ASICMINER: Entering the Future of ASIC Mining by Inventing It - page 538.

empoweoqwj

hero member

Activity: 518

Merit: 500

Quote from: Entropy-uc on December 27, 2013, 01:52:42 PM

Friedcat definitely missed a huge opportunity on the sales side. It was obvious in March that the demand for hardware would persist into the 10s of Ph/s. ASICMINER could have supplied a very large chunk of that.

I'm sure it was amusing selling hardware at >$500 / Gh/s to fools that would never see BTC breakeven on the gear. But he could have just as easily sold several Ph/s at $50 to smart investors who stood on the sidelines and kept their money in hand. That would have been >$100M in sales.

I actually contacted Friedcat regarding this direction as my family had resources in China and the US to support distribution at this scale. Even though I was a large shareholder at the time, he never bothered to respond.

Now he's planning to come to market with a 40 nm product when at least 5 groups will have a 28 nm device available before him. Talk about bringing a knife to a gun fight!

Not sure how FC could have supplied huge amounts of hash rate with their old hardware? We all wish they could have waved a magic wand and supplied all demand but business doesn't work like that. Where was he going to get several Ph/s from Huh

Equilux

sr. member

Activity: 353

Merit: 251

Quote from: Mabsark on December 27, 2013, 07:59:08 PM

Quote from: Equilux on December 27, 2013, 06:35:06 PM

Quote from: Mabsark on December 27, 2013, 06:28:58 PM

Quote from: jimmothy on December 27, 2013, 06:19:00 PM

Quote from: Mabsark on December 27, 2013, 06:07:08 PM

Quote from: jimmothy on December 27, 2013, 05:47:16 PM

Eta for next gen tape-out is jan 20th

Which does not mean next gen chips will be available then, in case that's what some of you were thinking. Sample chips will arrive about 2-3 months after tape out. For example, HashFast taped-out on 2013-09-05, the first wafers were completed on 2013-11-08, and sample chips were received on 2013-12-01.

So, next gen AM chips will likely be ready in March/April.

It is not really fair to compare asicminer to hashfast. However march/april does seem like a reasonable estimate for gen3 deployment.

I'm not comparing AM to HashFast, I'm just showing how long it typically takes for foundries to produce the wafers and for those wafers to be packaged into chips. The process is pretty much out of AM's or HashFast's hands until they get the samples back.

Wans't the tape-out from KnC 3 weeks before they delivered a working product? Or am I misremembering?

I couldn't find a date for KnC's tape out, but there is the following:

Quote from: kncminer on June 26, 2013, 09:14:10 AM

All,

I would like to say that EVERYTHING that needs to be done at this point in time for a September delivery has indeed been done.
We are on track for delivery and we know of nothing currently to the contrary.

There is not a single part in our systems that has a longer delivery time than the time we still have available.

Yes that does imply that things that don’t need to be ordered yet have not been ordered. The reason behind that is we simply don’t need to order them yet, We do have contracts with all suppliers we need to have contracts with Those contracts contain delivery times and the associated order dates by which we must confirm the amounts (and of course pay).

I hope this removes confusion

KnCMiner

and

Quote from: kncminer on July 30, 2013, 08:41:06 AM

Hi Guys

We would like to say that. Yes....we did tape-out of the ASIC some time ago. The exact details of the ASIC schedule, volumes, vendor etc. is kept internal since it's very sensitive information towards competitors. The product development tasks are on-track."

We are still on schedule to begin shipping in September.

Thanks
KnCMiner Team

So, tape out was likely around 2013-06-26 and 2013-07-30 at the latest.

From the KNC website:

Quote

Chip News
9/17/2013 6:29:00 AM

You have all been waiting for an update on our chips and we can today announce that the Fabrication process has finished and the wafers are on their way to the packing assembly house right now. They will be in the assembly house for a few days before they make their way to us in Sweden and of course via the fastest method possible.

and

Quote

Production News
9/30/2013 7:51:00 AM

We have relocated most of our operation to the factory over the weekend. The first delivery of the chips have arrived and have been populated on the boards. We have begun testing and initial results are good, we will have more updates throughout the day. We know many of you are calling and we are sorry if we cannot answer your call. This is a very busy time for us but we are all hands on deck we can assure you of that.

So, wafers were completed around 9/17/2013 and sample chips arrived on 9/30/2013. That's around 3 months between tape out and getting sample chips back.

