Pages:
Author

Topic: 0.19 J/GH - CoinBau looks for investors in German mining technology - page 5. (Read 22388 times)

full member
Activity: 129
Merit: 100
This is a forum for this manufacturer so I don't think it's ethical to quote other peoples work, but if you're curious why not ask some of the current 28nm based suppliers to give some data on what their chips run at in sub threshold? I think you might be surprised. The GH/mm2 is a useless metric in my opinion, the only reason used it was because Coinbau did - you can only base analysis on data a manufacturer gives you.

No offence, but isn’t the GH/s/mm² metric not directly related to the production costs of the pure silicon in terms of $/GH?
From my understanding, e.g. somebody who is able to realize 0.19 J/GH @ 0.7 GH/s/mm² in 28nm (instead of 1.4 GH/s/mm²) would have to pay roughly twice the $/GH/s price for the ASIC die.

With respect to the main competitor 28nm ASICs I see here a design which will have almost the same $/GH/s price in production (if they could really realize this specification), while only consuming half of the energy. Please correct me if I’m wrong.
 
sr. member
Activity: 441
Merit: 250
Forgot to add, if we take your 'nominal' chip speed as 100GH/s, as per your website, and your GH/mm2 (Huh) as 1.4, then your core chip size would be 70mm2, or approximately 79 mm2 die size, yes? Also means the chip has to dissipate 19W, but that's entirely feasible.

For that hashrate versus die area in 28nm, it suggests strongly that your chip isn't full custom at all, it's more likely a regular standard cell implementation. 70 mm2 gives a rough 'equivalent' gate count of around 115M allowing for your control circuit. Divided that by the 450k gate equivalents needed per pipeline means there are, surpirise, surprise 256 pipelines per chip.

To get your 100GH/sec it needs to run at 400 MHz, which for 28nm means you're either being ultra conservative in your chip timing, it's a really bad design - which I doubt - or it's running sub threshold which is absolutely fine, of course, but these calculations are based on standard designs.

In short, you're asking your investors to pay for the NRE & production of a chip which on the face of it doesn't offer any real advantage over solutions that are available from competitors.

Possibly I'm missing something here?



Very nice analysis, seams feasible. Sorry, maybe I missed something, but I don't know a competitor who offers 0.19 J/GH at 1.4 GH/mm2. Source?

This is a forum for this manufacturer so I don't think it's ethical to quote other peoples work, but if you're curious why not ask some of the current 28nm based suppliers to give some data on what their chips run at in sub threshold? I think you might be surprised. The GH/mm2 is a useless metric in my opinion, the only reason used it was because Coinbau did - you can only base analysis on data a manufacturer gives you.
full member
Activity: 129
Merit: 100
full member
Activity: 129
Merit: 100
What have stopped you from ordering more wafers of Gen1 design from GF and produce more miners during the last eight months? Even if you were doing MPW for Gen1, you have the time to make profit from Gen1 design.

If it were an MPW, the answer is quite easy.
An MPW wafer normally contains only one of your dies per reticle. In this case they would have not more than 200 working dies from one wafer.
Taking the price of an 28nm wafer of about $6k (probably much more in low volume), they would have a price of more than $3 per GH/s for their gen1 for the pure silicon.


newbie
Activity: 17
Merit: 0
Hello all,

CoinBau will be German Incorporated Company (Aktiengesellschaft): its shares are freely tradeable but not at stock exchange. So, yes, with an investment of $10,000 you can get a dividend payout after year 3 (and in each following year) of about $3300 plus the grown value of the shares.

About the technical stuff:
We never said we have a full custom implementation. In fact it is a combination of standard cell and few important special cells which were customly optimized.
Two factors are relevant for economical success of a miner: the acquisition costs in $/GH/s and the operational costs depending on J/GH. The goal is the optimization of the complete miner: bring as much hash power as possible into one case which can be supplied with a standard PSU at reasonable costs. The goal is not to have the fastest chip, not to have the biggest GH/mm². The die prize is only one part and not even the biggest of the production costs. Thus, economically it is better to put a few mm² of silicon more onto the PCB, run it at lowest possible voltage (voltage is quadratic in power consumption) resulting in very low energy consumption per GH. The quadratic influence of voltage on power is pretty important because raising voltage for higher frequency results in more energy consumption per GH. So, it ends up in a trade-off between additional energy costs (high frequency and voltage) or additional silicon area costs. We calculated a lot and since the silicon is only part of the production costs we came to the conclution it is better to utilize more silicon area and reduce energy costs.
So, I only can ask: do you know a competitor offering 0.19 J/GH at less than $1/GH/s?

Additionally: by investing into CoinBau you save all margins for hardware sellers or cloudhashing brokers for your own rate of return.

In an MPW run you have only a few chips per wafer and, thus, the costs per chip are pretty high and dominate everything. Since the production of a wafer takes 2 months or more, in March or April we wouldn't have been able to get a return of investment with this first gen hardware. We would have needed a full mask tapeout but did not want to waste the money on an old technology when our 2nd generation was nearly final.

