What Bitfury aren't telling you

[brontosaurus]

Firstly, congratulations to Bitfury on getting their chip working. Considering the number of asic mining chips that have failed to deliver in one way or another, it's always good to see a new one coming along.

However, I think they are being somewhat 'economical with the truth' in terms of the actual performance of the device, and that makes me a little suspicious of what they say. I'm especially wary of their power consumption figures, or should I say lack of said. Yes, they are given a headline figure of 0.06 J/GH, which is a meaningless metric as it should be J/(GH/sec) for any meaningful comparison, but then they claim 40Gh/sec at the same figure. Still no mention of clock speed, but we'll leave that just now.

Many month ago Guy Corem of Spondoolies posted that sub 0.1 J/(GH/sec) was possible in 28nm with 'extreme' design  and he is absolutely right, in fact Bitfury get their low power consumption with a very low voltage and presumably a very low clock speed.

This is hardly breaking edge design, especially in 16nm.

Bitfury go on to say the device can operate up to 192GH/sec but fail to mention the power consumption (or clock) at this speed. Why? If this device is as energy efficient as they say them surely it should be at least sub 0.1 J/(Gh/sec) to compare with what Bitmain can achieve in 28nm? Stepping own two process nodes (28 to 20 to 14/16) should result in energy saving of at least 50% so why hide it unless the actual figure is higher than expected? They also convenient ignore to specify the J/(GH/sec) at 55 Gh/sec, mentioning only 'convergence to a plateau' (from 40 Gh/sec) whatever the hell that means.

Then there is the matter of the die size. Again no mention of this in Bitfury's press release, but one of Guy's extreme design techniques involves using a lot of silicon area, so I'm hazarding a guess that this little beast is probably a good deal larger than one might expect. I'm taking a guess that the chip is running at around 1Ghz when producing 192 GH/sec in which case a cell based design would come out at around 25 square millimetres, 5mm on a side. I'd love to know how this one compares, naturally a bigger clock speed means a smaller die.

I don't have any Bitfury products or interest in their company save what they choose to print, so I have no axe of any kind to grind with them. They can write what they like, but please don't try to mislead or misinform people, I know they are not in the consumer market any more but take a leaf out of Spondoolies book and give people comprehensive REAL figures of what your devices can achieve in practice rather than carefully chosen 'sweet spots', especially if you are planning to ask them to part with money. That's what the real engineers do.

[1715383513]

1715383513

[1715383513]

1715383513

[1715383513]

1715383513

[1715383513]

1715383513

[1715383513]

1715383513

[1715383513]

1715383513

[philipma1957]

So?  Does it matter to anyone?

My estimates of the chip are  as follows.

They opened a 40mega watt plant

Based on the hash rates spikes last diff and this diff  the plant does   175 to 225ph.


This should be  35megawatt used since overheating they do not go full force.

This is .2 watts at worst and .155 watts at best .

Not close to .06  but pretty good compared to anyone else.

[mwizard]

However, I think they are being somewhat 'economical with the truth' in terms of the actual performance of the device, and that makes me a little suspicious of what they say. I'm especially wary of their power consumption figures, or should I say lack of said. Yes, they are given a headline figure of 0.06 J/GH, which is a meaningless metric as it should be J/(GH/sec) for any meaningful comparison, but then they claim 40Gh/sec at the same figure. Still no mention of clock speed, but we'll leave that just now.

I lost a lot of confidence in Bitfury when they announced their Bitcoin mining light bulbs earlier this year.  It seemed such a silly concept for a serious company.

Still they seem to have 16nm chips and are claiming good efficiency.  Bitfury claim efficency "ranges from 0.055 joules per gigahash to 0.07 joules per gigahash".  It is unlikely they would lie on what is basically an engineering issue easily checked by an investor or anyone involved with their hardware.

Bitfury seem to have used the correct units to measure efficiency.  Efficiency of miners is traditionally measured as the ration between energy (Joules) and Gigahash.  Either J/Gh or Gh/J depending on local convention.  

KNC seem to have had 16nm chips working for some time. KNC taped out in February and claimed working chips in June.   A couple of months ago they added about 15-20 Petahash to their pool.  Presumably their new data centre.  KNC claim 0.07 J/Gh.  

http://www.coindesk.com/kncminer-deploys-next-generation-16nm-bitcoin-asic/
http://www.cio.com.au/article/590688/bitcoin-miner-knc-planning-another-four-week-datacenter-build-out/

http://www.businesswire.com/news/home/20151216005453/en/BitFury-Announces-Mass-Production-Fastest-Effective-16nm

[brontosaurus]

So?  Does it matter to anyone?

My estimates of the chip are  as follows.

They opened a 40mega watt plant

Based on the hash rates spikes last diff and this diff  the plant does   175 to 225ph.


This should be  35megawatt used since overheating they do not go full force.

This is .2 watts at worst and .155 watts at best .

Not close to .06  but pretty good compared to anyone else.

It matters to people who are thinking of buying new kit, and from comments I've seen there are quite a few folks who have decided to wait for Bitfury's 'better' offerings rather than buying Bitmains product on the grounds the Bitfury product will be more efficient.

By the looks of things it won't and their released information is misleading. I though that one of the purposes of this forum is to share information and flag up data that simply isn't true?

They do not  have a cost effective solution that runs at 0.06 J/(GH/sec), their chip is designed to run at 192 GH/sec, not the 40GH/sec quoted at the 0.06 figure. Sure, you could use 5 x 40GH/sec chips running at 0.06 to get to 192, but then you use nearly 5 times the amount of energy! (and 5 times the chip, board and component cost).