Great digging there, thanks for the solid reply!

Mabsark

legendary

Activity: 826

Merit: 1004

Quote from: Equilux on December 27, 2013, 06:35:06 PM

Quote from: Mabsark on December 27, 2013, 06:28:58 PM

Quote from: jimmothy on December 27, 2013, 06:19:00 PM

Quote from: Mabsark on December 27, 2013, 06:07:08 PM

Quote from: jimmothy on December 27, 2013, 05:47:16 PM

Eta for next gen tape-out is jan 20th

Which does not mean next gen chips will be available then, in case that's what some of you were thinking. Sample chips will arrive about 2-3 months after tape out. For example, HashFast taped-out on 2013-09-05, the first wafers were completed on 2013-11-08, and sample chips were received on 2013-12-01.

So, next gen AM chips will likely be ready in March/April.

It is not really fair to compare asicminer to hashfast. However march/april does seem like a reasonable estimate for gen3 deployment.

I'm not comparing AM to HashFast, I'm just showing how long it typically takes for foundries to produce the wafers and for those wafers to be packaged into chips. The process is pretty much out of AM's or HashFast's hands until they get the samples back.

Wans't the tape-out from KnC 3 weeks before they delivered a working product? Or am I misremembering?

I couldn't find a date for KnC's tape out, but there is the following:

Quote from: kncminer on June 26, 2013, 09:14:10 AM

All,

I would like to say that EVERYTHING that needs to be done at this point in time for a September delivery has indeed been done.
We are on track for delivery and we know of nothing currently to the contrary.

There is not a single part in our systems that has a longer delivery time than the time we still have available.

Yes that does imply that things that don’t need to be ordered yet have not been ordered. The reason behind that is we simply don’t need to order them yet, We do have contracts with all suppliers we need to have contracts with Those contracts contain delivery times and the associated order dates by which we must confirm the amounts (and of course pay).

I hope this removes confusion

KnCMiner

and

Quote from: kncminer on July 30, 2013, 08:41:06 AM

Hi Guys

We would like to say that. Yes....we did tape-out of the ASIC some time ago. The exact details of the ASIC schedule, volumes, vendor etc. is kept internal since it's very sensitive information towards competitors. The product development tasks are on-track."

We are still on schedule to begin shipping in September.

Thanks
KnCMiner Team

So, tape out was likely around 2013-06-26 and 2013-07-30 at the latest.

From the KNC website:

Quote

Chip News
9/17/2013 6:29:00 AM

You have all been waiting for an update on our chips and we can today announce that the Fabrication process has finished and the wafers are on their way to the packing assembly house right now. They will be in the assembly house for a few days before they make their way to us in Sweden and of course via the fastest method possible.

and

Quote

Production News
9/30/2013 7:51:00 AM

We have relocated most of our operation to the factory over the weekend. The first delivery of the chips have arrived and have been populated on the boards. We have begun testing and initial results are good, we will have more updates throughout the day. We know many of you are calling and we are sorry if we cannot answer your call. This is a very busy time for us but we are all hands on deck we can assure you of that.

So, wafers were completed around 9/17/2013 and sample chips arrived on 9/30/2013. That's around 3 months between tape out and getting sample chips back.

Entropy-uc

hero member

Activity: 756

Merit: 501

Tapeout to finished silicon is ~45 days with a hot lot. After that you have bumping and packaging, board assembly and test, etc.

Equilux

sr. member

Activity: 353

Merit: 251

Quote from: Mabsark on December 27, 2013, 06:28:58 PM

Quote from: jimmothy on December 27, 2013, 06:19:00 PM

Quote from: Mabsark on December 27, 2013, 06:07:08 PM

Quote from: jimmothy on December 27, 2013, 05:47:16 PM

Eta for next gen tape-out is jan 20th

Which does not mean next gen chips will be available then, in case that's what some of you were thinking. Sample chips will arrive about 2-3 months after tape out. For example, HashFast taped-out on 2013-09-05, the first wafers were completed on 2013-11-08, and sample chips were received on 2013-12-01.

So, next gen AM chips will likely be ready in March/April.

It is not really fair to compare asicminer to hashfast. However march/april does seem like a reasonable estimate for gen3 deployment.

I'm not comparing AM to HashFast, I'm just showing how long it typically takes for foundries to produce the wafers and for those wafers to be packaged into chips. The process is pretty much out of AM's or HashFast's hands until they get the samples back.