Best regards,
Markus
legendary
Activity: 1666
Merit: 1185
dogiecoin.com
What have stopped you from ordering more wafers of Gen1 design from GF and produce more miners during the last eight months? Even if you were doing MPW for Gen1, you have the time to make profit from Gen1 design.

This atm. If even after successful tapeout of the new gen you're only able to retail for $1/GH, why stop with what yo had so early?
donator
Activity: 1120
Merit: 1001
What have stopped you from ordering more wafers of Gen1 design from GF and produce more miners during the last eight months? Even if you were doing MPW for Gen1, you have the time to make profit from Gen1 design.
full member
Activity: 129
Merit: 100
Forgot to add, if we take your 'nominal' chip speed as 100GH/s, as per your website, and your GH/mm2 (Huh) as 1.4, then your core chip size would be 70mm2, or approximately 79 mm2 die size, yes? Also means the chip has to dissipate 19W, but that's entirely feasible.

For that hashrate versus die area in 28nm, it suggests strongly that your chip isn't full custom at all, it's more likely a regular standard cell implementation. 70 mm2 gives a rough 'equivalent' gate count of around 115M allowing for your control circuit. Divided that by the 450k gate equivalents needed per pipeline means there are, surpirise, surprise 256 pipelines per chip.

To get your 100GH/sec it needs to run at 400 MHz, which for 28nm means you're either being ultra conservative in your chip timing, it's a really bad design - which I doubt - or it's running sub threshold which is absolutely fine, of course, but these calculations are based on standard designs.

In short, you're asking your investors to pay for the NRE & production of a chip which on the face of it doesn't offer any real advantage over solutions that are available from competitors.

Possibly I'm missing something here?



Very nice analysis, seams feasible. Sorry, maybe I missed something, but I don't know a competitor who offers 0.19 J/GH at 1.4 GH/mm2. Source?
sr. member
Activity: 441
Merit: 250
Forgot to add, if we take your 'nominal' chip speed as 100GH/s, as per your website, and your GH/mm2 (Huh) as 1.4, then your core chip size would be 70mm2, or approximately 79 mm2 die size, yes? Also means the chip has to dissipate 19W, but that's entirely feasible.

For that hashrate versus die area in 28nm, it suggests strongly that your chip isn't full custom at all, it's more likely a regular standard cell implementation. 70 mm2 gives a rough 'equivalent' gate count of around 115M allowing for your control circuit. Divided that by the 450k gate equivalents needed per pipeline means there are, surpirise, surprise 256 pipelines per chip.

To get your 100GH/sec it needs to run at 400 MHz, which for 28nm means you're either being ultra conservative in your chip timing, it's a really bad design - which I doubt - or it's running sub threshold which is absolutely fine, of course, but these calculations are based on standard designs.

In short, you're asking your investors to pay for the NRE & production of a chip which on the face of it doesn't offer any real advantage over solutions that are available from competitors.

Possibly I'm missing something here?

sr. member
Activity: 441
Merit: 250
Hello all,

There have been some questions concerning our technology and why we are able to realize a power efficiency others can't. It's time to answer them:

Our hashing core is a fully pipelined architecture whereas our key differentiator is a very area efficient custom pipeline implementation which we developed within the last 6 months based on a so called cell based design approach. We combined standard cells with custom optimized complex cells which save area and power.
Our high density core enabled us to realize much more hashing cores on the same silicon area, running at quite low frequencies. The low performance requirements of a single core allow us to apply an ultra-low supply voltage, which is much lower than the standard at this GLOBALFOUNDRIES 28nm process node. Therefore we optimized and re-characterized our in-house standard cell library for such low supply voltages. To ensure that this will work first-time right, we developed the complete concept of this ultra-low voltage design in close cooperation with the GLOBALFOUNDRIES design enablement team here in Dresden, which has detailed knowledge of the selected 28nm technology produced here in Dresden too.
Our unique design achieves in nominal mode 1.4 GH/mm² and 0.19 J/GH. However, similar to other hardware providers, we can change the frequency of the hashing cores in a high performance mode realizing 2.1 GH/mm² at an energy consumption of about 0.29 J/GH.

If there are any questions to be answered, do not hestitate to ask.

Best regards,
Markus


Much though I hate to sound like a party pooper, running in sub threshold mode is nothing new or innovative, it's an idea that's been around for decades and used in many commercial chips. It's the basic principle used in the Bitfury asics.

Low energy useage is fine, but it slows your system down considerably. To get the highest hashing speed (and hence earnings) if you're getting cheap energy then it makes much more sense to go faster unless your hardware is very cheap. Your own figures show that for a nominal die size you would need 1.5 times as much area to get the same hashing speed per second in nominal mode versus high performance mode, and so it would take the same amount of energy except you need to buy more die space!
member
Activity: 92
Merit: 10
Hello,

we expect less than $1 per GH/s retail costs and even less production costs based on our experience with our generation 1 miners.