In short, you can buy Bitmains product today and get just as cost efficient a product as Bitfury's 16nm offering.

All that NRE and development money for nothing.

[brontosaurus]

However, I think they are being somewhat 'economical with the truth' in terms of the actual performance of the device, and that makes me a little suspicious of what they say. I'm especially wary of their power consumption figures, or should I say lack of said. Yes, they are given a headline figure of 0.06 J/GH, which is a meaningless metric as it should be J/(GH/sec) for any meaningful comparison, but then they claim 40Gh/sec at the same figure. Still no mention of clock speed, but we'll leave that just now.

Bitfury seem to have used the correct units to measure efficiency.  Efficiency of miners is traditionally measured as the ration between energy (Joules) and Gigahash.  Either J/Gh or Gh/J depending on local convention.  

The only meaningful metric is J/(GH/sec), if you don't have a time factor then the figure may be correct but it may take the age of the universe to make a single calculation. Running a chip in it's 'sweet spot' is totally misleading, the sweet spot may be at 0.33V, 50 Mhz for a chip designed to run at 0.66V, 1.2Ghz where it's energy efficiency is much lower.

[Biodom]

Well, all these discussions are academic. It is obvious that they are deploying in bulk-just look at that difficulty rise.
It is a "killa" for everyone else as price is falling behind the diff increase.
At current speed of diff rise, S5 will stop producing any income in March (at $0.1/kwh), S7 in June.
A major bummer.

[GSpgh]

The only meaningful metric is J/(GH/sec), if you don't have a time factor then the figure may be correct but it may take the age of the universe to make a single calculation. Running a chip in it's 'sweet spot' is totally misleading, the sweet spot may be at 0.33V, 50 Mhz for a chip designed to run at 0.66V, 1.2Ghz where it's energy efficiency is much lower.

I see where you're coming from but I don't think that's a big deal

J/GH is what everyone else uses and it's fine for comparison

if they produce something woefully inefficient at high speeds like you're saying then it means they lied but they can also do what bitmain did - more chips at lower speed to meet their promise (batch 1 S7 at 600 MHz vs batch 8 at 700 MHz)

but since all we have so far is their PR then it's a schrodinger's miner anyway...

[HyperMega]

The only meaningful metric is J/(GH/sec), if you don't have a time factor then the figure may be correct but it may take the age of the universe to make a single calculation. Running a chip in it's 'sweet spot' is totally misleading, the sweet spot may be at 0.33V, 50 Mhz for a chip designed to run at 0.66V, 1.2Ghz where it's energy efficiency is much lower.

Sorry, meaningful would be either

W/(GH/s) (W for Watt, which is the unit for power)

or

J/GH (J for Joule, which is the unit for energy)

because 1 J = 1 Ws 

But in general I agree with you.
Without the (GH/s)/mm² @ 0.06 J/GH number and how much of their engineering samples really work at these ultra low voltages, this press release is misleading.
I would speculate too that at this operation point the chip would not be compatitive in terms of $/(GH/s).

And sidehack mentioned that the efficiency for different GH/s configurations could be calculated based on the information provided in the press release. He stated that it should end up somewhere around 0.12 J/GH, if the chip runs at 100 GH/s. But this still doesn't tell us anything about the die size. 

[QuintLeo]

$/GH is also important, not just J/GH

 If you have 10% better J/GH but your $/GH is ten times higher, you have a product that won't do well in the market and will probably never reach positive RoI.

 Reliability is also important, and in the 14/16nm generation will become a LOT more important as a long-term factor.

[Mitak]

We'll have to wait and see when the chip specs are published.
For sure my favourites in terms of IC design are Spondo and Bitfury.
The time will tell

[r00tdude]

None of this really matters though. To date, only Bitmain and Avalon have produced the newest devices and sold them to consumers - KnC has held back anything new they created to mine themselves and it seems like BitFury is doing the same at least for now... so, all that really matters here to us is that centralization of mining resources will continue to rise remarkably high for the near future.

What we really should be concerned about is what this centralization of resources will do to the bitcoin community/economy at large. Will people like us lose confidence in Bitcoin once the majority of the mining and pool resources are in the hands of companies like these? What will that do to adoption and, ultimately, price? One of the earliest drivers of value is the 'network effect', which could tamp down markedly if so much of the PH is in the hands of corporations.

Surprisingly, or maybe not-so suprisingly, ASIC 'hardened' scrypt currencies have seen little to no ASIC development and delivery since the onrush of the second gen rigs hit the market. While there are some chips on the horizon (Innosilicon A4's and SFARDS) they still aren't nearly as efficient, and are available in much smaller quantities... and those coin networks suffer from far less centralization or corporate influence.

Frankly, I see a come-uppance that Bitcoin will have to suffer at some point in late 2016 after the halving of the block rewards where the questions of corporate mining will have to be addressed and/or dealt with. I'm pretty sure during the run-up to the halving will be exciting to experience, but the time after is going to be interesting as we deal with the afterglow and what the landscape looks like with decreasing mining rewards sitting in everyone's face and increasing control over the processing of transactions by companies that refuse to sell their efficient tech to promote decentralization.

In that vacuum, I think it's likely that scrypt coinage will gain quite a bit against bitcoin. I'm betting that way, anyway. YMMV.

[hazenyc]

Joules = Watts/Seconds so 1 Watt per GH/s = 1 Joule per GH.

daily cost is what matters and = [price per kWh x J/GH x 24 hours per day] x [hashrate in GH/1,000]