Wans't the tape-out from KnC 3 weeks before they delivered a working product? Or am I misremembering?

Mabsark

legendary

Activity: 826

Merit: 1004

Quote from: jimmothy on December 27, 2013, 06:19:00 PM

Quote from: Mabsark on December 27, 2013, 06:07:08 PM

Quote from: jimmothy on December 27, 2013, 05:47:16 PM

Eta for next gen tape-out is jan 20th

Which does not mean next gen chips will be available then, in case that's what some of you were thinking. Sample chips will arrive about 2-3 months after tape out. For example, HashFast taped-out on 2013-09-05, the first wafers were completed on 2013-11-08, and sample chips were received on 2013-12-01.

So, next gen AM chips will likely be ready in March/April.

It is not really fair to compare asicminer to hashfast. However march/april does seem like a reasonable estimate for gen3 deployment.

I'm not comparing AM to HashFast, I'm just showing how long it typically takes for foundries to produce the wafers and for those wafers to be packaged into chips. The process is pretty much out of AM's or HashFast's hands until they get the samples back.

jimmothy

hero member

Activity: 770

Merit: 509

Quote from: Mabsark on December 27, 2013, 06:07:08 PM

Quote from: jimmothy on December 27, 2013, 05:47:16 PM

Eta for next gen tape-out is jan 20th

Which does not mean next gen chips will be available then, in case that's what some of you were thinking. Sample chips will arrive about 2-3 months after tape out. For example, HashFast taped-out on 2013-09-05, the first wafers were completed on 2013-11-08, and sample chips were received on 2013-12-01.

So, next gen AM chips will likely be ready in March/April.

It is not really fair to compare asicminer to hashfast. However march/april does seem like a reasonable estimate for gen3 deployment.

Mabsark

legendary

Activity: 826

Merit: 1004

Quote from: jimmothy on December 27, 2013, 05:47:16 PM

Eta for next gen tape-out is jan 20th

Which does not mean next gen chips will be available then, in case that's what some of you were thinking. Sample chips will arrive about 2-3 months after tape out. For example, HashFast taped-out on 2013-09-05, the first wafers were completed on 2013-11-08, and sample chips were received on 2013-12-01.

So, next gen AM chips will likely be ready in March/April.

explorer

legendary

Activity: 2016

Merit: 1259

Quote from: jimmothy on December 27, 2013, 05:47:16 PM

The current status of AM is debatable. But I think we can all agree you bought at the worst possible time. 4btc/piece would require like 40% network hashrate which will probably never happen.

Assuming zero profits from sales. It reached 4+ on the back of sales. Granted, that window is only open by a crack compared to last summer, but still counts for something. 4+ is a dream, perhaps, but 10% network + some sales shouldn't be unmanageable. Another point would be newer investors, happy with less than 33% APR - that could raise the price as well.

jimmothy

hero member

Activity: 770

Merit: 509

Quote from: glendall on December 27, 2013, 05:40:07 PM

How's asicminer doing these days?

I was an investor from around 4 btc a share down to .8 a share, haven't been keeping on the news.

Any eta on the next gen chips?

The current status of AM is debatable. But I think we can all agree you bought at the worst possible time. 4btc/piece would require like 40% network hashrate which will probably never happen.

Eta for next gen tape-out is jan 20th

glendall

legendary

Activity: 2156

Merit: 1018

Buzz App - Spin wheel, farm rewards

How's asicminer doing these days?

I was an investor from around 4 btc a share down to .8 a share, haven't been keeping on the news.

Any eta on the next gen chips?

Puppet

legendary

Activity: 980

Merit: 1040

Quote from: jimmothy on December 27, 2013, 04:44:20 PM

Thought I read somewhere that some mobile cpus were using 14nm but I realize that is not the case. Question answered thanks.

You probably read the deal Apple made with Samsung and TSMC for 14nm chips. But you missed the date, that process is not expected to be online before mid 2015.

jimmothy

hero member

Activity: 770

Merit: 509

Quote from: Puppet on December 27, 2013, 04:39:36 PM

Quote from: jimmothy on December 27, 2013, 04:29:23 PM

maybe you could explain why intel is seemingly so slow at adopting new manufacturing tech?

Lol? Intel is miles ahead of everyone else. While TSMC is still busy ordering 20nm production equipment (good luck KnC Neptune customers), Globalfoundries apparently stuck at 28nm, intel has been producing 14nm chips for some time. Not sure they started selling these chips, they did have yield issues, but that doesnt mean they are somehow behind.

/OT

Thought I read somewhere that some mobile cpus were using 14nm but I realize that is not the case. Question answered thanks.

Puppet

legendary

Activity: 980

Merit: 1040

Quote from: jimmothy on December 27, 2013, 04:29:23 PM

maybe you could explain why intel is seemingly so slow at adopting new manufacturing tech?

Lol? Intel is miles ahead of everyone else. While TSMC is still busy ordering 20nm production equipment (good luck KnC Neptune customers), Globalfoundries apparently stuck at 28nm, intel has been producing 14nm chips for some time. Not sure they started selling these chips, they did have yield issues, but that doesnt mean they are somehow behind.

/OT

jimmothy

hero member

Activity: 770

Merit: 509

Quote from: Entropy-uc on December 27, 2013, 03:35:41 PM

Friedcat is a smart guy. But it appears to me that his team has made more money than they ever dreamed of already, and don't particularly care about being the best any longer.

I disagree. Why would FC not want to be the best and make more money? He is going down what he feels is the most profitable path.

Side question: Having worked for intel maybe you could explain why intel is seemingly so slow at adopting new manufacturing tech? Why is Intel just now adopting 22nm when 14nm tech is possible?

Puppet

legendary

Activity: 980

Merit: 1040

Quote from: Entropy-uc on December 27, 2013, 03:35:41 PM

I expect the network to be 100 Ph/s by the end of march. It could happen sooner if the rumors I hear about silicon valley stealth operations turn out to be true.

Quite a bold prediction. I thought I was the resident pessimist lol.

Quote

What bitfury did was a herculean task. He hand designed every transistor for the bitcoin proof of work.

I understood friedcat (or whomever he hired) were doing similar hand optimizations. Now I happen to agree with you that was a strategic blunder. The time wasted doing that will cost far more than it will ever return. Paying a higher NRE for a smaller process node to get similar performance to market faster, KnC style, would have been my obvious choice. But thats water under the bridge, and now tape out is close, if not already behind us now? If they can indeed come up with a chip that uses on the order of 0.2W/J in the next 2 or 3 months, it doesnt really matter if thats achieved on 40 or 20nm. They will have a design thats going to be more than competitive for the foreseeable future (granted, in bitcoin world, that isnt very long

).

Entropy-uc

hero member

Activity: 756

Merit: 501

Quote from: Puppet on December 27, 2013, 03:12:04 PM

Quote from: Entropy-uc on December 27, 2013, 02:35:28 PM

40 nm power efficiency will be completely un-marketable against 28 nm.

It will be a while before that is the case. By the time power consumption will become the major factor, prices and margins will have tumbled enormously. I have no doubt that will happen, but not in Q1 and not before the network is over ~100PH.

Besides, current 28nm designs dont seem to pay particular attention to power efficiency. Its clear the focus was on time to market. Testatement to that is that 55nm bitfury's are in the same ball park as 28nm chips. A similarly well optimized 40nm chip should be competitive with existing and upcoming 28nm chips. I seem to remember Friedcat estimating 0.2J/GH ? If he manages that, he is better than any announced 28nm chip and probably better than KnC's future 20nm chip, which Im not expecting anytime soon.

TL;DR. The clock is ticking rapidly, but there is still time to extract a nice profit.

I expect the network to be 100 Ph/s by the end of march. It could happen sooner if the rumors I hear about silicon valley stealth operations turn out to be true.

What bitfury did was a herculean task. He hand designed every transistor for the bitcoin proof of work. And it is buggy as hell with 20% invalids due to design errors as I understand things. It sets a standard for everyone else to aim for though. And you can count on it that every design coming to market has benchmarked against that standard. Even if ASICMINER matches that design in 40 nm, he's going to be overwhelmed by the sheer volume of 28 nm devices. And on the retail side, you have customers that were ripped off (by their own choice) at prices that would never break even, and a lousy reputation because of the very high rate of failures being reported now.

Friedcat is a smart guy. But it appears to me that his team has made more money than they ever dreamed of already, and don't particularly care about being the best any longer.

Puppet

legendary

Activity: 980

Merit: 1040

Quote from: Entropy-uc on December 27, 2013, 02:35:28 PM

40 nm power efficiency will be completely un-marketable against 28 nm.

