Bitcoin and The Age of Bespoke Silicon; (Bitcoin mining reaches academia)

[perseus]

Hi all,

I wrote a short paper that talks about the many innovations of the Bitcoin mining community. It will be published in Oct. 2013 in an academic conference called CASES (http://esweek.acm.org/cases/).

My goal was to put the many amazing efforts on this forum in a positive light, and to package it so that the researcher community could see what amazing things were done here.

I believe that the Bitcoin ASIC efforts could be used as a model for new hardware startups in the future, since venture capitalists seem hesitant to fund new chip efforts these days. Many kudos to the many innovators and great thinkers on this forum---I have been super impressed and I hope that I have done you all justice.

I also talk about about some of the factors that will shape the upcoming 28-nm generation of Bitcoin ASICs, and what potential optimizations would be fruitful and what challenges (namely, Dark Silicon) will curtail performance.

Feedback and ideas appreciated!

Warm regards,

Michael Taylor (Perseus)
Associate Professor
Department of Computer Science and Engineering
University of California, San Diego
http://www.cse.ucsd.edu/~mbtaylor

-----------------

Bitcoin and the Age of Bespoke Silicon
CASES 2013
http://cseweb.ucsd.edu/~mbtaylor/papers/bitcoin_taylor_cases_2013.pdf

Recently, the Bitcoin cryptocurrency has been an international
sensation. This paper tells the story of Bitcoin hardware: how a group
of early-adopters self-organized and financed the creation of an
entire new industry, leading to the development of machines, including
ASICs, that had orders of magnitude better performance than what Dell,
Intel, NVidia, AMD or Xilinx could provide.

We examine this story for clues as to how we can foster greater
innovation in the semiconductor industry and enable this phenomenon to
occur more broadly for more application areas, spawning a new age of
hardware innovation tailored to emerging application domains---an Age
of Bespoke Silicon.

[1714864982]

1714864982

[1714864982]

1714864982

[digitalmagus]

A few names...

Bitfury - You should ask him for an interview, but be prepared to talk technical for a while. The guy deserves major credit. He worked as a one man team and taught himself how to design the most power efficient full custom 55nm ASIC in Bitcoin in a mere 1 year, with successful batch production on first run without MWP. Part of the feat was lots of optimizations related to SHA-256 core design. Meanwhile, everyone else with larger teams, some of them veterans (10-20 years of experience) in the industry, some of them large companies even with 28nm designs still can't really beat his power efficiency. The guy is simply some kind of genius.

Intron and c-scape also deserve some credit for the furiously fast PCB designing - some times without even a chip in hand or detailed chip specs to test the PCB with and later finding out it works on first try, although early versions of some of the PCBs weren't exactly perfect, but quite an accomplishment regardless.

Avalon Team  - After raising $300K from crowdfunding, they started several months behind Butterfly labs, and beat them to the market by many months with the first ever bitcoin ASIC at 110nm. They even had better SHA-256 optimizations than BFL. They were also touted for designing their device case in a single day, although not exactly pretty, it got the job done and product out the door fast. Prior to their ASIC design, they also did 2x FPGA designs, both of which were later fully open sourced. Oh and their ASIC PCB has also been open sourced. Yifu, the public face of avalon at one point had his face on T-shirts sold at bitcoin2013 with the word "Liberate" under it. I believe this was due to his decision to sell his ASICs (the first batch of which he could hold in a small case in his hand and supposedly accounted for some 25% of the entire bitcoin network's hashrate at the time), instead of self-mining them for personal profit he and his team decided to sell them, thus ensuring ASICs were widely distributed in the community and assuring the network security of bitcoin, which at the time was somewhat in doubt. Contrary to this, recent batches of his ASICs are rumoured to have been sold in very large  quantities to single individuals, further, and to be fair to their customers, as evidenced in the many threads, they are now having lots of problems delivering hundreds of thousands of ASIC chips they sold, to say nothing of very old Batch2 and 3 orders never being delivered upon.

friedcat of ASICminer - Whilst his  ASIC design was somewhat laughable with various instances of unedited copy/paste design details from available industry templates, he deserves major credit for business astuteness, having his company's share price valued in excess of 150 Million dollars at one point, far surpassing the value of any other company based on a single chip product.

ButterFly Labs - Deserves the what-not-to-do award... at least 10 times over, for about 20++ different reasons - including my personal favorite, putting a heatsink on top (instead of under) a QFN48 packaged ASIC chip. I'll let their customers elaborate ;-)

The following companies, all with their first products with ETAs for Q4 2013... Cointerra, KNCminer, HashFast and FastHash ...are still to be judged

And the Professionalism, Customer Service and Ethics award goes to  ... NO COMPANY (so far)! ...although buzzdave of 100TH has the best individual customer service and ethics I ran into.

[agath]

...although buzzdave of 100TH has the best individual customer service and ethics I ran into.

I can fully confirm this statement!

[k9quaint]

Can I pre-order a copy of the paper? 

[rethaw]

Thanks, I poked around your personal site.

Have you written on bitcoin mining and dark silicon?

EDIT: Read more.

[oaxaca]

My goal was to put the many amazing efforts on this forum in a positive light, and to package it so that the researcher community could see what amazing things were done here.

Yes, you have put BFL and Avalon (actually you should revise your paper.  Avalon is the name of their product, not the name of the company) in a positive light.  You have obviously missed the tremendous amount of anger towards these 2 actors here in this forum.  BFL is actually in the dictionary under the entry "scam".  While BitSyncom (who you call Avalon) will be next.  Simply look at the most recent threads on the same page where your thread currently resides and you will see a dozen regarding these 2 companies.

The favorable light you cast these 2 in might serve your purpose in the paper, but will only result in more anger here among the faithful.  Is there a presenter at the conference that specializes in Business Psychology or Criminal Law?  That would be a fantastic counterpoint.

Both of these companies are sitting on mountains of customer cash while NOT delivering the products you are glorifying. 

[oaxaca]

Avalon Team ... they are now having lots of problems delivering hundreds of thousands of ASIC chips they sold, to say nothing of very old Batch2 and 3 orders never being delivered upon.

Problems?  That might be the understatement of the year.

[oaxaca]

This just keeps getting worse.  It's the glorifying of outfits like BFL that cause situations like this:

Hi,

Few days ago I ordered 2 Monarch cards from BFL.
When I ordered they say delivery is estimated in October/November.
Now I see the first batch is sold out (http://www.butterflylabs.com/monarch/)
But the delivery dates they publish for the sold out dates are... "November/December"
Looks like they got another month delay just few days after they start taking orders...
If they finish the "tape out" few few days ago and at that time they estimate 60-90 days what can change now to get an estimation of 90-120 days.
With 60-90 days was a risky bet to buy... with 120 is a suicide.
I know may I have been an idiot buying from them.

Regards

This poor noob just gave them $4,680 and will never see it again.  I feel terrible for him.  I'm not saying he read your paper, but he might have seen somebody else's work where they just wanted to put things in a positive light.  Maybe you should interview him for your paper.

[perseus]

Thanks, Digitalmagus -- some great points in here!

A few names...

Bitfury - You should ask him for an interview, but be prepared to talk technical for a while. The guy deserves major credit. He worked as a one man team and taught himself how to design the most power efficient full custom 55nm ASIC in Bitcoin in a mere 1 year, with successful batch production on first run without MWP. Part of the feat was lots of optimizations related to SHA-256 core design. Meanwhile, everyone else with larger teams, some of them veterans (10-20 years of experience) in the industry, some of them large companies even with 28nm designs still can't really beat his power efficiency. The guy is simply some kind of genius.

Intron and c-scape also deserve some credit for the furiously fast PCB designing - some times without even a chip in hand or detailed chip specs to test the PCB with and later finding out it works on first try, although early versions of some of the PCBs weren't exactly perfect, but quite an accomplishment regardless.

Avalon Team  - After raising $300K from crowdfunding, they started several months behind Butterfly labs, and beat them to the market by many months with the first ever bitcoin ASIC at 110nm. They even had better SHA-256 optimizations than BFL. They were also touted for designing their device case in a single day, although not exactly pretty, it got the job done and product out the door fast. Prior to their ASIC design, they also did 2x FPGA designs, both of which were later fully open sourced. Oh and their ASIC PCB has also been open sourced. Yifu, the public face of avalon at one point had his face on T-shirts sold at bitcoin2013 with the word "Liberate" under it. I believe this was due to his decision to sell his ASICs (the first batch of which he could hold in a small case in his hand and supposedly accounted for some 25% of the entire bitcoin network's hashrate at the time), instead of self-mining them for personal profit he and his team decided to sell them, thus ensuring ASICs were widely distributed in the community and assuring the network security of bitcoin, which at the time was somewhat in doubt. Contrary to this, recent batches of his ASICs are rumoured to have been sold in very large  quantities to single individuals, further, and to be fair to their customers, as evidenced in the many threads, they are now having lots of problems delivering hundreds of thousands of ASIC chips they sold, to say nothing of very old Batch2 and 3 orders never being delivered upon.

friedcat of ASICminer - Whilst his  ASIC design was somewhat laughable with various instances of unedited copy/paste design details from available industry templates, he deserves major credit for business astuteness, having his company's share price valued in excess of 150 Million dollars at one point, far surpassing the value of any other company based on a single chip product.