The annual general meeting of all shareholders decides on the dividend payout based on the achieved profit, so we cannot preset the dividend payouts now. However, we would like to pay 50% of the profit (in our scenario these are about 10 million US dollar) as dividend to the shareholders after year 3. The rest we want to use for reinvestments. In the following years the profit and dividend payout will be even higher - assumed Bitcoin stays at least stable.
We know, we will require two years to build up the mine by reinvestment of the mined Bitcoins in order to achieve significant market share. That's the reason for no dividend payout in the first two years.

In order to build up the mine with a reasonable growth rate our calculations recommend a total of 15 million US dollar. We can also start with 10 million US dollar but in that case, the groth rate is much smaller and significantly less dividend payout will be possible after year 3.

Best regards,
Markus

So because you only issue 50% of your shares to you shareholders, which would invest 15 million in total, that means if I invest 10'000$ I would receive 3333$ as dividend in your example right? will I be able / allowed to sell me shares?
newbie
Activity: 17
Merit: 0
Our chips work between 60°C and 80°C depending in their speed grade, the adjusted fan speed and their position on the board.

Markus
hero member
Activity: 854
Merit: 500
einc.io
What are the temperature of the chips of each cards?
newbie
Activity: 17
Merit: 0
Hi,

up to now Bitcoin has been a very volatile business, an effect we find in many extremely new business fields like the internet in 90s. But we expect the Bitcoin to consolidate and stablilize. We want to support the Bitcoin in this stabilization and want to be a long-term part of the ecosystem. We believe in Bitcoin and we hope others do as well.

The company will be a German one and its headquarters will be in Dresden. The datacenter where the miners are physically located will be in other countries, e.g. in Iceland as our generation 1 miners, due to much lower power costs.

Best regards,
Markus
hero member
Activity: 854
Merit: 500
einc.io
And where in Europe are you going to build your mine?
legendary
Activity: 1904
Merit: 1007
I find 3 years a very long period in the Bitcoin ecosystem.
newbie
Activity: 17
Merit: 0
Hello,

we expect less than $1 per GH/s retail costs and even less production costs based on our experience with our generation 1 miners.

The annual general meeting of all shareholders decides on the dividend payout based on the achieved profit, so we cannot preset the dividend payouts now. However, we would like to pay 50% of the profit (in our scenario these are about 10 million US dollar) as dividend to the shareholders after year 3. The rest we want to use for reinvestments. In the following years the profit and dividend payout will be even higher - assumed Bitcoin stays at least stable.
We know, we will require two years to build up the mine by reinvestment of the mined Bitcoins in order to achieve significant market share. That's the reason for no dividend payout in the first two years.

In order to build up the mine with a reasonable growth rate our calculations recommend a total of 15 million US dollar. We can also start with 10 million US dollar but in that case, the groth rate is much smaller and significantly less dividend payout will be possible after year 3.

Best regards,
Markus
member
Activity: 92
Merit: 10
I read in your pager that you are planning to pay no dividends till the end of the third year. I also so a lot of numbers about profits etc., but what your expected annual rate of return for an investor for the 3., 4. and the 5. year? How much do you payout to your investors after the third year?

For everbody who is interested -> https://coinbau.com/files/cb_investment_brochure_web_en.pdf
hero member
Activity: 924
Merit: 1000
Keen on seeing this come out.

What is the total investment you require to get this produced?

full member
Activity: 129
Merit: 100
Hello all,

There have been some questions concerning our technology and why we are able to realize a power efficiency others can't. It's time to answer them:

Our hashing core is a fully pipelined architecture whereas our key differentiator is a very area efficient custom pipeline implementation which we developed within the last 6 months based on a so called cell based design approach. We combined standard cells with custom optimized complex cells which save area and power.
Our high density core enabled us to realize much more hashing cores on the same silicon area, running at quite low frequencies. The low performance requirements of a single core allow us to apply an ultra-low supply voltage, which is much lower than the standard at this GLOBALFOUNDRIES 28nm process node. Therefore we optimized and re-characterized our in-house standard cell library for such low supply voltages. To ensure that this will work first-time right, we developed the complete concept of this ultra-low voltage design in close cooperation with the GLOBALFOUNDRIES design enablement team here in Dresden, which has detailed knowledge of the selected 28nm technology produced here in Dresden too.
Our unique design achieves in nominal mode 1.4 GH/mm² and 0.19 J/GH. However, similar to other hardware providers, we can change the frequency of the hashing cores in a high performance mode realizing 2.1 GH/mm² at an energy consumption of about 0.29 J/GH.

If there are any questions to be answered, do not hestitate to ask.

Best regards,
Markus


Thanks Markus,

You also give some specification about your 3rd gen 14nm ASIC in your investment document. 
Do you really think it is feasible to reach 0.08 J/GH with 14nm? How was this estimation done?
Pages:
Jump to: