafers, so millions of chips. The question is: will Bitcoin mining chips ever be produced in the millions?

[DeathAndTaxes]

However, this would be doable only at sufficiently large production volumes. With thousands of wafers, so millions of chips. The question is: will Bitcoin mining chips ever be produced in the millions?

Will Bitcoin ever be used by millions of users.  Maybe, maybe not.  But if it was I could see someday the network consisted of millions of low GH/s devices that are essentially plug and play. 

Imagine something as simple as small box with Ethernet port and power jack.  The back side of it has a QR code w/ private key for importing into a wallet and it just hashes 24/7 solo sending those rewards to its static address.  No drivers, no SDK just very high electrical efficiency and simplicity.   People who don't want the relatively small rewards could donate the private key to open source development, or a bitcoin non-profit.