It will be a while before that is the case. By the time power consumption will become the major factor, prices and margins will have tumbled enormously. I have no doubt that will happen, but not in Q1 and not before the network is over ~100PH.

Besides, current 28nm designs dont seem to pay particular attention to power efficiency. Its clear the focus was on time to market. Testatement to that is that 55nm bitfury's are in the same ball park as 28nm chips. A similarly well optimized 40nm chip should be competitive with existing and upcoming 28nm chips. I seem to remember Friedcat estimating 0.2J/GH ? If he manages that, he is better than any announced 28nm chip and probably better than KnC's future 20nm chip, which Im not expecting anytime soon.

TL;DR. The clock is ticking rapidly, but there is still time to extract a nice profit.

jimmothy

hero member

Activity: 770

Merit: 509

Quote from: ?? on ??

Quote from: jimmothy on December 27, 2013, 02:39:31 PM

Quote from: Entropy-uc on December 27, 2013, 02:35:28 PM

Quote from: superduh on December 27, 2013, 02:23:53 PM

Quote from: Entropy-uc on December 27, 2013, 01:52:42 PM

Friedcat definitely missed a huge opportunity on the sales side. It was obvious in March that the demand for hardware would persist into the 10s of Ph/s. ASICMINER could have supplied a very large chunk of that.

I'm sure it was amusing selling hardware at >$500 / Gh/s to fools that would never see BTC breakeven on the gear. But he could have just as easily sold several Ph/s at $50 to smart investors who stood on the sidelines and kept their money in hand. That would have been >$100M in sales.

I actually contacted Friedcat regarding this direction as my family had resources in China and the US to support distribution at this scale. Even though I was a large shareholder at the time, he never bothered to respond.

Now he's planning to come to market with a 40 nm product when at least 5 groups will have a 28 nm device available before him. Talk about bringing a knife to a gun fight!

missed opportunity, sure. but 40 vs 28 chip isn't really a big deal if the 40 nm is designed well. he'll go on a smaller size later as will everyone

40 nm power efficiency will be completely un-marketable against 28 nm.

And ASICMINER's facilities will just be spit in a bucket for internal deployment.

Source?

Last time 40nm vs 28nm was debated it was concluded that the process size doesn't mean shit when it comes to final hardware specs. 28nm may have higher electricity efficiency if optimized but in the end it is really just a gimmick.

Two decades working for Intel.

There is no question that design efficiency can have a significant impact. But 28 nm starts out with a 30% advantage. On both power and clock speed. Out of 5 teams with 28 nm designs, at least a couple will utilize the process node at least as well as Friedcat.

On top of that, time is eroding the reward for deployment. It's fairly easy to calculate the power available in ASICMINER's facilities. Fill that with 40 nm hardware in March and you get network share that is respectable but isn't going to support the stock.

Is a 30% advantage really worth the nre cost and potential delays? I think it will come down to the cost/gh and it is very possible that 40nm provides that optimum ratio.

I do agree that eventually AM will have to jump on the 28nm train but that will be only when the competition has improved on their 28nm designs which could be 6+ months.

I am just mot convinced that rushing to 28nm provides a significant benefit over a fully optimized 40nm chip.

If AM 110nm tech can compete cost wise with knc 28nm, how will 40nm compare?

damiano

legendary

Activity: 1246

Merit: 1000

103 days, 21 hours and 10 minutes.

Quote from: Puppet on December 27, 2013, 04:48:15 AM

I know its rhetorical question, but allow me to answer this one nonetheless

Quote from: Endlessa on December 27, 2013, 12:27:13 AM

They did something awesome once, but can they do it again?

The awesome thing they did was being the first (and for quite some time, only) industrial bitcoin mining operation producing their own asics, at a time when almost no such chips existed. That is something no one will ever be able to do again.

Also, I would put this:

Quote

Do they have enough liquidity to produce their now called "3rd Generation" chips (even though there has only been 1 gen for AM)

Firmly in the "pro" list, rather than with the "cons". AM has a huge stockpile of money, thats a serious advantage in any industry. They dont have to worry about finding funding or taking in enough preorders to pay for some NRE. THey can just do what they think makes sense and if need be, think big. They even have enough money to develop several alternatives (be it chip designs or whatever) in parallel, if they think that makes sense. I dont see how that is a negative.

Well said

Topic: ASICMINER: Entering the Future of ASIC Mining by Inventing It - page 538. (Read 3917468 times)