ButterFly Labs - Deserves the what-not-to-do award... at least 10 times over, for about 20++ different reasons - including my personal favorite, putting a heatsink on top (instead of under) a QFN48 packaged ASIC chip. I'll let their customers elaborate ;-)

The following companies, all with their first products with ETAs for Q4 2013... Cointerra, KNCminer, HashFast and FastHash ...are still to be judged

And the Professionalism, Customer Service and Ethics award goes to  ... NO COMPANY (so far)! ...although buzzdave of 100TH has the best individual customer service and ethics I ran into.

[perseus]

Thanks, Oaxaca. I appreciate your feedback; you have great points, especially about the issue of customer frustration. In the article, I intentionally tried not to focus on the gold-rush aspects of bitcoin mining, or to advertise large fortunes earned by purchasers of BFL or Avalon machines, since that is not the common case. Instead, I wanted to focus on the hardware innovations of the respective efforts --- small groups creating a new class of chip is almost unheard of these days.

Regarding Bitsyncom; interesting. I noted the use of that name earlier on, but I used Avalon because that is how they currently refer to themselves (as in their blog postings on http://www.avalon-asics.com/).

In the article, I do emphasize several times that extreme delay of the BFL units and also the anxiety that customers (including myself) experienced: "frustration and animosity from many enthusiastic customers anxiously looking at the rising BTC difficulty curve and wondering if they had bet on the slow horse."  I also emphasize that the most important factors in mining profitability is your chosen vendor's ability to deliver a best-of-class product before the others, and the position you are on their wait list. Unfortunately, given that those are the primary two factors that determine your payoff, then bitcoin mining is inherently fraught with exactly the kinds of risk and frustration that folks have encountered and come from a requirement to "play the game" well.  

Thinking about what you are saying, I agree there are good and meaty questions regarding the issue of pre-order financing and whether stricter guarantees should be made about the number of units produced (at least for Bitcoin mining; may not apply for other kinds of Bespoke ASICs) and to what degree a manufacturer may legitimately represent that they will not refund moneys even in cases of extreme delays. I imagine this could also easily happen on Kickstarter as well -- for instance, an Android gaming console that ends up being obsoleted by an Nvidia product that is released before all of the consoles have shipped. From this perspective, the ASICMINER IPO model may be the best one, since the possibility of loss of funds is implicit and well understood in stock purchasing.

This just keeps getting worse.  It's the glorifying of outfits like BFL that cause situations like this:

Hi,

Few days ago I ordered 2 Monarch cards from BFL.
When I ordered they say delivery is estimated in October/November.
Now I see the first batch is sold out (http://www.butterflylabs.com/monarch/)
But the delivery dates they publish for the sold out dates are... "November/December"
Looks like they got another month delay just few days after they start taking orders...
If they finish the "tape out" few few days ago and at that time they estimate 60-90 days what can change now to get an estimation of 90-120 days.
With 60-90 days was a risky bet to buy... with 120 is a suicide.
I know may I have been an idiot buying from them.

Regards

This poor noob just gave them $4,680 and will never see it again.  I feel terrible for him.  I'm not saying he read your paper, but he might have seen somebody else's work where they just wanted to put things in a positive light.  Maybe you should interview him for your paper.

[perseus]

Hi Rethaw,

In the paper, I do have a section that talks about the impact of dark silicon on future generations, including 28 nm.

Thanks, I poked around your personal site.

Have you written on bitcoin mining and dark silicon?

EDIT: Read more.

[perseus]

Just to follow up on this, digitalmagus, thanks for the shoutouts on Bitfury.

His design is particularly fascinating because, unlike the others, he opted to go full-custom, and as a result was able to skip out on the expensive standard-cell CAD tool flow by leveraging his unique "sea of hashes" approach rather than the traditional pipelining unrolling approach. This is a neat solution to the problem of CAD tool financing! Doing all of this and hitting an advanced process node like 55 nm for his first chip is simply amazing. I am planning on incorporating his work in a future version of the paper, targeted at a wider audience.

A few names...

Bitfury - You should ask him for an interview, but be prepared to talk technical for a while. The guy deserves major credit. He worked as a one man team and taught himself how to design the most power efficient full custom 55nm ASIC in Bitcoin in a mere 1 year, with successful batch production on first run without MWP. Part of the feat was lots of optimizations related to SHA-256 core design. Meanwhile, everyone else with larger teams, some of them veterans (10-20 years of experience) in the industry, some of them large companies even with 28nm designs still can't really beat his power efficiency. The guy is simply some kind of genius.

Intron and c-scape also deserve some credit for the furiously fast PCB designing - some times without even a chip in hand or detailed chip specs to test the PCB with and later finding out it works on first try, although early versions of some of the PCBs weren't exactly perfect, but quite an accomplishment regardless.

[fpgaminer]

A great read, perseus!  Thank you for sharing.

[perseus]

Thanks, fpgaminer!

A great read, perseus!  Thank you for sharing.

[Phinnaeus Gage]

Just read it. Well-written. Thanks for your contribution, perseus.

~TMIBTCITW